#cs183

Sonia Arrison: I spend most of my time looking at biotech, so I’ll talk about the biotech slice first. It is clear that biology is quickly becoming an engineering problem. I got interested in biotech several years ago when my CS friends started picking up biology books. They thought, probably accurately, that the next big thing in coding would be bio, not computers. This is now a mainstream view. Bill Gates has said something like this, along with several others. Great hackers go into biotech. In 30 or 40 years, the bio-as-engineering paradigm could make the world look radically different. There is a sense in which genomics is moving faster than Moore’s law. Prices are falling; the first human genome sequencing was around $3 billion. Now it can be done for around $1,000. There is work being done on a genomic compiler, which would make it easier to hack all sorts of organisms’ genomes, which would in turn open up all kinds of possibilities.

The big complaint right now is that, despite the fact that the first draft of the human genome was sequenced in 2000, twelve years later not that much has actually happened in terms of new treatments or cures based on the technology.  . This criticism is weak because it misses an important point: for most of those 12 years genomic sequencing was so expensive that very few scientists could do the work they wanted to do using genomes. Of course, now that prices have fallen substantially, barriers are falling in a serious way. Things will happen—people are working on radical new things. Gene therapy promises to cure diseases. It’s possible that we can develop new kinds of fuels. There is a Kickstarter project that involves taking an oak tree and splicing firefly genes into it. The end result would be trees that glow. More than just cool in it’s own right, maybe you could use those firefly trees to illuminate roads instead of streetlamps. That’s awesome. And there is so much more that we can’t even fathom right now. A lot can, and will happen at the nexus of bio and engineering.

In the short run and outside of biotech, the shift to online education seems like it will dramatically change how people learn. Things like the Stanford AI class, Udacity, the Kahn Academy—we don’t know exactly how it will all play out, but it’s safe to say that there are a lot of things to look forward to on this front.

Peter Thiel: Let’s engage on the culture question: why do most people think you’re crazy?  

Michael Vassar: For whatever reason, having opinions about the future is seen as strange. Only a small minority of people forms opinions about the future—even the near future. Perhaps this is because thinking about the future is uncomfortable and kind of difficult. People prefer to work with models that involve one variable changes in linear trajectories, while everything else stays the same. We know that that’s nonsense, of course; the world doesn’t work like that. But it makes for easy conversation. Keeping the discourse at that simplistic level allows us to focus on one thing and work together today. Factoring in 100 variables would in some sense break that dynamic. But thinking about the future is very important, and right now that can be isolating. Diverging from people means that there are fewer people you can talk to. There are fewer connotations; people tend not to understand where you’re coming from.

But this is not to say that people just have different beliefs than we do. Usually, they don’t. You don’t usually encounter anti-singularity views. Maybe some global warming people or apocalypse people are affirmatively anti-singularity. But most people aren’t substantively engaging. What is perceived as crazy isn’t the substance of the belief itself, but rather having the belief in the first place. 

Aubrey de Grey: I disagree a bit. People do tend have some view of the future. They usually project relative stagnation. People tend to believe that, not only will most things not change, but what will change won’t change very quickly. People who criticize my views on biotech and aging, for instance, do not identify bad logical steps or seize on anything substantive. Rather, they choose not to believe what I’m saying because it conflicts with their bias toward stagnation. They walk away quite sure that the rate of progress in anti-aging and longevity technology will never accelerate. That is pretty striking. 

I try and dispose of this by pointing out that if you were to ask someone in 1900 how long it would to cross the Atlantic in 1950, they would make a prediction drawing from ocean liner speed trajectories up to that point. They wouldn’t be able to foresee the airplane. And so their calculation would be off by orders of magnitude.

Of course, everyone knows how much technological change has happened in the past few centuries and decades. Everyone knows what the Internet did in recent years. But there is a huge reluctance to apply any of this as precedent for what might or is likely to happen in the future.

There’s also a desirability aspect to it. Fear of the unknown is such a deep-seated emotion. When people encounter a radical new proposition, they are biased to think that things will go way wrong. It is very hard for people to consider the reasonable likelihood of those scenarios unfolding, so they exaggerate risks. More rational aspects to the conversation go out the window.

Sonia Arrison: For the record, no one thinks that I’m crazy. 

Peter Thiel: You’re the best disguised…

Sonia Arrison: Well, “crazy” is a hard claim to make since I focus on actual technology that is grounded in reality. I write about tissue engineering, regenerative medicine, and biohacking, for instance. That exists now. And it’s going to continue to develop and, I think, really change the world. There are three reasons that people sometimes have a problem with this stuff. First, they don’t understand it. Second, they don’t believe it. Third, they fear it.

Think about the firefly/oak tree street lamps for a second. Just the idea of that terrifies some people. It’s completely different from how things are now. Some people respond with knee-jerk reactions: “Don’t mess with nature!” “Don’t play God!” This reaction is understandable, but it stands in the way of progress. It’s not the best reaction. In a lot of ways it doesn’t really make sense.

Peter Thiel: Is the best approach to ignore those people, then?

Sonia Arrison: Better than ignoring them is trying to educate them. It is important to explain things clearly. Technology that people do not understand looks a lot like magic sometimes. And Magic is scary. But if you distill and explain—“this is x, this is what it does”—you can sell them on it. It’s just a matter of clearly communicating the benefits vs. the costs. “This will drive out dirty fossil fuels,” for example, might be one persuasive line of argument in favor of the firefly/tree hybrid street lamps.

Peter Thiel: There’s a compelling case that we’ll very likely see extraordinary or accelerated progress in the decades ahead. So why not just sit back, grab some popcorn, and enjoy the show?

Another cut at the question is this: In Kurzweil’s The Singularity is Near, progress follows an exponential growth curve. It’s a law of nature. In a sense, the singularity is happening regardless of what individual people actually do to make it happen. The assumption was that there will always be enough people who try things, so you, as an individual, don’t actually have to do anything and you can just wait for things to happen. Is there anything wrong with that argument?

Aubrey de Grey: Yes, there is. It doesn’t only matter that these technologies are developed. When they are developed is hugely important as well. Take anti-aging science, for instance. Very close to 150,000 people die everyday. About 100,000 of these daily deaths are aging-related. (Probably about 90% of deaths in Western countries are aging-related). So each day that you don’t delay saves 100,000 lives. From this perspective, it doesn’t matter how inevitable the singularity is. Inevitable is cold comfort to the people losing their lives or loved ones now. We want the defeat of aging by medicine as soon as possible, for the simple reason that more suffering is alleviated the sooner we achieve it.

Michael Vassar: I strongly agree. It is important to work toward making good effects happen, and avoiding bad things. Inevitability can cut both ways; sometimes you want it to happen, if the effects are good, but sometimes you don’t want certain things to happen. Focusing just on inevitability misses other important pieces. If death is or seems inevitable and we are basically dead in the long run, there is still some chance at survival, and we should give it a damn good fight. 

Besides, popcorn is bad for you. Though I guess Aubrey might figure out a way to make it not so bad for you… 

Sonia Arrison: Focusing on inevitability alone is dangerous because it allows people to get complacent about bad systems in place. People might ignore the many perverse incentives that often thwart or frustrate the many scientists working on radical technologies. Too few people are thinking about how the FDA might be blocking very important developments. If it’s all going to happen anyway, there’s less of a sense that it is important to reform what we have now so we can better realize our goals. But of course that kind of reform is terribly important, and it won’t happen if we don’t work towards it.

Peter Thiel: So who do you think is going to do this? Who is going to forge the technological future?

Michael Vassar: You.  [laughter…] 

Peter Thiel: [pause] Michael... you’re supposed to be motivating the people in this class…

Michael Vassar: But I’m serious. It’s a short list of people. You, Elon, Sean…

Sonia Arrison: My take is that innovation comes from two places: top-down and bottom-up. There’s a huge DIY community in biology. These hobbyists are working in labs they set up in their kitchens and basements. On the other end of the spectrum you have DARPA spending tons of money trying to engineer new organisms. Scientists are talking to each other from different countries, collaborating on synthetic bio projects. All this interconnectedness matters. All these interactions in the aggregate will bring the change.

Aubrey de Grey: I disagree. My answer is Oprah Winfrey.

Yes, there are a few people like Peter. There are a very few visionary people who can make a real difference at the formative early stage. But there are also many people with Peter’s net worth who aren’t doing this. It’s not that these people don’t understand the issues or the value of technology. They understand these things very well. But they are held back by social opinion. They probably can’t articulate this well to themselves, let alone to others. But they face viscerally emotional blockades that the people around them erect. Just because you’re rich doesn’t mean you don’t fear people laughing at you. Many potential visionaries are held back by little more than social pressure to conform.

This is why mainstream opinion formers are absolutely pivotal. Perhaps no other subset of people could do more to further radical technology. By overpowering public reluctance and influencing the discourse, these people can enable everyone else to build the technology. If we change public thinking, the big benefactors can drive the gears.

Michael Vassar: I do not think that progress will come from the top-down or from the bottom-up, really. Individual benefactors who focus on one thing, like Paul Allen, are certainly doing good. But they’re not really pushing on future; they’re more pushing on individual thread in homes that it will make the future come faster. The sense is that these people are not really coordinating with each other. Historically, the big top-down approaches haven’t worked. And the bottom-up approach doesn’t usually work either. It’s the middle that makes change—tribes like the Quakers, the Founding Fathers, or the Royal Society. These effective groups were dozens or small hundreds in size. It’s almost never lone geniuses working solo. And it’s almost never defense departments or big institutions. You need dependency and trust. Those traits cannot exist in one person or amongst thousands.

Peter Thiel: That’s three different opinions on who makes the future: a top-down bottom-up combo, social opinion molders, and tribes. Let’s run with some version of Michael’s tribe theory. Suppose it’s just a small cabal of tech people that drives it.

Aubrey de Grey: I think the tribe argument is right. Michael is right that single people don’t make the difference. There is too much infrastructure. Working in biology costs a fair bit of money. Developing algorithms can be quite costly too. Individuals have to fit themselves into the network of money flow, whether that network is entrepreneurial, philanthropic, or public funding. But the truly radical technology discussed in this class is so early that philanthropic support will probably play the largest role for awhile longer. That can change fast as these technologies advance and more people start to see the commercial viability. When public opinion changes, the people who want to get elected will fund the things that people want, and we’ll start to see more funding for these things.

Sonia Arrison: In some sense asking for a single source of progress  is the wrong question. It can come, and almost always does come, from lots of places. Things are interconnected. Ideas build on top of each other, and often ideas that once seemed unrelated can come together later on.

Question from the audience: We know that progress has happened in the past. But fairly rarely did that progress look like what people were expecting beforehand. So how do you know that your claims as to how progress is going to happen in the future are right? What do you make of the line that “most discussion about the future is either fantasy or bullshit”?

Michael Vassar: People used to predict the future in a pretty determinate way. Suppose you’re looking for oil. That involves making fairly concrete predictions: there is x amount of oil at y place, and it will last z number of years.

People have largely stopped doing that. Recent science fiction is a bit more on point than the science fiction of old. It used to be hard to predict the distant future. It may be that it’s actually quite easy to predict what the late 2020s look like, relative to what it used to be. But it is unusually hard to make any statement about 2040. 

People were much better at predicting the future before movies and mass media. The tools were logic and trend analysis, not what looked cool on the big screen. Modern forecasts of the future are often more about looking credible than about making reasonably accurate predictions. 

Consider things like Neal Stephenson’s Snow Crash—some very good abstraction there, somewhat satirical. There are lots of details that probably aren’t going to play out like that in the actual 2020’s. But we can think of them as being about as reasonable as Kurzweil’s descriptions of possible future technology. 

Sonia Arrison: The question basically says, “Well, a lot of people were wrong about the future in the past, so we shouldn’t talk about it now.” That’s nonsense. Yes, people will be wrong. But we’re not talking pie-in-the-sky guesses about the future. We’re talking about what is here now, and reasonably extrapolating from that. This isn’t science fiction. Gene splicing and gene therapy exist. We can create living code, as Craig Venter demonstrated. The questions are how long will this take and how fast can we go. These are difficult questions to answer. But that doesn’t mean we can’t think about them. We should think about them. That people have various perspectives doesn’t invalidate the project.

  Question from the audience: Will the future be a science problem or engineering problem? 

Aubrey de Grey: We are right in the middle at this point. In medicine and computation, for instance, we are seeing a shift from inherently exploration-based, science-based perspectives to engineering perspectives. 

Michael Vassar: Science matters much more than engineering does. But it’s easier to talk about engineering. So one should use engineering to discard the 99.9% of people who have no clue what’s going on. But then one should get into the science with the remnant. That is where the upside will come from.

Sonia Arrison: There is also a knowledge aggregation problem. It is hard or impossible for one human brain to know everything. So people don’t know what other people are doing, and they sometimes work on overlapping or redundant things. To the extent computers can better organize knowledge, people’s efforts will be further streamlined, whether they are scientific or engineering-focused.

Question from the audience: On the hardware side, Moore’s Law seems like it’s going to continue to hold. But on the software side, the process of software engineering and collaboration seems to be improving only linearly. Is there a leveragability problem or some hidden limit there?

Michael Vassar: Linear growth in capabilities can get you over key hurdles. There is a feedback loop. Linear growth can be enough for you you to nail down a process, leverage it, and get positive feedback to face transitions that then have the exponential growth arcs. And then you’re back to growing linearly.

This is true for probably all of psychology and for AI (which is essentially psychology-as-engineering).

Peter Thiel: We know that, in practice, timing is very important. So while we don’t know exactly when radical technology of the future will come to be, the timing does make a great deal of difference. If it’s all crazy science fiction that’s barely plausible, it might not make sense to work on it now. That would be like the Chinese man who tried to launch a rocket into space in the 11th century. No one was or should have been working on supersonic flight in the Middle Ages. That would be paddling way to far in advance of the wave.

Aubrey de Grey: I’m not sure the timing question is so critical. There must always be stepping-stones to an eventual goal. In the 11th century, the goal may have been to travel to the moon. But the technology then only permitted, say, a prospective space traveler to get one foot off the ground. So at that time, you’d get the equivalent of your PhD if you could make a system that got you 10 feet off the ground.

The question is thus which trajectories will lead toward the ultimate goal and which ones will fail. We must identify the good trajectories and prioritize them. But without the long-term goal, you can’t organize competing trajectories, and you’ll never get there.

Peter Thiel: So perhaps a 20-year goal with lot of milestones along the way would be a good approach. The problem there is that too many milestones make the achievability of the end goal rather speculative.

Aubrey de Grey: You have to see that coming, and avoid the wrong turns.

There are also humanitarian reasons to set the sights large. We must remember that 100,000 lives are saved each day that the solution to aging comes quicker. In that light, 20 years is dramatically better than 21.

Sonia Arrison: People usually become deterred if a goal seems too hard or impossible. We can’t expect everyone to be a tireless visionary. So showing traction is key. We can grow blood vessels and tracheas and bladders in the lab. So maybe we can get to hearts. The stepping-stones are key, since without them, fewer people will be as excited about the prospects of engineering new hearts.

Michael Vassar: The Apollo project was a tremendous 10-year project with lots of technological convergence. That was more than 40 years ago. At this point we probably can’t even go to the moon anymore.

Framing the U.S. Constitution was an incredible accomplishment. The Founders had the knowledge to do that. They wrote for a particular socioeconomic and technological context. They didn’t intend to write the end-all governing document for the entire world for all eternity. And yet, when we take over a Middle Eastern country today, we basically copy our Constitution. We have no idea how to do what our Framers did some 200 years ago. We’ve lost the ability to make such a culturally nuanced system. Applied history is underrated.

  Question from the audience: No trend can run without running into limits.  Where is the future asymptotic? When do we reach the limits of physical world? How long does the exponential part go, and when does it stop?

Michael Vassar: It’s hard to say where it stops. Probably not for a good while; there’s much more to be done. If something happens x times in a row, and no other variable is at play, one way to think about the chance of it happening again is to estimate it at (x+1)/(x+2). It’s a really crude technique, but can be quite useful too.

Aubrey de Grey: Kurzweil acknowledges that you get S-curves. But those curves tend to be replaced by new S-curves with each paradigm shift. Merge all those curves into one and you get a mega S-curve. Obviously there’s only so far you can go within physical laws. But we’re not hitting those problems yet.

Sonia Arrison: At some point, things decelerate. But that’s okay. Necessity is mother of all invention. There will be other things to tackle. There will always be a new exponential curve.

  Question from the audience: We at the Stanford Transhumanist Association are interested in open dialogue about the consequences of technological change, so we do a lot of research on how core emotions like fear or empathy come into play when people evaluate technology.

What do you think are the most effective ways to get people interested in and comfortable with transhumanist ideas?

Sonia Arrison: Sometimes it’s possible to just appeal to the humanistic side. Certain aspects of transhumanism would, fully realized, alleviate lots of suffering. Some issues fit in that category pretty well. So if you frame it right, the conclusion becomes a no-brainer. No one wants net suffering.

Other things don’t fit in that category as well. These are the things that just look radically different from the status quo—we might think they’re cool, but that’s not others’ default. The emotional argument on these things is that people should be free to be individuals. But there can be a serious fear factor on freedom. Some people are generally scared of it. So the problem is much harder.

Michael Vassar: You could appeal to people’s sense of wonder. If you’ve ever interacted with an Alzheimer’s sufferer or someone who has a mental disability, you might have gotten a sense that they were missing something. Well, so are we. The gap between them and us is practically adjacency in the space of possibilities. We’re probably missing out on a great many things. Shouldn’t we try and fix things so we’re missing less?

II. Closing Thoughts (from Peter Thiel)

This course has largely been about going from 0 to 1. We’ve talked a lot about how to create new technology, and how radically better technology may build toward singularity. But we can apply the 0 to 1 framework more broadly than that. There is something importantly singular about each new thing in the world. There is a mini singularity whenever you start a company or make a key life decision. In a very real sense, the life of every person is a singularity.

The obvious question is what you should do with your singularity. The obvious answer, unfortunately, has been to follow the well-trodden path. You are constantly encouraged to play it safe and be conventional. The future, we are told, is just probabilities and statistics. You are a statistic. 

But the obvious answer is wrong. That is selling yourself short. Statistical processes, the law of large numbers, and globalization—these things are timeless, probabilistic, and maybe random. But, like technology, your life is a story of one-time events. 

By their nature, singular events are hard to teach or generalize about. But the big secret is that there are many secrets left to uncover. There are still many large white spaces on the map of human knowledge. You can go discover them. So do it. Get out there and fill in the blank spaces. Every single moment is a possibility to go to these new places and explore them. 

There is perhaps no specific time that is necessarily right to start your company or start your life. But some times and some moments seem more auspicious than others. Now is such a moment. If we don’t take charge and usher in the future—if you don’t take charge of your life—there is the sense that no one else will.

There are four basic explanations for such a strange, inverted distribution. The first two reflect the familiar nature vs. nurture debate:

      1. It is natural. Founders really are different. Max Levchin really has 3 kidneys.

      2. It is developed, or nurtured. Cultural feedback makes founders different.

But the nature vs. nurture paradigm assumes that the distribution is real. It may, in fact, be mythology. To the extent that it’s fictional, there are two explanations:

      3. It is self-created (exaggerated by the founders).

      4. It is other-created (exaggerated by everyone else).

Thinking about founders involves thinking about which of these explanations fit and which do not. The complicated answer is that generally all four apply to some extent. It is very hard to disaggregate them. In practice, they tend to all feed into each other in important but complicated ways. 

The dynamic might work like this. People start out being different. They are nurtured to develop their already somewhat extreme traits. Those traits become more important, and they learn to exaggerate them. Others perceive that inflated importance and exaggerate in turn. The founders thus end up being even more different than they were before. And we cycle and repeat.

In practice, the arrows could be reversed. Or the interactions might not make a clean circle, and the feedback loop would be much more complicated. The point is that some interactive combination, and not just one static piece, is driving the process.

D. Applied

Anecdotally, we can apply this framework to any founder figure.

Take Sir Richard Branson, for instance. The big question is whether Branson should be king. He has been called:

The king of publicity;

The Virgin king; 

King of the desert (and space);

The king of branding;

The ice king; and even

King of the Muppets   

Let’s start with the haircut. He sort of looks like a lion. In fact, in the picture above he is actually dressed up as a lion. It seems kind of redundant. Anyway, one suspects that Branson wasn’t actually born with that exact hairstyle. There is probably some degree to which he cultivated and nurtured his traits over time.  Reconstructing the truth is tricky. It is very hard to actually know the precise dynamic—nature, nurture, or some kind of fiction—because stories about heroic founder figures get told in very exaggerated, morphed forms. 

Jack Dorsey is another figure we can pick on. He’s hit all of the extremes and very little of the average. At the outset he donned a nose ring and unkempt hair. He got a nerdy tattoo. Then he transformed to the other extreme side of the inverse distribution. Now he wears Prada suits and fashionable shirts. His branding went from extreme outsider to extreme insider. And this is all going off nothing but totally superficial appearances.

Sean Parker might be the paradigmatic example of the extreme founder figure. There was a rise, fall, rise, fall, and then a rise again. His experience in founding multiple things has been a pastiche of extremes. He didn’t go to college. Maybe he didn’t even finish high school. He was involved in various underground hacking circles in ‘90s. He did Napster as teenager. That had a crazy up-down arc to it. Criminal, of course, is the ultimate outsider category. There were all sorts of questions on whether Napster was really a criminal undertaking. Per the Digital Millennium Copyright Act, companies had to list a phone number for people to call for support inquires. At Napster, that number was Sean’s cell phone. He spent a lot of time in the early 2000s assuring concerned Midwestern mothers that their children weren’t going to get locked up for having downloaded a Metallica album. 

And then there are the wacky drug allegations and the crazy celebrity part. Sean made the cover of the Forbes 400 issue; he found a way to be distinctive even amongst the set of the richest people in the world. Justin Timberlake, of course, played Sean in the Facebook movie. There is a person at Clarium who looks pretty similar to these guys. When he travels outside of Silicon Valley, people ask him if he’s Justin Timberlake. But in Silicon Valley, people ask him if he’s Sean Parker.

Sean seems as exciting to people as he does dangerous. One random anecdote involves the Founders Fund surfing trip to Nicaragua over New Year’s 2007. We took the jet down to Managua. We were probably the only people in the country with a private jet. We drove to a remote town on the coast. Everything started off great. We threw a terrific New Year’s party. Except it kept getting crazier and crazier. Our professional security guard had to displace some people when various drug dealers and other sketchy types started showing up. In Sean’s mind at least, things got weirder from there. 36 hours later, by the morning of Jan 2nd, Sean was all but convinced that our security guard was plotting against him and he was about to be kidnapped. He went from extreme insider to extreme outsider very quickly. He and his girlfriend ditched their luggage and fled to Managua international airport in a cab. The rest of us thought that this was exaggerated paranoia, so we stayed as planned. Sure enough, the security guard became visibly distressed when he noticed Sean was no longer there. We nervously told him that Sean had mentioned that he’d be leaving tomorrow—that way he’d already be gone when they tried to nab him at the airport. Ultimately there was a happy ending and no one got kidnapped. But there will probably not be any more Founders Fund trips to Nicaragua.

This segues to the pure celebrity version, best epitomized by Lady Gaga. Born This Way is her recent hit album and song. On one level, the whole thing is obviously completely fictional. It’s probably safe to say that she was not, in fact, born like this. The big piece must be nurture. But on another level, maybe it is nature. What sort of people would actually do this to themselves? Maybe one actually does have to be born that way in order to do these things. Who really knows for sure? Is Gaga self-created myth? A myth created by other people? Everything all pulled together at once?

II. Mythology 

Oddly enough, classical mythology overlaps with the inverted bell curve distribution. There are monsters and there are gods. And very often they are one and the same.

In what sense are founders like mythical heroes? Myths about the founding of things are very common. Are mythical heroes actually any different? Did they have extreme traits? Develop them? Did they exaggerate themselves? Did others exaggerate their stories?

Consider Oedipus.  He was both an extreme insider and an extreme outsider. He was the king. He was so brilliant that he was able solve the riddle of the sphinx. But he was abandoned to die on a hill as an infant. He was a foreigner from a different place. And then he had the incest accusations and ensuing downfall.

Achilles is another mythological hero who was active at the extremes. He was incredibly strong and perfect, except where he was weak and flawed. 

Perhaps the most classic founding of all is the founding of Rome. Romulus and Remus were disadvantaged, common orphans who were raised by wolves. They were outsiders. But then they became founders and lawmakers. Romulus killed his brother and became a lawbreaker and king. If there is a hierarchy to it—if killing your brother is worse than killing a random person and killing your twin brother is even worse than that—then Romulus was an unusually bad criminal. 

Legend has it that what prompted the murder is Remus’ leaping over the imaginary boundary line that Romulus had established as the edge of Rome. The rule was codified with blood: anyone who jumps over the walls of Rome will be destroyed. Does this make Romulus a criminal outlaw? Or does it make him the king who defined Rome? It depends. Maybe he was both.

Remus obviously had a bad ending. Romulus’ ending is more ambiguous. In Livy’s account, there was a huge storm that terrified the people. When the storm cleared up, Romulus had disappeared. It was announced that he had become a god. But Livy also notes an alternate account; a group of conspiratorial senators caught up with Romulus and used the chaos of the storm as cover to kill him and dispose of the body.

One other mythical element was the 12 eagles that Romulus saw from Palatine Hill. They stood for the 12 centuries that Rome would endure, after which point the debt of the founding crime would have to be repaid. Approximately 12 centuries later, Attila the Hun apparently thought it would be a good idea to copy Romulus, and killed his brother Bleda. Incidentally, fratricide is probably no longer best practice for founding things.

III.  Archaic Cultures

A. The Sacrificial Cycle

The founder/extremeness/infamous dynamic, or something very much like it, was an incredibly important part of ancient cultures. The fundamental problem in these cultures was that there were all sorts of conflicts everywhere. People didn’t know what to do. There were no rules—a striking parallel to the tech startup context. Amidst all the chaos there was war of all against all.

Various enlightenment theorists have insisted that, to escape this warring state of nature, people got together, had a good chat, and drew up a social contract. But nothing of the sort ever happened. Where warring civilizations didn’t just collapse entirely, the most common resolution involved polarizing and channeling all the hostility into one particular person. Depending on the culture, witches were burned or people had their hearts cut out. The details differed. But the dynamic—a crazed community rallying around the sacrificial scapegoat—was the same. 

In cultures that had some degree of permanence, this became a cyclical process. Absent strong institutions, peace never lasted. Things would go wrong. Maybe disease struck. Or maybe there was some other kind of internal (and less often, external) conflict that led to complete chaos. And then people would gang up, unite against a scapegoat, and perform the sacrifice. Peace was restored. And the cycle repeated ad infinitum.

It’s clear that the scapegoat is extremely powerful. Scapegoats can turn conflict into peace. This makes the scapegoat omnimalevolent; if peace follows his killing, he must have been very bad indeed. Or maybe it’s omnibenevolent, since it trades its life so that others may live in peace. Probably the right answer is that it’s some of both.

We can speculate that in many cultures, this process became ritualized. People realized the power of the scapegoat and abstracted it away from localized contexts. Instead of waiting for random uncontrolled chaos, sacrifice became planned. Of course, there were probably cultures that never figured this out. They couldn’t systematize the isolation of the scapegoat. So everyone just killed everyone and the culture blew up. One suspects that the cultures that managed to ritualize and repeat the cycle were the ones who lasted for awhile. 

B. Finding the Victim

There are all sorts of questions on how to go about finding the scapegoat. Sometimes the processes are random. In Gaelic Scotland, people would bake a cake over the fires of Beltane and cut it into pieces. One piece would be marked with charcoal. Men would choose a piece from a bonnet. Whoever got the black piece was the devoted, and was sacrificed to Baal. Residual forms of this persisted up through the 18th century, where the devoted would have to just jump through the fire instead of perish in it.

The ancient Gauls took a more objective approach. Someone would have to be sacrificed on the eve of battle to win the gods’ favor. But which person would that be? Rather than complicate things, the Gauls just held a footrace to the battlefield. Naturally, the slowest person was the one who got sacrificed.

C. Anatomy of the Scapegoat

The perfect scapegoat is someone at both extremes. He must be both an extreme outsider and an extreme insider. It can’t be a completely random person drawn from a homogeneous lot. It must be some sort of outsider, lest the people in the crowd get introspective and realize that the sacrificed was essentially just like them (and, next time, may well be them). But neither can the scapegoat be entirely different from the crowd; he must be an insider, since the pretext behind the ritual is that he is responsible for the internal community strife. 

D. The Roots of Monarchy 

Not all scapegoats were hated all the time. Very often, they would be worshipped before they were sacrificed. People would give the scapegoat a certain amount of power before tearing him apart. That scapegoats were either worshipped or demonized follows from their being all-powerful.

One working theory is that monarchy originated this way. The Aztecs, for instance, would basically crown someone a quasi-god king for a period of time, after which point he would be sacrificed. Kings became scapegoats who had not yet been killed. Every king was a living god. Every god was a murdered king. 

Arguably Egyptian pharaohs started off as scapegoats. Perhaps the first pyramids were the piles of stones that entombed people who were stoned to death. Later, when Pharaohs became powerful kings and it was unthinkable to kill them during their lifetimes, people kept putting increasingly large piles stones on top of them after they died.

Given this dynamic, we can imagine how monarchy came into being. The scapegoat simply figured out how to maintain his power and indefinitely delay his execution. 

The Zulu Kingdom was a warlike African monarchy in the 19th century. The Zulu king had to be strong and powerful. He could have hundreds of wives and do pretty much whatever else that we wanted. But once he started to get white hair and wrinkles, his power faded. He would be deemed unfit to be king, deposed, and killed. It’s hardly a surprise, then, that upon first contact with the British, the Zulu kings were more interested in hair coloring lotion than in anything else. Whether phenomena like this continue to exist in our society today is a question well worth asking. 

E. The Politics of Sacrifice

According to Aristotle, tragedy functioned so as to reduce common peoples’ anger toward successful people. The lesson in all tragedy is that even the greatest people have tragic flaws. Everybody falls. It was thus cathartic for ordinary people to see terrible things happen to extraordinary people, if only on stage. Tragedies were political tools that transformed envy and anger into pity. Commoners would retreat contentedly to their small houses instead of plotting against the upper class.

Julius Caesar was a classic insider/outsider. Eventually, of course, he was assassinated. Every subsequent Roman emperor pretty much had to be a Caesar. And the sacrificial cycle repeated an infinitum for centuries thereafter. 

Being an extreme insider is great, until it all goes very wrong. Marie Antoinette was such an insider. But people turned on her. She was an Austrian, i.e. a foreigner. She faced accusations strikingly similar to those from the Oedipus mythology. It’s not clear whether the “let them eat cake” line was fictitious or not. But all great revolutions could be described as the rapid shift from insider to outsider. During the French Revolution, there was an interesting legal debate on whether the king should get a trial. Robespierre and the revolutionaries vehemently argued against a trial. The king, they should, should be slaughtered like a wild beast. Having a trial meant that the king might be innocent, which, in turn, meant that the people might be guilty. But it was unthinkable that the people might be guilty. So the solution was to just kill the king. 

IV. Sacrifice Endures 

A. In Culture 

A modern version of this is the 12-person jury in the criminal context. The unlucky 13th person is the criminal who gets punished or killed. It is the classic scapegoating-type mechanism. The 13th person is assumed to be—and probably is—different. It’s never really a jury of your peers. If you’re a murderer, you aren’t judged by 12 murderers. If you’re rich, they don’t find 12 rich people to decide your fate. It is very much unclear whether a jury trial works well for its stated goals at all. It seems to work in contexts where people perceive things as they are. But other contexts, it is just scapegoating gone crazy. 

Another modern version has to do with celebrities, and resurrects the monarchical dynamic that people assume has long since died. We literally anoint our stars as kings. Elvis was the King of Rock. Michael Jackson was the King of Pop. Brittney Spears was the Princess of Pop—I guess Madonna was the Queen. You start to run out of titles pretty quickly.

Then, at some point, things go wrong. The anointed are put on pedestals only to be torn down. Elvis self-destructed in the ‘70s. Michael Jackson obviously went downhill. The picture below depicts Brittney Spears at height of the paparazzi insanity. A few years ago, the paparazzi industry was a $400 million/year industry. Brittney Spears drove $100 million of that. There were between 1,000 and 2,000 people who made their living doing nothing but chasing her around and taking pictures of her. What went wrong? Was Brittney naturally crazy? Did she become crazy after having been isolated as a child superstar? Maybe the crowd got to her. Or maybe she intentionally acted in weird ways for the publicity.

Regardless, these kind of stars all enjoy a very strange afterlife. In life, they are torn down from pedestals. But after they die, they are resurrected as god-kings. Things come full circle. 

Another example of this is the Forever 27 club, whose members include Janice Joplin, Jimi Hendrix, Jim Morrison, Kurt Cobain, Amy Winehouse, etc. This is the set of famous musicians who all died at age 27. “They tried to make me go to rehab, I said, ‘No, no, no.’” There are all sorts of questions one could ask. But there is a sense in which these people will live on as iconic cultural figures.

The “from destructive to immortalized” dynamic goes way back to mythology. Alexander the Great was 32 when he died. He would frequently engage in hardcore quasi-religious drinking marathons. Apparently the game was to consume alcohol until someone died, and Alexander felt that he had to prove that that someone would not be him. It was a strategic error. But he will forever be known as a great conqueror.

B. In Politics 

The political version involves certain ideological distortions. People on the left and the right tend to focus and even obsess on people from the other side. Everybody from the other column becomes the crazy person and the legitimate scapegoat. In reality, the truth is that it tends to involve some strange combination of both. 

Two of our greatest presidents had this sort of strange heroic arc to their story. Abraham Lincoln was an extreme outsider turned insider. He was born in an isolated log cabin. He was probably our poorest President. He was very smart and also very ugly. And he, probably intentionally, uglified himself even further with his strange beard. Lincoln was always on both extremes. His end involves a very strange return to the Cesar question. John Wilkes Booth, believing that he was reenacting Cesar assassination, shouted “Sic semper tyrannis” as he shot Lincoln—which is, of course, what Brutus is reputed to have said as he stabbed Caesar.

A strange counterpoint point to this comes from one of Lincoln’s first public speeches ever. The future president delivered what is now called the Lyceum Address to a small crowd in Springfield Illinois in 1837, when he was 28 years old. It is worth reading in its entirety. It opens:

As a subject for the remarks of the evening, "The perpetuation of our political institutions" is selected.

Lincoln spoke about how there could not be any more founding moments in the United States. The founding had been done, in the 18th century. It was over. At this point all that one could do was preserve and maintain things. There was nothing truly new that anyone could ever hope to do in our government.

About halfway through the speech, things get really interesting. Lincoln asks whether ambitious people would ever try to be founders anyways, or whether they would be fully satisfied with existing institutions. He answers yes and no, respectively:

The question then is, Can that gratification be found in supporting and maintaining an edifice that has been erected by others? Most certainly it cannot. Many great and good men, sufficiently qualified for any task they should undertake, may ever be found whose ambition would aspire to nothing beyond a seat in Congress, a gubernatorial or a presidential chair; but such belong not to the family of the lion or the tribe of the eagle. What! think you these places would satisfy an Alexander, a Caesar, or a Napoleon? Never! Towering genius disdains a beaten path. It seeks regions hitherto unexplored. It sees no distinction in adding story to story upon the monuments of fame erected to the memory of others. It denies that it is glory enough to serve under any chief. It scorns to tread in the footsteps of any predecessor, however illustrious. It thirsts and burns for distinction; and if possible, it will have it, whether at the expense of emancipating slaves or enslaving freemen. Is it unreasonable, then, to expect that some man possessed of the loftiest genius, coupled with ambition sufficient to push it to its utmost stretch, will at some time spring up among us?

The takeaway is that we have to be really careful because such people might exist.

Kennedy’s story was different but the underlying dynamic was the same. He was one of richest people—worth about $1 billion in today’s money—to become president. His father was criminal bootlegger. He was on amphetamines most of the time. He stopped being a rich insider when he found himself an outsider to whatever plot or conspiracy it was that led to his assassination. 

C. In Tech Companies

This dynamic recurs over and over again in the tech company founder context. Let’s focus on 3 instances: Bill Gates, Howard Hughes, and Steve Jobs. 

Those old enough to remember will remember the “Bill Gates is god” phase in the ‘90s. The president of the U.S. always has a quasi-divine status. So when you get compared to the sitting president, it’s pretty extreme. All the same questions apply to Gates. Was it nature or nurture? He was a Harvard insider but a dropout outsider. He wore big glasses. Did he become a nerd unwillingly? Did he prosper by accentuating his nerdiness? It’s hard to tell.

What is clear, however, is that the good times didn’t last:

One (admittedly unconventional) theory is that Bill Gates is still being tortured and punished for his fall. He has to go to all sorts of boring charity events, pretend that the people there are saying interesting things, and then give them his money to boot. And adding insult to injury is the fact that these are the same people who ganged up on him in the late ‘90s. 

Howard Hughes was one of the greatest founders in the 20th century. His life had a very extraordinary arc to it from about 1930 to 1945. He started off as reasonably successful. He went on to have incredible parallel careers in movies and aviation, which, in retrospect, were the two booming tech sectors of the 1930s. He became the wealthiest person in the U.S. by age 45. If Hughes had died in the plane crash that he had in 1946, he would have gone down as greatest entrepreneur of 20th century.

One of Hughes’ favorite tricks was to pretend to be crazy on the theory that no one would take the time and energy to try to stop or compete with a crazy person. A large part of his mythology was fictionally constructed; he claimed, for instance, to have been born on December 25th, 1905. One has to wonder if he was really born on the same day of the year as Christ, or whether that was an intentional ploy.

Hughes’ fall from grace began after the’46 crash, when he became addicted to painkillers. He more or less holed up in various penthouse lofts for 30 years, hooked up to IV machines and refusing to eat. Looking back the story has a pretty crazy color to it. The craziness continued even after Hughes died; as there was no authoritative will, all sorts of distant descendants and questionable figures began a long and vicious fight to inherit the estate.

And then there’s the Steve Jobs version. You could probably tell a few different versions of the Jobs version. Let’s focus on the one from the ‘70s and ‘80s. He had all the classic extreme outsider and extreme insider traits. He dropped out of college. He was eccentric and had all these crazy diets. He started out phreaking phones with Steve Wozniak. He took LSD.

Ultimately he was kicked out of apple and was replaced John Sculley, who was seen as the much more normal, adult-type person that should be in charge. 

Circling back to the bit about archaic cultures. Isn’t this dynamic roughly the same now as it was then? We tend to think of monarchy as a dead and defunct institution. But is it really? Time magazine put Marc Andreessen on the cover in February 1996—sitting on a throne-like chair! He was later vilified quite a bit when things went bad at Netscape. Now he seems to have recovered quite nicely. 

Mary Meeker had a similar rise and fall and then rise again. Dubbed the "Queen of the Net," Meeker was an influential stock market analyst who was probably the most bullish person on net in the ‘90s. If she wrote about your company, your stock would go up. She received a much more negative reassessment from the public after the ‘90s tech bubble exploded. She was torn down from the pedestal. But she stuck through it at Morgan Stanley and has come back to being very successful, now as a venture capitalist.

D. Can It Be Escaped? 

How much of this can be avoided? How do you avoid becoming a sacrificial victim? The simple answer, of course, is that if you really don’t want to get killed, you shouldn’t sit on the throne. But this seems suboptimal. Wearing the crown is obviously an attractive thing. The question is whether you can decouple it with getting executed.

That is the danger with being an extreme insider. Push too hard and the poles reverse; you end up as an extreme outsider and it all goes to pot. There have been 44 American presidents. Four of them—9% of presidents—were assassinated while in office. Four more were almost killed. Your odds of not dying a violent death are dramatically lower if you’re not the president. That’s at least worth thinking about if being president is your goal.

This is not to say that people can or should escape by abdicating the throne. Sometimes the risk is worth it. And maybe you can reduce the risk. There have to be CEOs and founders. Those people are expected to wear the crown. That necessarily involves a certain amount of playing with fire. The tricky part is that, while mistakes get made, they are incredibly hard to spot at the time. They are more easily analyzed in retrospect. Bill Gates was incredible through the 1990s, until Larry Ellison and Scott McNealy and a bunch of CEOs from other tech companies effectively started a “We Hate Gates” club, stirred up attention at the DOJ, etc. From Gates’ perspective, he was on perpetual winning arc of never-ending progress. Everything was perfect, and the haters were just envious and pathetic. But once it turns it can turn pretty quickly. The falls are so big that it’s hard to fully recover.

V. Extending the Founding 

A. Forms and Theory

One strategy to avoiding becoming a scapegoat is to extend the founding moment. With the big caveat that there is probably no single silver bullet solution—the founder turned god turned victim dynamic is probably inescapable to some extent—let’s work through some ideas on how to negotiate this dangerous ground.

You can plot out the various forms of government on 1-dimentional axis:

A startup is basically structured as a monarchy. We don’t call it that, of course. That would seem weirdly outdated, and anything that’s not democracy makes people uncomfortable. But look at the org chart:

It is certainly not representative governance. People don’t vote on things. Once a startup becomes a mature company, it may gravitate toward being more of a constitutional republic. There is a board that theoretically votes on behalf of all the shareholders. But in practice, even in those cases it ends up somewhere between constitutional republic and monarchy. Early on, it’s straight monarchy. Importantly, it isn’t an absolute dictatorship. No founder or CEO has absolute power. It’s more like the archaic feudal structure. People vest the top person with all sorts of power and ability, and then blame them if and when things go wrong.

We are biased toward the democratic/republican side of the spectrum. That’s what we’re used to from civics classes. But the truth is that startups and founders lean toward the dictatorial side because that structure works better for startups. It is more tyrant than mob because it should be. In some sense, startups can’t be democracies because none are. None are because it doesn’t work. If you try to submit everything to voting processes when you’re trying to do something new, you end up with bad, lowest common denominator type results.

But pure dictatorship is unideal because you can’t attract anyone to come work for you. Other people want some power and control too. So the best arrangement is a quasi-mythological structure where you have a king-like founder who can do more than in a democratic ruler but who remains far from all-powerful.

B. Occupy

We can reimagine our old 0 to 1 (technology) and 1 to n (globalization) paradigm by putting a monarchy/democracy overlay on it. Monarchy involves going from 0 to 1. Democracy involves going from 1 to 99.

The 99% vs. the 1% is the modern articulation of this classic scapegoating mechanism. It is all minus one versus the one. And it has to just be the one. 99.99 people or percent is too granular. Scapegoating 0.1 doesn’t really work. You need a whole person to play the victim. Similarly, 98-2 doesn’t quite have the same ring to it either.  

C. Extending the Moment, Escaping the Trial

The normal company arc involves an initial monarchical founding period and then a normal period where founders are gone and more conventional people come in and run things. In the U.S., there were the founding fathers. And then there have been everybody else. Perhaps some figures like Lincoln or FDR were exceptions to this. But the two phases are generally clear and distinct.

If you want to be a founder and stay a founder, can you extend the founding period? In tech companies, foundings last as long as technological innovation continues. The question is thus how long it takes for the substantive technology focus to yield to process. Once you shift toward ossified, process-based normality, much less gets done. Every founder would thus to do well to never stop wondering whether there are strategies to extend the founding in one form or another. 

This probably requires a healthy amount of paranoia. You might conceive of every board meeting as a trial. At best, the board is jury (though probably not of your peers). At worst, it is a mob and is looking to make you the sacrificial victim. Your job as founder is to survive the trial. You must make sure that you do not get executed. The boardroom is not the only place where things can go wrong, of course. But it is typically where things go wrong internally, and most fatal wounds come from internal, not external conflict.

Even something as seemingly innocuous as holding the title of CEO may actually be quite dangerous. Maybe you can figure out ways to minimize it. Augustus never said he was king. It was dangerous to be a king after Brutus killed Caesar.  So Augustus was just the “first among equals.” Whether that equality was anything more than pure fiction, of course, is very questionable.

In October of 2000, things were pretty crazy at PayPal. The burn rate was $10 million/month. There were about 4.5 months of runway left. When I returned as CEO, it wasn’t all of a sudden. I was the Chairman and came back as the interim CEO. We went through a 6-7 month process of trying to find a permanent one. The one decent candidate that we found sort of didn’t work out. Things were going well, so the board agreed to have me be CEO. But the company was about to go public, so the board insisted that there be a Chief Operating Officer (COO) too. COO, of course, is code for the #1 replacement candidate for CEO—it’s like the Vice President in U.S. politics, only more adversarial. I was able to convince the board to make David Sacks COO, which was probably a good, safe move since David was perceived to be crazier than I was. Thinking carefully about these things can lead to powerful insurance policies against getting deposed or executed at trials board meetings. 

The dual founder thing is worth mentioning. Co-founders seem to get in a lot less trouble than more unbalanced single founders. Think Hewlett and Packard, Moore and Noyce, and Page and Brin. There are all sorts of theoretical benefits to having multiple founders such as more brainstorming power, collaboration, etc.  But the really decisive difference between one founder and more is that with multiple founders, it’s much harder to isolate a scapegoat. Is it Larry Page? Or is it Sergey Brin? It is very hard for a mob-like board to unite against multiple people—and remember, the scapegoat must be singular. The more singular and isolated the founder, the more dangerous the scapegoating phenomenon. For the skeptic who is inclined to spot fiction masquerading as truth, this raises some interesting questions. Are Page and Brin, for instance, really as equal as advertised? Or was it a strategy for safety? We’ll leave those questions unanswered and hardly asked.

D. Return of the King

The return of the founder is not to be underestimated. Apple is the paradigmatic example. There were 12 crazy years from 1985 to ’97. There were very conventional CEOs. They couldn’t figure out anything new to build. Obviously there was something very powerful in bringing the founder back; from 1997-2011 Apple changed course entirely and had an incredibly powerful arc. 

The options backdating scandal has been relegated to a minor footnote in the Apple mythology. Apple stock kept going up, and the board kept backdating options grants, giving Steve Jobs a fairly big windfall. 

It probably wasn’t just building great products or being a good insider that saved Steve Jobs. His being terminally ill part was probably a very important variable. There is much less power in scapegoating someone who’s power—indeed, whose life—is waning anyways.

I met Steve Jobs once, at Marc Andreessen’s wedding in 2006. He was already very frail then. At 9 pm, he got up from the table and announced that he had to get back to the office to work. One couldn’t help but wonder: Was this real? Was Jobs really working this hard? Or was it an excuse? Maybe he was just bored talking to me.

Resurrections are possible. But you can only be resurrected after you die. Founders should think carefully about how to preserve the original founding moment for as long as possible. The key is to encourage and achieve perpetual innovation. It is very important to avoid, or at least delay, the shift to a horrible bureaucracy where no one can do anything and everyone is circumscribed.

The usual narrative is that society should be organized to cater to and reward the people who play by the rules. Things should be as easy as possible for them. But perhaps we should focus more on the people who don’t play by the rules. Maybe they are, in some key way, the most important. Maybe we should let them off the hook. 

So AI is a very large space—so large that people’s normal intuitions about its size are often off base by orders of magnitude.

One of the big questions in AI is exactly how smart it can possibly get. Imagine an intelligence spectrum with 3 data points: a mouse, a moron, and Einstein. Where would AI fall on that scale? 

We tend to think of AI as being marginally smarter than an Einstein. But it is not a priori clear why the scale can’t actually go up much, much higher than that. The bias is toward conceiving of things that are fathomable. But why is that more realistic than a superhuman intelligence so smart that it’s hard to fathom? It might be easier for a mouse to understand the relativity than it is for us to actually understand how an AI supercomputer thinks. 

A future with artificial intelligence would be so unrecognizable that it would unlike any other future. A biotech future would involve people functioning better, but still in recognizably human way. A retrofuture would involve things that have been tried before and resurrected. But AI has the possibility of being radically different and radically strange.

There is a weird set of theological parallels you could map out. God may have been to the Middle Ages what AI will become to us. Will the AI be god? Will it be all-powerful? Will it love us? These seem like incomprehensible questions. But they may still be worth asking.

II. The Strangeness of AI

The Turing test is the classic, decades-old test for AI that asks whether you can build a machine that behaves as intelligently as a human does. It focuses on the subset of human behavior that is intelligent. Recently the popular concern has shifted from intelligent computers to empathetic computers. People today seem more interested in whether computers can understand our feelings than whether they are actually smart. It doesn’t matter how intelligent it is in more classic domains; if the computer does not find human eye movement emotionally provocative, it is, like Vulcans, still somehow inferior to people. 

The history of technology is largely a history of technology displacing people. The plow, the printing press, the cotton gin all put people out of business. Machines were developed to do things more efficiently. But while displacing people is bad, there’s the countervailing sense that these machines are good. The fundamental question is whether AI actually replaces people or not. The effect of displacement is the strange, almost political question that seems inextricably linked with the future of AI. 

There are two basic paradigms. The Luddite paradigm is that machines are bad, and you should destroy them before they destroy you. This looks something like textile workers destroying factory cotton mills, lest the machines take over the cotton processing. The Ricardo paradigm, by contrast, holds that technology is fundamentally good. This is economist David Ricardo’s gains from trade insight; while technology displaces people, it also frees them up to do more. 

Ricardian trade theory would say that if China can make cheaper cars than can be made in the U.S., it is good for us to buy cars from China. Yes, some people in Detroit lose their jobs. But they can be retrained. And local disturbances notwithstanding, total value can be maximized.

The charts above illustrate the basic theory. With no trade, you get less production. With joint production and specialization, you expand the frontier. More value is created. This trade framework is one way to think about technology. Some cotton artisans lose their jobs. But the price of shirts from the cotton factory falls quite a bit. So the artisans who find other jobs are now doing something more efficient and can afford more clothes at the same time.

The question is whether AI ends up being just another version of something you trade with. That would be straight Ricardo. There’s a natural division of labor. Humans are good at some things. Computers are good at other things. Since they are each quite different from each other, the expected gains from trade are large. So they trade and realize those gains. In this scenario AI is not substitute for humans, but rather a compliment to them. 

But this depends on the relative magnitudes of advantage. The above scenario plays out if the AI is marginally better. But things may be different if the AI is in fact dramatically better. What if it can do 3000x what humans can do across everything? Would it even make sense for the AI to trade with us at all? Humans, after all, don’t trade with monkeys or mice. So even though the Ricardo theory is sound economic intuition, in extreme cases there may be something to be said for the Luddite perspective.

This can be reframed as a battle over control. How much control do humans have over the universe? As AI becomes stronger, we get more and more control. But then AI hits an inflection point where it goes superhuman, and we lose control altogether. That is qualitatively different from most technology, which gives people more control over the world with no end. There is no cliff with most technology. So while computers can give us a great deal of control, and help us overcome chance and uncertainty, it may be possible to go too far. We may end up creating a supercomputer in the cloud that calls itself Zeus and throws down lightning bolts at people. 

III. The Opportunity of AI 

Hugeness and strangeness are interesting questions. But whether and how one can make money with AI may be even more interesting. So how big is the AI opportunity?

A. Is It Too Early for AI?

Everything we’ve talked about in class remains important. The timing question is particularly important here. It might still be too early for AI. There’s a reasonable case to be made there. We know that futures fail quite often. Supersonic airplanes of the ‘70s failed; they were too noisy and people complained. Handheld iPad-like devices from the ‘90s and smart phones from ’99 failed. Siri is probably still a bit too early today. So whether the timing is right for AI is very hard to know ex ante.

But we can try to make the case for AI by comparing it to things like biotech. If you had a choice between doing AI and the biotech 2.0 stuff we covered last class, the conventional view would be that the biotech angle is the right one to pick. Arguably the bioinformatics revolution is being or will soon be applied to humans, whereas actual application of AI is much future out. But the conventional view isn’t always right. 

B. Unanimity and Skepticism

Last week in Santa Clara there was an event called “5 Top VCs, 10 Tech Trends.” Each VC on the panel made 2 predictions about technology in the next 5 years. The audience voted on whether they agreed with each prediction. One of my predictions was that biology would become an information science. When the audience voted, it was a sea of green. 100% agreed with that prediction. There wasn’t a single dissenter. Perhaps that should make us nervous. Unanimity in crowds can be very disconcerting. Maybe it’s worth questioning the biotech-as-info-science thesis a little bit more.

The single idea that people thought was the worst was that all cars would go electric. 92% of the audience voted against that happening. There are many reasons to be bearish on electric cars. But now there is one less.

The closest thing to AI that was discussed was whether Moore’s law would continue to accelerate. The audience was split 50-50 on that. If it can accelerate—if it can more than double every 18th months going forward—it would seem like you’d get something like AI in just a few years. Yet most people thought AI was much further away than biotech 2.0. 

C. (Hidden) Limits

One way to compare biotech and AI is to think about whether there are serious—and maybe even hidden—limits in each one. The biotech revolution narrative is that we’re going to figure out how to reverse and cure all sorts of maladies, so if you just live to x, you can stay alive forever. It’s a good narrative. But it’s also plausible that there are invisible barriers lurking beneath the surface. It’s possible, for example, that various systems in the human body act against one another to reach equilibrium. Telomerase helps cells split unbounded. This is important because you stop growing and start to age when cells don’t split. So one line of thinking is that you should drink red wine and do whatever else you can to keep telomerase going.

The challenge is that unbounded cell splitting starts to look a lot like cancer at some point. So it’s possible that aging and cancer have the effect of cancelling each other out. If people didn’t age, they would just die of cancer. But if you shut down telomerase sooner, you just age faster. Fix one problem and you create another. It’s not clear what the right balance is, whether such barriers can be overcome, or, really, whether these barriers even exist.

A leading candidate for an invisible barrier in AI is the complexity of the code. The might be some limit where the software becomes too complicated as you produce more and more lines of code. Past a certain point, there is so much to keep track of that no one knows what’s going on. Debugging becomes difficult or impossible. Something like this could be said to have happened to Microsoft Windows over a number of decades. It used to be elegant. Maybe it has been or can be improved a bit. But maybe there are serious hidden limits too it. In theory, you add more lines of code to make things better. But maybe they will just make things worse. 

The fundamental tension is exponential hope versus asymptotic reality. The optimistic view is the exponential case. We can argue for that, but it’s sort of unknown. The question is whether and when asymptotic reality sets in.

and the AI version:

D. AI Pulls Ahead 

There are many parallels between doing new things in biotech and AI. But there are three distinct advantages to focusing on AI:

Engineering freedom

Regulatory freedom

Underexplored (contrarian)

Engineering freedom has to do with the fact that biotech and AI are fundamentally very different. Biology developed in nature. Sometimes people describe biological processes as blueprints. But it’s much more accurate to describe them as a recipe. Biology is a set of instructions. You add food and water and bake for 9 months. There is a whole series of constructions like this. If the cake turns out to have gotten messed up, it’s very hard to know how to fix it simply by looking at the cookbook. 

This isn’t a perfect analogy. But directionally, AI is much more of a true blueprint. Unlike recipe-based biotech, AI is much less dependent on a precise sequence of steps. You have more engineering freedom to tackle things in different ways. There is much less freedom in changing a biological recipe than there is in designing a blueprint from scratch. 

On the regulatory side, the radical difference is that biotech very heavily regulated. It takes 10 years and costs $1.3 billion to develop a new drug. There are lots of precautionary principles at work. There are 4,000 people at the FDA.

AI, by contrast, is an unregulated frontier. You can launch just as quickly as you can build software. It might cost you $1 million, or millions. But it won’t cost $1 billion. You can work from your basement. If you try to synthesize Ebola or smallpox in your basement, you could get in all sorts of trouble. But if you just want to hack away at AI in your basement, that’s cool. Nobody will come after you. Maybe it’s just that politicians and bureaucrats are weird and have no imagination. Maybe the legislature simply has no mind for AI-kind of things. Whatever the reason, you’re free to work on it.

AI is also underexplored relative to biotech. Picture a 2x2 matrix; on one axis you have underexplored vs. heavily explored. On the other you have consensus vs. contrarian. Biotech 2.0 would fall in the heavily explored, consensus quadrant, which, of course, is the worst quadrant. It is the new thing. The audience in Santa Clara last week was 100% bullish on it. AI, by contrast, falls in the underexplored, contrarian quadrant. People have been talking about AI for decades. It hasn’t happened yet. Many people have thus become quite pessimistic about it, and have shifted focus. That could be very good for people who do want to focus on AI.

PayPal, at Luke Nosek’s urging, became the first company in the history of the world that had cryogenics as part of the employee benefits package. There was a Tupperware-style party where the cryogenics company representatives made the rounds trying to get people to sign up at $50k for neuro or $120k for full body. Things were going well until they couldn’t print out the policies because they couldn’t get their dot matrix printer to work. So maybe the way get biotech to work well is actually to push harder on the AI front.

IV. Tackling AI  

We have people from three different companies that are doing AI-related things here to talk with us today. Two of these companies—Vicarious and Prior Knowledge—are pretty early stage. The third, Palantir, is a bit later. 

Vicarious is trying to build AI by develop algorithms that use the underlying principles of the human brain. They believe that higher-level concepts are derived from grounded experiences in the world, and thus creating AI requires first solving a human sensory modality. So their first step is building a vision system that understands images like humans do. That alone would have various commercial applications—e.g. image search, robotics, medical diagnostics—but the long-term plan is to go beyond vision and build generally intelligent machines.

Prior Knowledge is taking a different approach to building AI. Their goal is less to emulate brain function and more to try to come up with different ways to process large amounts of data. They apply a variety of Bayesian probabilistic techniques to identifying patterns and ascertaining causation in large data sets. In a sense, it’s the opposite of simulating human brains; intelligent machines should process massive amounts of data in advanced mathematical ways that are quite different from how most people analyze things in everyday life.

The big insight at Palantir is that the best way to stop terrorists isn’t regression analysis, where you look at what they’ve done in the past to try to predict what they’re going to do next. A better approach is more game theoretic. Palantir’s framework is not fundamentally about AI, but rather about intelligence augmentation.It falls very squarely within the Ricardo gains from trade paradigm. The key is to find the right balance between human and computer. This is a very similar to the anti-fraud techniques that PayPal developed. Humans couldn’t solve the fraud problem because there were millions of transactions going on. Computers couldn’t solve the problem because the fraud patterns changed. But having the computer do the hardcore computation and the humans do the final analysis, while a weaker form of AI, turns out to be optimal in these cases.

So let’s talk with D. Scott Brown from Vicarious, Eric Jonas from Prior Knowledge, and Bob McGrew from Palantir.

V. Perspectives

Peter Thiel:  The obvious question for Vicarious and Prior Knowledge is: why is now the time to be doing strong AI as opposed to 10-15 years from now?

Eric Jonas: Traditionally, there hasn’t been a real need for strong AI. Now there is. We now we have tons more data than we’ve ever had before. So first, from a practical perspective, all this data demands that we do something with it. Second, AWS means that you no longer need to build your own server farms to chew through terabytes of data. So we think that a confluence of need and computing availability makes Bayesian data crunching make sense. 

Scott Brown: If current trajectories hold, in 14 years the world’s fastest supercomputer will do more operations per second than the number of neurons in the brains of all living people. What will we do with all that power? We don’t really know. So perhaps people should spend the next 13 years figuring out what algorithms to run. A supercomputer the size of the moon doesn’t do any good on it’s own. It can’t be intelligent if it’s not doing anything. So one answer to the timing question is simply that we can see where things are going and we have the time to work on them now. The inevitability of computational power is a big driver. Also, very few people are working on strong AI. For the most part, academics aren’t because their incentive structure is so weird. They have perverse incentive to make only marginally better things. And most private companies aren’t working on it because they’re trying to make money now. There aren’t many people who want to do a 10-year Manhattan project for strong AI, where the only incentives are to have measurable milestones between today and when computers can think.

Peter Thiel:  Why do you think that human brain emulation is the right approach?

Scott Brown: To clarify, we’re not really doing emulation. If you’re building an airplane, you can’t succeed by making a thing that has feathers and poops. Rather, you look at principles of flight. You study wings, aerodynamics, lift, etc., and you build something that reflects those principles. Similarly, we look at the principles of the human brain. There are hierarchies, sparsely distributed representations, etc.—all kinds of things that represent constraints in the search space. And we build systems that incorporate those elements. 

Peter Thiel:  Without trying to start a fistfight, we’ll ask Bob: Why is the correct intelligence augmentation, not strong AI?

Bob McGrew: Most successes in AI haven’t been things that pass Turing tests. They’ve been solutions to discrete problems. The self-driving car, for instance, is really cool. But it’s not generally intelligent. Other successes, in things like translation or image processing, have involved enabling people to specify increasingly complex models for the world and then having computers optimize them. In other words, the big successes have all come from gains from trade. People are better than computers at some things, and vice versa. 

Intelligence augmentation works because it focuses on conceptual understanding. If there is no existing model for a problem, you have to come up with a concept. Computers are really bad at that. It’d be a terrible idea to build an AI that just finds terrorists. You’d have to make a machine think like a terrorist. We’re probably 20 years away from that. But computers are good at data processing and pattern matching. And people are good at developing conceptual understandings. Put those pieces together and you get the augmentation approach, where gains from trade let you solve problems vertical by vertical.

Peter Thiel:  How do you think about the time horizon for strong AI? Being 5-7 years away from getting there is one thing. But 15-20 years or beyond is quite another.

Eric Jonas: It’s tricky. Finding the right balance between company and research endeavor isn’t always straightforward. But our goal is simply to build machines that find things in data that humans can’t find. It’s a 5-year goal. There are compounding returns if we build these Bayesian systems so that they fit together. The Linux kernel is 30 million lines of code. But people can build an android app on top of that without messing with those 30 million lines. So we’re focusing on making sure that what we’re building now can be useable for the big problems that people will tackle 15 years from now. 

Peter Thiel:  AI is very different from most Silicon Valley companies doing web or mobile apps. Since engineers seem to gravitate toward those kind of startups, how do you go about recruiting?

Scott Brown: We ask people what they care about. Most people want to make an impact. They may not know what the best way to do it is, but they want to do it. So we point out that it’s hard to do something more important than building strong AI.  Then, if they’re pretty interested, we ask them how they conceive of strong AI. What incremental test would something have to pass in order to bea stepping stone towards AI? They come up with a few tests. And then we compare their standards to our roadmap and what we’ve already completed. From there, it becomes very clear that Vicarious is where you should be if you’re serious about building intelligent machines.

  Question from the audience: Even if you succeed, what happens after you develop AI? What’s your protection from competition?

Scott Brown: Part of it is about about process. What enabled the Wright brothers to build the airplane wasn’t some secret formula that they come up with all of a sudden. It was rigorous adherence to doing carefully controlled experiments. They started small and built a kite. They figured out kite mechanics. Then they moved onto engineless gliders. And once they understood control mechanisms, they moved on. At the end of the process, they had a thing that flies. So the key is understanding why each piece is necessary at each stage, and then ultimately, how they fit together. Since the quality comes from process behind the outcome, the outcome will be hard to duplicate. Copying the Wright brothers’ kite or our vision system doesn’t tell you what experiments to run next to turn it into an airplane or thinking computer. 

Peter Thiel: Let’s pose the secrecy questions. Are there other people who are working on this too? If so, how many, and if not, how do you know? 

Eric Jonas: The community and class of algorithms we’re using is fairly well defined, so we think we have a good sense of the competitive and technological landscape. There are probably something like 200—so, to be conservative, let’s say 2000—people out there with the skills and enthusiasm to be able to execute what we’re going after. But are they all tackling the exact same problems we are, and in the same way? That seems really unlikely. 

Certainly there is some value to the first mover advantage and defensible IP in AI contexts. But, looking ahead 20 years from now, there is no a priori reason to think that other countries around world will respect U.S. IP law as they develop and catch up. Once you know something is possible—once someone makes great headway in AI—the search space contracts dramatically. Competition is going to be a fact of life. The process angle that Scott mentioned is good. The thesis is that you can stay ahead if you build the best systems and understand them better than anyone else. 

Peter Thiel:  Let’s talk more about avoiding competition. It’s probably a bad idea to open a pizza restaurant in Palo Alto, even if you’re the first one. Others will come and it will be too competitive. So what’s the strategy?

Scott Brown: Network effects could offer a serious advantage. Say you develop great image recognition software. If you’re the first and the best, you can become the AWS of image recognition. You create an entrenching feedback loop; everyone will be on your system, and that system will improve because everyone’s on it. 

Eric Jonas: And while AWS certainly has competitors, they’re mostly noise. AWS has been able to out-innovate them at every step. It’s an escape velocity argument, where a sustainable lead builds on itself. We’re playing the same game with data and algorithms. 

Scott Brown: And you keep improving while other people copy you. Suppose you build a good vision system. By the time other people copy your V1, you’ve been applying your algorithms to hearing and language systems. And not only do you have more data than they have, but you’ve incorporated new things into an improved V1.

Peter Thiel:  Shifting gears to the key existential question in AI: how dangerous is this technology?

Eric Jonas: I spend a lot less time worrying about dangers of the underlying tech and more about when we’re going to be cash flow positive. Which is why I plan on naming my kid John Connor Jonas...

More seriously, we do know that computational complexity bounds what AI can do. It’s an interesting question. Suppose we could, in a Robin Hansonian sense, emulate a human in a box. What unique threat does that pose? That intelligence wouldn’t care about human welfare, so it’s potentially malevolent. But there might be serious limits to that. Being Bayesian is in some sense the right way to reason in uncertainty. To the extent that I'm worried about this, I’m worried about it for the next generation, and not so much for us right now.

Scott Brown: We think of intelligence as being orthogonal to moral intuition. An AI might be able to make accurate predictions but not judge whether things are good or bad. It could just be an oracle that can reason about facts. In that case, it’s the same as every technology ever; it’s an inherently neutral tool that is as good or as bad as the person using it. We think about ethics a lot, but not in a way the popular machine ethicists tend to write about it. People often seem to conflate having intelligence with having volition. Intelligence without volition is just information. 

Peter Thiel:  So you’re both thinking it will all fundamentally work out.

Scott Brown: Yes, but not in a wishful thinking way. We need to treat our work with the reverence you’d give to building bombs or super-viruses. At the same time, I don’t think hard takeoff scenarios like Skynet are likely. We’ll start with big gains in a few areas, society will adjust, and the process will repeat.

Eric Jonas: And there is no reason to believe that the AI we build will be able to build great AI. Maybe that will be true. But it’s not necessarily true, in an a priori sense. Ultimately, these are interesting questions. But the people who spend too much time on them may well not be the people who end up actually building AI.

Bob McGrew: We view the dangers of technology a little differently at Palantir, since we’re doing intelligence augmentation around sensitive data, not trying to build strong AI. Certainly computers can be dangerous even if they’re not full-blown artificially intelligent. So we work with civil liberty advocates and privacy lawyers to help us build in safeguards. It’s very important to find the right balances.

Question from the audience: Do we actually know enough about the brain to emulate it?

Eric Jonas: We understand surprisingly little about the brain. We know about how people solve problems. Humans are very good at intuiting patterns from small amount of data. Sometimes the process seems irrational, but it may actually be quite rational. But we don’t know much about the nuts and bolts of neural systems. We know that various functions are happening, just not how they work. So people take different approaches. We take a different approach, but maybe what we know is indeed enough to pursue an emulation strategy. That’s one coin to flip.

Scott Brown: Like I said earlier, we think emulation is the wrong approach. The Wright brothers didn’t need detailed models of bird physiology to build the airplane. Instead, we ask: what statistical irregularities would evolution have taken advantage of in designing the brain? If you look at me, you’ll notice that the pixels that make up my body are not moving at random over your visual field. They tend to stay together over time. There’s also a hierarchy, where when I move my face, my eyes and nose move with it. Seeing this spatial and temporal hierarchy to sensory data provides a good hint about what computations we should expect the brain to be doing. And lo and behold, when you look at the brain, you see a spatial and temporal hierarchy that mirrors the data of the world. Putting these ideas together in a rigorous mathematical way and testing how it applies to real-world data is how we’re trying to build AI. So the neurophysiology is very helpful, but in a general sense.

Question from the audience: How much of a good vision system will actually translate over to language, hearing, etc.? If it were so easy to solve one vertical and just apply it to others, wouldn’t it have been done by now? Is there reason to think there’s low overhead in other verticals?

Scott Brown: It depends on whether you think there’s a common cortical circuit. There is good experimental support for it being a single circuit, whether incoming data is auditory or visual. One recent experiment involved rewiring ferrets’ brains to basically connect their optic nerves with the auditory processing regions instead of visual regions. The ferrets were able to see normally. There are a lot of experiments demonstrating related findings, which lends support to the notion of a common algorithm that we call “intelligence.” Certainly there are adjustments to be made for specific sensory types, but we think these will be tweaks to that master algorithm, and not some fundamentally different mechanism.

Eric Jonas: My co founder Beau was in that ferret lab at MIT. There does appear to be enough homogeneity across cortical areas and underlying patterns in time series data. We understand the world not because we have perfect algorithms, but also because tremendous exposure to data helps. The overarching goal—for all of us, probably—is to learn all the prior knowledge about the world in order to use it.  It’s reasonable to think that some things will map over to other verticals. The products are different; obviously building a camera doesn’t help advance speech therapy. But there may be lots over overlap in the underlying approach.

Peter Thiel:  Is there a fear that you are developing technology that is looking for a problem to solve? The concern would be that AI sounds like a science project that may not have applications at this point. 

Eric Jonas: We think there are so many opportunities and applications for understanding data better. Finding the right balance between building core technology and focusing on products is always a problem that founding teams have to solve. We do of course need to keep an eye on the business requirement of identifying particular verticals and building products for particular applications. The key is to get in sync with the board and investors about the long-run vision and various goals along the way.

Scott Brown: We started Vicarious because we wanted to solve AI. We thought through the steps someone would need to take to actually build AI. It turns out that many of those steps are quite commercially valuable themselves. Take unrestricted object recognition, for instance. If we can just achieve that milestone, that alone would be tremendously valuable. We could productize that and go from there. So the question becomes whether you can sell the vision and raise the money to build towards the first milestone, instead of asking for a blank check to do vague experiments leading to a binary outcome 15 years down the road.

Bob McGrew: You have to be tenacious. There’s probably no low-hanging fruit anymore. If strong AI is the high- (or maybe even impossible) hanging fruit, Palantir’s intelligence augmentation is medium-hanging fruit. And it took us three years before we had a paying customer. 

Peter Thiel: Here’s a question for Bob and Palantir. The dominant paradigm that people generally default to is either 100% human or 100% computer. People frame them as antagonistic. How do you convince the academic people or Google people who are focused on pushing out the frontier of what computers can do that the human-computer collaborative Palantir paradigm is better?

Bob McGrew: The simple way to do it is to talk about specific problem. Deep Blue beat Kasparov in 1997. Computers can now play better chess than we can. Fine. But what is the best entity that plays chess? It turns out that it’s not a computer. Decent human players paired with computers actually beat humans and computers playing alone. Granted, chess is a weak-AI in that it’s well specified. But if human-computer symbiosis is best in chess, surely it’s applicable in other contexts as well. Data analysis is such a context. So we write programs to help analysts do what computers alone can’t do and what they can’t do without computers.

Eric Jonas: And look at mechanical turk. Crowdsourcing intelligent tasks in narrowly restricted domains—even simple filtering tasks, like “this this is spam, this is not”—shows the increasingly blurring line between computers and people.

Bob McGrew: In this sense, Crowdflower is Palantir’s dark twin; they’re focusing on how to use humans to make computers better.

  Question from the audience: What are the principles that Palantir thinks about when building its software?

Bob McGrew: There is no one big idea. We have several different verticals. In each, we look carefully at what analysts need to do. Instead of trying to replace the analyst, we ask what it is that they aren’t very good at. How could software supplement what they are doing? Typically, that involves building software that processes lots of data, identifies and remembers patterns, etc.

   Question from the audience: How do balance training your systems vs. making them full-featured at the outset?  Babies understand facial expressions really well, but no baby can understand calculus.

Scott Brown: This is exactly the sort of distinction we use to help us decide what knowledge should be encoded in our algorithms and what should be learned. If we can’t justify a particular addition in terms of what could be plausible for real humans, we don’t add it.

Peter Thiel:  When there is a long history of activity that yields only small advances in a field, there’s a sense that things may actually just be much harder than people think. The usual example is the War on Cancer; we’re 40 years closer to winning it, and yet victory is perhaps farther away than ever. People in the ‘80s thought that AI was just around the corner. There seems to be a long history of undelivered expectations. How do we know this isn’t the case with AI? 

Eric Jonas: On one hand, it can be done. There's an easy proof of concept; All it takes to create a human-level general intelligence is a couple of beers and a careless attitude toward birth control. On the other hand, we don't really know for sure whether or when strong AI will be solved. We're making what we think is the best bet.

Peter Thiel:  So this is inherently a statistical argument? It’s like waiting for your luggage at the airport: The probability of your bag showing up goes up with each passing minute. Until, at some point, your luggage still hasn’t shown up, and that probability goes way down.

Eric Jonas: AI is perceived to have a lot of baggage. Pitching AI to VCs is pretty difficult. Those VCs are precisely the people who expected AI to have come much easier than it has. In 1972 a bunch of people at MIT thought they would all just get together and solve AI over a summer. Of course, that didn’t happen. But it's amazing how confident they were that they could do it—and they were hacking on PDP-10 mainframes! Now we know how incredibly complex everything is. So this is why we are tackling smaller domains. Gone are the days where people think they can just gather some friends and build an AI this summer.

Scott Brown: If we applied the baggage argument to airplanes in 1900, we’d say “People have been trying to build flying machines for hundreds of years and it’s never worked.” Even right before it did happen, many of the smartest people in the field were saying that heavier than air flying machines were physically impossible.

Eric Jonas: Unlike things like speed of light travel or radical life extension, we at least have proofs of possibility.

  Question from the audience: Do you focus more on the big picture goal or on targeted milestones? 

Eric Jonas: It’s always got to be both. It’s “we are building this incredible technology” and then “here’s what it enables.” Milestones are key. Ask what you know that no one else does, and make a plan to get there. As Aaron Levie at Box says, you should always be able to explain why now is the right time to do whatever it is you’re doing. Technology is worthless without good timing and vice versa. 

Scott Brown: Bold claims also require extraordinary proof. If you’re pitching a time machine, you’d need to be able to show incremental progress before anyone would believe you. Maybe your investor demo is sending a shoe back in time. That’d be great. You can show that prototype, and explain to investors what will be required to make the machine work on more valuable problems.

It’s worth noting that, if you’re pitching a revolutionary technology as opposed to an incremental one, it is much better to find VCs who can think through the tech themselves. When Trilogy was trying to raise their first round, the VCs had professors evaluate their approach to the configurator problem. Trilogy’s strategy was too different from the status quo, and the professors told the VCs that it would never work. That was an expensive mistake for those VCs. When there’s contrarian knowledge involved, you want investors who have the ability to think through these things on their own. 

Peter Thiel:  The longest-lasting Silicon Valley startup that failed was probably Xanadu, who tried from 1963 to 1992 to connect all the computers in the world. It ran out of money and died. And then Netscape came the very next year and ushered in the Internet. 

And then there’s the probably apocryphal story about Columbus on the voyage to the New World. Everybody thought that the world was much smaller than it actually was and that they were going to China. When they were sailing for what seemed like too long without hitting China, the crew wanted to turn back. Columbus convinced them to postpone mutiny for 3 more days, and then they finally landed on the new continent.

Eric Jonas: Which pretty much makes North America the biggest pivot ever.

To some extent, the U.S. has fallen a bit behind in the effort. Life expectancy here is several years below the global max. There are all sorts of idiosyncratic explanations for this; Americans eat bad food, are too inactive, etc. But a little U.S. lag notwithstanding, there has been a relentless trend upwards.

Another way to think about it is this: every day you survive, you add 5 of 6 hours to your life. That is a startling realization. The question is what happens next. Is the straight line going to continue? Slow down? Accelerate? Before 1840, life expectancy was pretty flat for thousands of years. Only recently has it really picked up. Whether this is a short burst that will stagnate or just the beginning of a fierce acceleration remains to be seen.

II. Luck, Life, and Death

A. Death as Bad Luck

In a sense, longevity is the opposite of bad luck. At the broadest level, you get into trouble when something unlucky happens to you. Think of everything that can go wrong. Maybe a piece of your DNA mutates and starts a cancer. Maybe you get run over by a car. Or maybe you get hit by an asteroid. There are many different unlucky things that could happen. So the question of longevity can be rephrased as the question of whether and to what extent luck can be overcome.

From the 17th to the mid-19th centuries, the prevailing view was that we could overcome all these accidents. Francis Bacon’s New Atlantis was the classic vision of an accident-free utopia. It was a new Atlantis because, unlike the old one that the Gods destroyed, new Atlanteans had complete mastery over nature.

This view has been receding since about 1850. Luck and indeterminacy have become increasingly dominant as frameworks for thinking about the future. This shift was probably driven by the emergence of actuarial science and life insurance. When people started to map out the data, they realized that life and death could be reduced to probability functions. A 30-year-old has a 1 in 1000 chance of dying in given year. But at age 100 that chance is 50%.

If we run with this math for a bit, living forever becomes just a matter of solving a simple equation:

Unchecked probabilistic thinking can be dangerous. It defeats one’s ability to shape the future. No County For Old Men captures this well; eventually, your luck runs out and you get shot in a deli in Texas. If everything is just a probability distribution, you have to resign to it. But that ignores your ability to think and avoid playing games that are too dependent on luck.

Random historical footnote: 1700, the claim that people could live forever seemed stronger than it would today, simply because there were people running around claiming to be 150 years old. Since record-keeping wasn’t always great back then, good salespeople could persuade others that they were, in fact, radically old. Today, of course, it’s easy to identify these longevity salesmen’s motives. If you’re 70 years old in early 18th century London, you’re perceived as kind of wretched and you get no special treatment. But if you’re 150 years old, that’s really something special. You might even get a pension from the King. 

B. Shift to Determinacy?

Can we move biology away from the realm of the statistical/probabilistic and toward being something that is determinate and solvable? 

It depends.

You can think of death as an accident. There are different kinds of accidents. You can lay these out on a spectrum, from microscopic accidents (genetic mutations) to macroscopic accidents (car crashes) to cosmic accidents (asteroid strikes). To solve the longevity problem completely, you have to get rid of all of these kinds of accidents. But there’s a sense in which certain macro and cosmic accidents are and will continue to be pretty probabilistic things. There is good reason to take those on later; if we can just get to the microscopic solution, the best estimates have people living to between 600 and 1,000 years.

III. CS and Biology

A. Difficulty of the Problem

Like death itself, modern drug discovery is probably too much a matter of luck. Scientists start with something like 10,000 different compounds. After an extensive screening process, those 10,000 are reduced to maybe 5 that might make it to Phase 3 testing. Maybe 1 makes it through testing and is approved by the FDA. It is an extremely long and fairly random process. This is why starting a biotech company is usually a brutal undertaking. Most last 10 to 15 years. There’s little to no control along the way. What looks promising may not work. There’s no iteration or sense of progress. There is just a binary outcome at end of a largely stochastic process. You can work hard for 10 years and still not know if you’ve just wasted your time.

In Internet businesses, the basic rule is that the company succeeds if every round of financing is an up round. In biotech, it’s very hard to do that. Investors get tired. Things don’t work. Some biotech investors are so candid as to state that they don’t really care about valuations, since everything will get wiped out in their favor once a company has the inevitable down round. Why negotiate valuation if luck dominates everything?

To be fair, we must acknowledge that all the luck-driven, stats-driven processes that have dominated people’s thinking have worked pretty well over the last few decades. But that doesn’t necessarily mean that indeterminacy is sound practice. Its costs may be rising quickly. Perhaps we’ve found everything that is easy to find. If so, it will be hard to improve armed with nothing but further random processes. This is reflected in escalating development costs. It cost $100 million to develop a new drug in 1975. Today it costs $1.3 billion. Probably all life sciences investment funds have lost money. Biotech investment has been roughly as bad a cleantech.

B. The Future of Biotech

Drug discovery is fundamentally a search problem. The search space is extremely big. There are lots of possible compounds. An important question is thus whether we can use computer technology to reduce scope of luck. Can CS make biotech more determinative? At the most basic level, biological processes can be thought of as involving some quantum of luck in the face of irreversible degradation. Traditional therapeutics largely mirrors those processes. But computational processes are reversible. You can dive in and re-program things as necessary. So one big question is the extent to which biological problems can be reduced to computer problems?

The cost of DNA sequencing is falling rapidly. It cost $500 million to sequence a genome in 2000. Now that’s down to something like $5,000. Within a year or two, it will probably cost $1,000. The question is whether we can do as much as people have been assuming we can with all the information this will yield.

The Human Genome Project was seen as incredibly revolutionary in late 1990s. But it hasn’t quite lived up to the hype. Perhaps it was all too early or too costly. But the second cut may be that it’s because the main problem is not a sequencing problem at all. The biggest problem may be that we just don’t know what to do with the data. Exactly how much of biology is computational is still an open question.

IV. Examples

We’ll highlight and then have a discussion with people from 3 companies who are doing very interesting things in biotech: Stem CentRx, Counsyl, and Emerald Therapeutics.

Of these 3, Stem CentRx is the closest to traditional biotech. But there is still a heavy computational piece to it. The basic goal is to cure all cancer. Their claim is that cancers have stem cells that are very different from core cancer cells. This stem-cell subset drives cancer/tumor growth. So they aim to target the stem cells and thereby knock out the cancer.

Backing up a step, the problem is that chemotherapy can be really ineffective at threating cancer. It is very hard to get chemo dosages exactly right. Too low a dosage is ineffective at stopping the cancer. Too high a dosage kills the patient along with the cancer. So if you can identify the subset of cancerous cells that are driving the growth and target those cells precisely, chemotherapy would be much less destructive and considerably more effective. So far, Stem CentRx’s studies on mice have been very promising. We should find out whether the approach works for human cancers over the next year or two.

Counsyl is a bioinformatics company whose goal is to become the default for pregnancy genetic screening. They have developed single simple test for the 100 or so genes that can be screened for inherited traits. They focus just on the Mendelian diseases, since beyond that it is currently very hard to know how more complicated genetic combinations work. So Counsyl has identified a tractable and well-defined subset of the problem. Today, Counsyl is involved in screening about 2% of all births in the U.S., and expects that figure to rise quite dramatically in coming years.

Emerald Therapeutics is the most computational of these companies. The basic goal is to cure all viral infections by reprogramming cells, i.e. turning cells into code-based machines. The idea is to build a molecular machine that tags cells that contain viruses, and then to release a sequence that causes those cells to self-destruct. Emerald is in stealth mode, so we can’t say too much. But the high degree of paranoia for companies doing programmable anti-viral therapies is understandable. These are big secrets that play out over long time horizons, not web apps that have a 6-week window to take over the world.

So with that, we’ll have a discussion with Brian Slingerland of Stem CentRx, Balaji Srinivasan of Counsyl, and Brian Frezza of Emerald Therapeutics.

V. Perspectives

Peter Thiel: Marc Andreessen visited this class a few weeks ago. His claim about the Internet in the late ‘90s was that many of the ideas were right, but were just too early. Even if one agrees that next phase in biotech is about to start—things are going to get much more computational—how do you know that now is the right time? How do you know you’re not paddling too early?

Balaji Srinivasan: The sequencing of the genome is like the first packets being sent over ARPANET. It’s a proof of concept. This technology is happening, but it isn’t yet compelling. So there is a huge market if one can make something compelling enough for people to actually go and get a genome sequenced. It’s like e-mail or word processing. Initially these things were uncomfortable. But when they become demonstrably useful, people leave their comfort zones and adopt them. Pregnancy testing is a major on ramp. People find it important to make sure their children are as healthy as possible. And then there is likely to be tons of positive things that can be done with the data beyond that.

Peter Thiel: So the question is how you can overcome pervasive fear of getting genome sequencing? And the answer is: “Do it for the kids?”

Balaji Srinivasan: Yes. No one spends $1000 to get computer so they can use Twitter. But once you have computer, there is zero marginal cost to use Twitter. So solving the install problem is the first step. Empirically, we’re starting to see very strong adoption. So we are confident that we can solve the install problem. 

Peter Thiel: Tackling the cancer problem is exciting but also worrisome at the same time. It’s an old problem. Nixon said in 1970 that we’d win the War on Cancer by ’76. People have been working on it for 40 years. So while we’re 40 years closer to a solution, it also seems farther away than ever. Doesn’t the fact that it’s taken so long mean that it’s an incredibly hard problem that won’t be solved soon?

Brian Slingerland: People have largely followed the same path over the past 40 years. The usual approach to cancer is to carpet bomb it with chemotherapy or the like. The approaches that have been attempted are remarkably similar. So we decided to take really different path. 40 years of failures have taught us something important. The endpoint that everyone focuses on is therapeutic efficacy measured by tumor shrinkage. But this isn’t the best metric; tumors can shrink and then come back. Focusing on shrinkage may lead to attacking the wrong cells. Embracing bioinformatics helps us illuminate better approaches. So we disagree that the problem won’t be solved soon; we strongly believe that we have a very good chance of doing just that.

Brian Frezza: Half of the timing question is taken care of for us; we’re certainly not too late, since viruses still exist. So are we too early? I don’t think so. People’s perspective on healthcare development is quite different from reality. Industry players tend to be very paranoid and secretive until they have a product to release. People discount that quite a bit, since they just pay attention to what is brought to market and when. What most people see at any given point was started decades before they even thought about it.

Biotech got quite a burst in late 70s early 80s, with new recombinant DNA and molecular biology techniques. Genentech led the way from the late 70s to the early 80s. Nine of the 10 biggest American biotech companies were founded during this really short time. Their technology came out some 7-8 years later. And that was the window; not very many integrated biotech companies have emerged since then. There was a certain amount of stuff to find. People found it. And before Genentech, the paradigm was pharma, not biotech. That window (becoming an integrated pharmaceutical company) had been closed for about 30 years before Genentech.

So the bet is that while the traditional biotech window may be closed, the comp bio window is just opening. Whoever gets in during that window gets installed. There are enormous monopoly barriers to getting to market. Here, first mover advantage often becomes last mover advantage. Imagine if IE or Chrome had to go through clinical trials just to get to market. It would be much harder to get in the game. So whoever manages to develop great technology and get it out first is in good shape. 

Peter Thiel: Talk about your corporate strategy. Even if your technology works, how do you distribute it?

Balaji Srinivasan: If you think of drugs, biotech, and now genomics as qualitatively different entities, you’ll see that genomics companies can do things quite differently. Genomics is much more computational than pharma or traditional biotech. With molecular diagnostics—but unlike traditional therapeutics—once you’ve assayed a sample, you’re on info superhighway. Internet rules apply. You can go from conception to product and sales within 15-18 months. It’s not quite as fast as Internet businesses. But it’s considerably faster than the 7-8 years it takes in biotech. In the early ‘90s there was an opening for web 1.0. In the late ‘90s there was the web 2.0 window. Now it’s genomics. We think that bio should just be sensors and gathering data. Everything else should be done at the command line.

Brian Slingerland: Stem CentRx has more of a traditional biotech process. We spent 3 years on the proof of concept phase. Now that we’ve finished all the efficacy studies in cancers in mice, we are in full-blown drug development mode. This process can be accelerated by adopting best practices from tech culture. 

Brian Frezza: We are creating a platform, not an isolated product. We create infrastructure for all sorts of future antiviral technologies. So being able to handle the science in a routine and scalable way is key. The culture is an important part too. Even though we have PhD organic chemists and molecular biologists, we shoot for a hardcore tech startup culture. We automate processes in the lab using advanced robotics. We use git to track our lab notebooks. We write a ton of software. We are the first movers in our space, and we’re trying to move very quickly, but we're also building a platform that's designed to scale exponentially.

Peter Thiel: How do you know that there isn’t someone else secretly pursuing the same strategy? And if you’re confident as to what other people are or aren’t doing, how do you know that they don’t know about you and that your secrecy is working?

Balaji Srinivasan: It’s like the Rumsfeld quip: there are known unknowns. Ultimately we think most people miss the key secrets in health industry because they are so caught up in the status quo that they actually can’t think their way to good solutions. Contrast the healthcare industry with the fitness industry. Ultimately, your fitness is your responsibility. You can join good gyms or get personal trainers. All that’s great. But the buck stops with you—you have to take the initiative. But consider how that initiative plays out in healthcare. If you come to a doctor’s appointment wanting to talk about something you’ve researched, doctors get pissed. You are either undermining their authority or you’re an idiot. But that’s odd; you are with your body for a lifetime, whereas the doctor is with you for 20 minutes each year. The one area of medicine that works—fitness—operates orthogonally to the rest of medicine in practice.

When these systems get build up, it’s very hard to clear away the overhang. People have thought themselves out of thinking about non status quo solutions. Stuff that actually works is perceived as crazy.

Brian Frezza: One bad vestige of the biotech boom is what happened to patents. In pharma, traditionally compounds got patents. But in biotech, general techniques became patentable. Genentech, for example, managed to patent recombinant antibodies as a general concept . So biotech is littered with really broad patents. Some biotech companies literally generate millions in revenue just from patent licensing; they produce no drugs at all. So it's best not to generate a large amount of public interest in new techniques you're developing if you don't want to encourage stray IP to accumulate.

Ultimately, you can’t prove a negative. It is distinctly possible that there is a Ruby Therapeutics out there that is doing the same thing we are. But we very much doubt it, given how unique what we're doing is. Even knowing that they may be out there, it still makes sense for both companies to stay quiet until you’re ready for revenue.

Brian Slingerland: There’s really no rush to spill the secret plans. This space is very much unlike fast-moving consumer Internet startups. Here, if you have something unique, you should nurse it. One good rule of thumb is to issue no press releases until you push a drug. That said, it’s a balancing act. Since our approach has been proven out and we’ll be moving to human trials, we are becoming a more public-facing company. No one wants to take a drug made by a stealth company with no info on its website. You just want to make sure that you don’t divulge too much too early. 

Of course, people should assume there are 10 companies coming after them. It’s always safe to assume that you have to work better and faster to come out ahead.

Peter Thiel: If you thought that a Ruby Therapeutics—or 10 different versions of them—was actually out there, wouldn’t it make sense to be more open and collaborate? And how do you recruit people if you’re so secretive? 

Balaji Srinivasan: In ecology, when you want to know how many species are in a jungle, you take a sample and project out. Sizing up the competition is a similar task. If you take a hard look at your network—search through the Silicon Valley part of the forest—and see no capital being deployed and no one working on the same problems, you can be reasonably confident that you’re alone. People would really have to come out of nowhere.

Personal referrals are very important for recruiting. We try and get each engineer to refer 2 people. 2^n scales very well. You get great people, but also get to stay under the radar.

Peter Thiel: That recruiting strategy has worked well in every company that I’ve been involved with. You have to keep a clear head about it. If you ask MBAs to refer talented people who are good to work with, you’ll get far too many recruits. But if you put the same question to engineers—and maybe it’s their friends who you really want to recruit—you may get a shocking silence because they are too shy. So you have to find a way to get them comfortable with referring people.

Brian Frezza: One strategy that works for that is to sit down with your engineers and go through their Facebook friends with them, one by one, and ask them who is good and who they’d like to work with. 

Peter Thiel: The bias for Silicon Valley entrepreneurs is to go work on a web or mobile app. Why should more people think about doing biotech/computational stuff instead?

Balaji Srinivasan: The thing to remember is that the next big thing won’t look like the last big thing. Search didn’t look like the desktop. Social didn’t look like search.

The human genome will never become obsolete. Mobile/local/social? It’s hard to say. Mobile seems to have a lot of growth ahead of it. So that’s at least a reasonable bet. Local? There’s not really a defensible advantage anymore. And social has been colonized. Flags have been planted.

Ultimately you simply have to care about what you’re doing. Another dating app really doesn’t matter. It’s hard to bleed/sweat/cry for. Meaningfulness is a big part of why people should think differently. And we think genomics is really meaningful.

Brian Frezza: Elon Musk is a master recruiter. The narrative is stark and simple: “We don’t pay as well as Google. But this is the most exiting project you can work on in your life.” You want to attract the people who find that narrative attractive. People can always try to find a lottery ticket of a startup. But the satisfaction of creating a tech revolution is much bigger than what comes from just chasing dollars.

Brian Slingerland: One reason that CS people may be ignoring biotech is that they think that they’ll be relegated to supporting roles. But that’s far from true at many biotech companies. CS people are at the very core of what we do at Stem CentRx. So if CS people have an interest in curing cancer or things like that, it’s certainly something they should think about. Have I mentioned that we’re hiring?

Peter Thiel: We usually say that advertising works best if it is hidden. But sometimes it actually works if it’s completely transparent. [laughter]

Being opaque can be so tiring. The standard wisdom in the VC world is: “If you want money, ask for advice. If you want advice, ask for money.” That game is exhausting. Sometimes it can be refreshing to hear someone say, “I really just want money.”

  Question from the audience: [unintelligible]

Peter Thiel: It is very odd that the FDA has a bottleneck on global drug development. There has to be some tipping point beyond which the U.S. no longer gets to dictate what drugs are developed in the entire world. Past that inflection point, the U.S. may have to compete with China on how quickly drugs can be developed. That could be a huge paradigm shift. So while things look pretty bad now, the future may be quite promising. SpaceX was very heavily regulated at first, but persevered and got through it. And the aero regulations have eased up in the last decade. So the sheer unfriendliness of the baseline could be a great opportunity.

Question from the audience: When you disclose your secret to prospective hires, do they try to use that knowledge as leverage to hold you up and negotiate more?

Brian Frezza: It hasn’t been a problem at all. VCs don’t sign NDAs, but job candidates will. And it would take years for anyone we talk with to replicate our technology on their own. 

Question from the audience: What role does HIPPA play in tech innovation?

Balaji Srinivasan: HIPAA could be seen as tech problem. How can we follow such and such standards, etc.

But it’s also an interesting genomics problem. A person’s genome provides a great deal of information about their relatives. On one hand, this data is private medical data. On the other hand, it’s inherently statistical and requires aggregation to do anything very useful with it. So the trick is to figure out how to do private aggregations. To get value out of your genome, you simply must allow some computation on it. The challenge is catalyzing this very important social shift toward becoming okay with that while preserving strong privacy controls. 

  Question from the audience: Unlike the web, where you can get feedback in minutes, how do you know if you’re on the right track in computational biotech? 

Brian Frezza: We use physical models and actual validation experiments. We don’t just use statistical approaches. But our processes are internal. We don’t go outside and seek external validation. Just like Instagram doesn’t get outside people to come in and appraise its code base. You develop a plan and execute it internally. Sometimes you have a multi-year cycle to get data back. You just work as efficiently as possible to shorten cycle times.

Balaji Srinivasan: Slow iteration is not law of nature. Pharma and biotech usually move very slowly, but both have moved pretty fast at times. From 1920-1923 Insulin moved at the speed of software. Today, platforms like Heroku have greatly reduced iteration times. The question is whether we can do that for biotech. Nowhere is it written in stone that you can’t go from conception to market in 18 months.

Brian Frezza: That depends very much on what you’re doing. Genentech was founded the same year as Apple was, in 1976. Building a platform and building infrastructure take time. There can be lots of overhead. Ancillary things can take longer than a single product lifecycle to accumulate.

  Question from the audience: How does biotech VC compare to regular VC?

Brian Slingerland: We never did the classic venture capital route because VC is broken with respect biotech. Biotech VCs have all lost money. They usually have time horizons that are far too short. VCs that say they want biotech tend to really want products brought to market extremely quickly.

Brian Frezza: “Integrated drug platform” is an ominous phrase for VCs. More biotech VCs are focused on globalization than on real technical innovation. VCs typically found a company around a single compound and then pour a bunch of money into it to push it through the capital-intensive trial process. Most VCs not interested in multi-compound companies doing serious pre-clinical research.

  Question from the audience: How useful are end-stage trials in trying to figure out how to cure cancer? Don’t you get inaccurate or just different genome data from terminal patients?

Brian Slingerland: Not being able to trial on earlier stage people is always a challenge. But our technology it is designed to apply to patients at all stages. All I can say is that our approach is stage agnostic for a variety of technical reasons. But generally speaking yours is a valid concern. That’s why traditional drugs that show initial progress often fizzle out in extended trials.

  Question from the audience: What were some of your early struggles or challenges?

Brian Frezza: The amount of time it took to set up a lab was shockingly large. 100% of our time went into acquiring equipment, negotiating price, dealing with initialization failures, etc. We greatly underestimated the time required to get up and running because we were coming out of existing, well-supplied labs. It basically takes a whole year to get up and running. There’s just a huge difference from the computer/Internet tech world.

Peter Thiel: With Internet businesses, you can be up and running without doing hardly anything. At PayPal, the biggest interface with reality was that, on Max’s orders, people had to assemble their own desks. But Luke Nosek thought even that was too much. So he found a company called Delegate Everything, who dispatched this elderly woman handyperson out to assemble the desk for him so that he could do more work on the computer.

Balaji Srinivasan: Startups are always hard at the start. There are futons and ironing boards in the office. You have to rush to clean up for meetings. But maybe the hardest thing is just to get your foundation right and make sure you plan to build something valuable. You don’t have to do a science fair project at the start. You just have to do your analytical homework and make sure what you’re doing is valid. You have to give yourself the best chance of success as things unfold in the future.

If you're interested in these companies, do check out jobs.counsyl.com and stemcentrx.com/careers.