#frontier technology

This is an internal note I sent to a few investors/friends a few months ago.

Over the past 6+ months there seems to have been a noticeable drought in the creation of new deep tech or just broadly technically ambitious companies (ex-biotech). This is a sentiment that has been echo'd heavily from other investors, and has only accelerated with COVID. This became especially noticeable as startup fundraising has massively increased in adjacent areas in H2 2020. 

While there isn't a data-driven reason as to why this has occurred (or if it has, indeed occurred) my working hypothesis is three-fold.

1) The maturation of platforms and no clear outbreak of new platforms

After a fervor of excitement surrounding areas like VR, ML, and Robotics from 2014-2018, we've largely seen unabashed excitement die down and be traded with hopeful skepticism. VR was too early and is looking increasingly like a platform that will be subsidized and dominated in early innings by incumbents due to heavy R&D costs. 

Machine Learning has shown a commoditization curve that erodes value for much core ML development, while also has been incredibly difficult to implement at scale to build vertical specific companies that accrue value (read our 2019 annual letter for more thoughts on this). ML companies have largely turned into product-centric organizations versus true research engineering orgs, which many investors aren't quite sure how to parse mid to long-term defensibility or economics (see Martin Casado’s wonderful post).

The closest next platforms that people seem to be centering on are AR and Quantum. AR has its own difficulties, again, largely a value accrual in the short-term towards incumbents (Snap, Apple, MSFT) and at the application layer very little differentiation in technology (my view is that most enterprise AR companies are building product orgs with little technical moats and early product UI/UX that will be considered cringey in 2-4 years). 

Quantum has a usability problem in addition to a timeline narrative that "this will take awhile and is incredibly capital intensive" so the only vertical use-cases we've seen really have been pharma and finance. In addition, some argue that ML algorithm development has kept pace in lockstep with quantum in reality.

So TLDR, investors need a narrative to drive them towards a Schelling point surrounding "what's next" but can't find consensus. 

Again, Computational bio and further on the spectrum, biotech seems to be only area, but has debates around moats and is incredibly hard to parse and pattern match for technology investors.

Thus we default to what we know has killer cash flow dynamics in a time where TAM has materially expanded (SaaS/dev tools/infra), or follow narratives that we want to be (and may be!) true, like social being ready for VC investment again.

2) COVID & Founder Profile: “It's time to build" wasn't meant for deep tech founders.

Founding a company becomes sexy as generational startups scale and with a large amount of money in stock options, employees feel the desire to work on a new problem where they can seek a new form of status, both social and financial (early ESOP or founder title). It's no longer cool to be at the $10B+ tech company and there are hundreds of people at these companies starting funds, syndicates, and companies, so you should too.

Some subset of employees on the technical side look at internal tools that were built and democratize them (think data science teams, but there are tons of examples).

Another subset are PMs and software engineers who read Marc's essay, saw a year of remote work ahead of them and thought "great I can move out of SF, live cheaply and stomach going from $250k/year to $100k/year, and be a Founder. 

These people like building software that can quickly be iterated upon and is quite realistic to raise $1-$3M to run at for 24 months. In addition, these people likely feel time pressure because multiple other teams are starting startups similar to their idea. It’s time to build because everyone is building.

Deep/frontier/emerging/whatever we call them tech founders sometimes follow a bit of a different founding story. Many have dedicated 4-10 years of their life becoming an expert in a given industry, or banging their head against the wall in academia, in pursuit of applying a technology to the real world, or bringing some other initially non-sexy technology to production level. The time pressure for these companies and founders could be viewed as less existential, as many companies aren’t speedrunning capital markets for a 12 month seed to series A sprint comprised of MVP -> early pilots -> highly extrapolated MRR/ARR. 

Instead, many deep tech companies are operating with a 2-4 year R&D window, hoping to show a step function in technology reality and value creation by month 18 to raise a Series A, where they then start to either begin converting pilots, or testing hypotheses on commercialization over the next 2-4 years. 

Candidly, there ain’t that many people signing up for that trajectory.

So when you're a deep tech profile founder (not to be overly prescriptive but often something like PhD/academic, eng/product lead at incumbent R&D group, Engineering at prior deep tech co, Engineering at scaling tech company with material infrastructure innovation, etc.) your calculus is a bit different.

If you're thinking about hardware, you might say, you don't want to deal with the dynamics of having possible supply chain headaches (not as much a concern today vs. 6 months ago), you don't want to deal with dynamics of really not being able to do remote work, and you know VCs are probably a bit risk averse on funding hardware at this stage as a whole. In addition, while customers might have a strong desire if you're automation focused, you won't be able to ship product for another 12 months probably so, why not just wait to see how the election/transfer of power/vaccine timeline shakes out to truly understand the economy in 2021, before starting something that could be de-railed from random economic/geopolitical policy.

Calculating the risk factor and why now of your business. The reality is, you must convince a smaller universe of investors every 12-36 months in the future you believe in and hope to pull forward, so why *now* matters more when these futures are less consensus. So just doing simple calculus, all of these things once again could lead to waiting. Bank your $300k+ salary at bigco through 2020 and see where the world is in 2021. You're operating on a true long-term time horizon so there isn't as much deep desire to uproot your life/family just to build asap.

TLDR - Deep tech founders maybe viewed as slightly less beholden to the time pressures of the market as many are solving previously unsolvable problems. I will say, many do underestimate the commoditization curve they are running up against though.

3) Burn Out

This point is not talked about enough publicly but is a consistent sentiment I’ve noticed over the past few years.

The first wave of "deep tech as a category" really hasn't created many successful businesses yet (we can debate the recent run of SPACs in deep tech on it creating successful outcomes). As you talk to more veterans in this space, many are pretty disenchanted by their past 7+ years of work. 

Their friends at Uber, AirBnB, and Stripe are multi-millionaires (some on paper), same with their friends at FAMGA, and they possibly are too but don't really have anything material to point to in the real world. They’ve spent a good chunk of their professional lives to hopefully get swallowed up in an offensive acquisition (Boston Dynamics, Skybox Intelligence, Zoox, Cruise) or early commercialization that is strategic (Kiva, Six River, Blue River), and a bunch more are sitting at these companies that weren’t able to quite deliver on the future they believed in. 

I think many are kind of waiting for a "moment" that feels like everything is once again possible. I don't think those moments are obvious in the present, but alas, it's a real sentiment.

--

I know I’ve painted a fairly jaded point of view on both sides of the market but ultimately I think there are massive tailwinds for all of tech over the next few years and believe COVID has pulled forward some futures we were expecting to take 3+ years.

What is your responsibility as investors when investing in these potentially scary or damaging technologies?

I've gotten a form of this question asked on the last two panels I've been on over the last two weeks (LDV Summit: Investing in Visual Technologies, Betaworks Render: Investing in Synthetic Media). It's a fair and topical question that in the last case was then followed by "Do you think Facebook is evil?” Clearly we're working out our feelings on this in real-time, and just now really starting to care about our answers.

This question makes some investors nervous because it puts a microphone to your mouth and asks you to take a stance on capitalistic desires, broader life principles, and really tests the limits of just how "mission driven" you are, despite your job really just being to take $1 and turn it into $4+ over 10 years. In an industry that loves nothing more than signaling our undying passion for mission, to the point where HBO's Silicon Valley repeatedly mocked us, there is some brand risk here.

Despite being oddly caught off guard and having no real politically correct answer to this question, I've tried to be transparent in showing both my strong capitalistic and somewhat philanthropic side. My answer at these events could be paraphrased as something like this:

I think this question bifurcates people. Either there are those that claim to take a very hard moral stance on evaluating companies entirely for good and then claim to deeply care about their responsibility in the trickle down effects, OR there are those that believe trickled down inevitabilities are exactly that, inevitable, and thus we must solve hard problems first and be mindful of long-term issues, but not destructive of innovation because of them.

I fall closer to the latter. I moreso believe that these technologies are indeed inevitable and their shockwaves will be felt regardless of if I back them or not, on some time horizon that is within a decade or so. I don't feel comfortable investing in some companies that I can see a clear, core, and very direct path to negative behavior (think Juul), especially in the short-term. However I'm not going to pass on a company I feel strongly about because there's a scenario where a decade or two down the round, the technology could be maybe negatively used by some small number of bad actors. My moral obligation does hold towards founders though, in the sense that humans that have previously acted immorally at certain levels, I will not back regardless of their pedigree, network, idea etc.

And then I usually mumble that it's a difficult decision and I don't think anyone has the perfect answer here.

-----------

The day after I was on the Betaworks panel, I read a paper by Nick Bostrom which breaks down what is called The Vulnerable World Hypothesis. The paper talks about possible inventions and how we've largely avoided a catastrophic outcome from an invention that destabilizes civilization because...well we've been lucky. The paper describes this innovation as such:

One way of looking at human creativity is as a process of pulling balls out of a giant urn. The balls represent possible ideas, discoveries, technological inventions. Over the course of history, we have extracted a great many balls—mostly white (beneficial) but also various shades of grey (moderately harmful ones and mixed blessings).

...What we haven’t extracted, so far, is a black ball: a technology that invariably or by default destroys the civilization that invents it. 

The paper then goes on to mention ways in which we could prevent or risk mitigate various types of "black ball" inventions.

Nick uses illustrative examples to hypothesize future black balls and types of vulnerabilities, many of which involve areas I very directly and specifically invest in, such as AI/ML, synthetic biology, and genomics. This got me thinking more about the question I mentioned at the top of this note.

Then I came across this excerpt:

The principle of differential technological development is compatible with plausible forms of technological determinism. For example, even if it were ordained that all technologies that can be developed will be developed, it can still matter when they are developed. The order in which they arrive can make an important difference—ideally, protective technologies should come before the destructive technologies against which they protect; or, if that is not possible, then it is desirable that the gap be minimized so that other countermeasures (or luck) may tide us over until robust protection become available.

This spoke to me with respect to this question on ethics and investing in emerging technologies, because it accepts inevitability, but doesn’t accept the often overly optimistic “humans will figure it out, we always have” stance that many have. It instead tells us to keep in mind the sequencing of these innovations so countermeasures may tide us over and risk-mitigate.

A story related to this:

I quickly became enamored by GANs from Ian Goodfellow’s first paper in 2014, then went on to study what I called digital asset manipulation, and then researched and wrote about how we must build technologies to detect these manipulations in 2016.

I’ve now spent ~3 years talking to multiple researchers, founders, and engineers working on the detection side. It was then that I reached the view and I wrote about how society must shift for any real defense to take place (Moving from a default trust to default skeptic society), and that detection technology won’t help drastically curtail many of the risks of these technologies (Truth Algorithms aren’t helpful for fake news) in the near to mid-term.

To be more direct, all of this technology is great, but my belief is that humans have only enacted widespread change when forced with terrible historical outlier events. Yes, we “figure it out”, but we do so after suffering.

Bostrom discusses this with relation to an increase in the importance of access control in bioscience that came due to the 2001 “Amerithrax” incident:

Some black-ball technologies, when they first emerge from the urn, might be difficult to use and require specialized equipment. There could be a period of several years before such a technology has been perfected to the point where an average individual could master it. 

During this early period, the set of empowered actors could be quite limited; for example, it might consist exclusively of individuals with bioscience expertise working in a particular type of lab. Closer screening of applicants to positions in such labs could then make a meaningful dent in the risk that a destructive individual gains access to the biotech black ball within the first few years of its emergence. 

-----------

As it stands today, we have research repositories and active theses in areas including generative media/creative AI, the future of family planning (including “designer children”), Robots/AI and how they impact social dynamics, and many other areas that could have scary downstream effects. I have and will continue to invest in companies in all of these categories today.

I write all of this to say that I believe the overarching view on our history and The Vulnerable World Hypothesis is one that should be shared more with blind human-optimists that believe we will always figure it out, because in our short time as a civilization we have. But past performance does not equal future results.

Some of the frameworks proposed for mitigation I believe are idealistic, but will help inform how I evaluate companies and their mid-term societal threats moving forward, as well as how I answer the initial question surrounding moral obligation as an investor. Specifically, the risk mitigating systems we put in place at both a company and societal level is one that I believe should be openly discussed for areas that are less nascent and (we believe) are slightly easier to project forward. Bostrom calls this “partial protection” in his conclusion.

Ultimately, I imagine I’ll remain in my previously described grouping of capitalistic, future-optimist, driven by a desire to bring the future to the present at a faster rate, but better equipped to analyze the “strings or elastic bands between some of the balls, so that when we pull on one of them we drag along several others to which it is linked.”

And I also imagine that many will quietly believe that people like me are why we’re in this social media debacle, and many other issues we’ll continue to see unravel in the future.

I don’t really have a great ending so I’ll end with a separate quote from a Fred Wilson post in 2012 that I’ve always loved that speaks to the need to continually evolve.

How advanced driver assistance systems will become mandatory and could be the immediate opportunity in mobility.

Seat belts (paired with airbags) fundamentally changed the safety dynamic of driving for consumers to a point where they became required in all vehicles. I believe advanced driver assistance systems (ADAS) will have a similar trajectory.

Advanced driver assistance systems (ADAS) are systems developed to automate/adapt/enhance vehicle systems for safety and better driving. Safety features are designed to avoid collisions and accidents by offering technologies that alert the driver to potential problems, or to avoid collisions by implementing safeguards and taking over control of the vehicle. Adaptive features may automate lighting, provide adaptive cruise control, automate braking, incorporate GPS/ traffic warnings, connect to smartphones, alert driver to other cars or dangers, keep the driver in the correct lane, or show what is in blind spots.

Basically ADAS uses technology to make you a better/more aware driver.

First, a brief history on seat belts

The first variation of the seat belt was invented in the mid-19th century by English engineer George Cayley. Many years later, they made their way into cars when Nash and Ford offered seat belts in 1949 and 1955, and Volvo integrated the first three point seat belt in the late 1950s.

The next decade, in 1966, Congress passed the National Traffic and Motor Vehicle Safety Act, which formally established Federal Motor Vehicle Safety Standards which made seat belts mandatory in all US vehicles.

Today it is illegal in every state except New Hampshire to drive without a seat belt.

Safety Statistics & ADAS

Seat belts reduce the risk of death of front-seat passengers by 45% and the risk of serious injury by 50%.

According to a report by BCG commissioned by MEMA, ADAS could prevent 28% of all crashes in the US, approximately 9,900 deaths, and saving $251B to society if new-car buyers invested in the most common currently available ADAS features.

It’s statistics like these that make it abundantly clear that ADAS will go the way of seat belts and become mandatory in all new vehicles in the near future.

When does this happen?

It’s already starting to happen. By 2018 all cars will be required to have rear view cameras. With increasingly autonomous features creeping their way into vehicles today (whether it’s the more simple lane-keeping and blind spot warning tech, or the more advanced systems like Tesla Autopilot), the value of these technologies will become abundantly clear to both regulators and consumers. More OEMS are going to need these technologies to continue to sell cars, and if ownership models shift from full to partial or distributed ownership (a la Uber) then the importance of safety and tech could outweigh status and looks.

This provides an opportunity for both aftermarket companies (if they are willing to deal with regulatory hurdles) as well as directly integrated ADAS startups to be acquired by a wide-range of OEMs or Tier 1/2 suppliers. (I’ve written about this before)

When talking about consumer Virtual Reality adoption, many voice concerns around activity, over-immersion, and fatigue. Most consumers just want to “VR and Chill” however some of the appeal that has created insane usage statistics in things like mobile gaming or social networks is “passive activity.”

Today when consumers watch TV or other content, they are often multi-tasking with their mobile device, checking Twitter, Instagram, Snapchat, email, or a slew of other apps. This is the definition of passive activity.

The elongated satisfaction curve of serial television no longer stimulates us enough as we search for what many believe are fast dopamine rushes which equal satisfaction.

First, let’s clarify what dopamine actually is:

Dopamine causes seeking behavior. Dopamine causes you to want, desire, seek out, and search. It increases your general level of arousal and your goal-directed behavior. From an evolutionary stand-point this is critical. The dopamine seeking system keeps you motivated to move through your world, learn, and survive. It’s not just about physical needs such as food, or sex, but also about abstract concepts. Dopamine makes you curious about ideas and fuels your searching for information. Research shows that it is the opioid system (separate from dopamine) that makes us feel pleasure.

“wanting” (dopamine) and the “liking” (opioid) are complementary. The wanting system propels you to action and the liking system makes you feel satisfied and therefore pause your seeking.

So dopamine = seeking, opioid = resulting pleasure.

What the first wave of addictive mobile applications did was create what people call “dopamine loops”, where seeking and pleasure happens so quickly (think: tweet -> favorite, or text -> response) that we continually seek more and more. And perfectly enough, dopamine is most effective when only a small amount of information is presented, like 140 characters, or a short text message, because it keeps us seeking and doesn’t fully satisfy our information craving.

And that’s the opportunity for VR content.

In some ways, VR has the potential to end the constant seeking because we get more complete experiences. Stimulants will create dopamine rushes, and then VR will satisfy the opioid system. A 360 photo will fill in more gaps than a normal photo, a 360 video will tell you more of the story than a tweet. It’s the most fulfilling technological platform for passive activity we’ve ever had. 

In 2011, Chris Dixon wrote a post titled Selling pickaxes during a gold rush. He explained the dynamics of pickaxes as follows:

When a major new technology trend emerges – say, the rise of online video or social media – entrepreneurs can try to capitalize on the trend by creating a consumer product (mining for gold), or by creating tools to enable consumer products (selling pickaxes).

The idea of pickaxes has now become a mainstream, buzzy term used by VCs and entrepreneurs all throughout startupland. Like online video or social media in the mid to late 2000′s, we are now seeing similar companies within frontier tech aiming to be the pickaxes of their given vertical. 

With more nascent sectors, infrastructure/platforms are often one of the key factors that enables that given sector to break out (think charging stations for EVs, early Oculus/Vive dominance for VR, and even earlier moves within ATC & Fly/No Fly for Drones). One such area I’ve been spending some time in recently that seems to be overcrowded is undoubtedly satellite launchers, and others agree.

Doug Messler took notice as well, and documented just a few of the more well-known launchers here , while SpaceWorks even included a case study on the rise of smallsat launchers in this year’s report (slide 15).

in the past 4 months we have seen over 15 early-stage startups fundraising for launch services for satellites (most commonly sub-50kg sats). Here are the catalysts for this:

- Many new satellite startups are emerging from the agile development trend made famous by companies like Planet Labs which rely on constellations of satellites in low-earth orbit (LEO). Many of these nano/microsats are incredibly cheap to build but suffer one major flaw...

- Launching is very expensive. Spaceflight projects upwards of $295K for a 3U CubeSat rideshare, and all the way to $1.75M for a 50kg satellite.

- The majority of today’s launches occur as secondary payloads on larger vehicles (like those owned by SpaceX or Orbital ATK). These launches happen irregularly, can suffer massive delays, and the satellite companies have almost no leverage in the event that something doesn’t go according to plan.

- 3000+ nano/microsats are expected to be launched by 2020. A smaller number of smallsats will also be launched in this time.

- As more of these startups/constellations get going, many of the micro/nanosats suffer from poor reliability as well as lack of adjustability once in orbit, so in order to keep a constellation working reliably, companies will have to continually be replacing satellites over time.

All of these points lead us to a massive race by startups to provide launch services hoping to solve the cost, supply, and frequency for the coming space economy.

While the general premise makes sense, differentiation between these launch providers continues to be difficult to sift through, especially when many are in early stages of testing and won’t be launching for at least a few years (with the exception of a few like Rocket Lab). In addition, as Bilal astutely points out in his tweet above, it will be a race to the bottom.

In the ever changing world of global data communications with the help of internet connections and fast-paced software development, IT security is becoming a matter of concern. IT security is the key matter that affects every business.

A combination of smart decisions and intelligent securitystrategy is required for achieving a superior information security for the business firms. Companies must…