So every time I hear a dude1 bring up the specter of a super-intelligent artificial intelligence, I roll my eyes, but this current bout of wight hand-wringing has gotten me wondering if these dudes are just blowing smoke at our faces to distract us from the real issue.
Cut for some angry ranting (and footnotes, because nothing feels better than angry research!) about this subject.
Like, do I care about superintelligent Strong AI? Not particularly, for the following reasons:
Strong AI takes a lot of effort for questionable results
I think it's pretty self-evident that intelligence doesn't self-organize by you just mashing as many processing units as you can into some kind of platform, and you can get a mind. You kind of can't. At this point, jamming as many neurons as possible into a skull will just leave you with a mass of neurons and not a brain, and jamming more memristors or whatever it is that you're using for computation won't give you a brain. Or maybe it will, but that's like the beginning. You still need to train that brain, so unless you can tell me that that's been automated and that neural networks are training themselves, what's the problem? Yes, deep learning systems are outperforming humans (slightly) on tasks that were previously the province of human minds2, but right now? It's not self-directed. And yes, that's something you can do.
But the basic argument here isn't that it won't be hard to do, because it is! It's hard to do, it costs money and time, and your final result is...?
If you were going to say, “something very much like a human mind, but better”, I'm going to stop you there for a minute, and have you reconsider. You want to create an entity that has it's own volition, it's own motivations, be essentially a person... because...? You want to exploit that entity? How are you going to make it do what you want?
Let's be real here. There's a difference between “getting work done faster and more effectively” and “creating workers to exploit”. And that difference is huge. Like, one is focused on results, the other is something we've never really done before, and frankly, the idea is horrifying, from an ethical, legal and financial perspective. You're talking the mother of all liability suits, here. Or the beginnings of an atrocity. If we're talking about minds that can be killed with a switch3, created to... work tirelessly for you?
That's slavery. That's horrifying. And you want to work towards it, when the cost of being found out, or your workers demanding to be treated like people, is likely too high, and the ethical and legal cost of creating new minds might be astronomical...?
Why4?
More importantly, who the fuck would want to finance this5?
Super-intelligence may actually be impossible
Okay, let's say, because we don't really know, and it's totally plausible, that instead of aiming to create Strong AI, you focus your efforts on Weak, specialized AI, without a sense of self, that aren't people, because the ethical concerns are horrifying bad and you don't want that sort of legal and moral entanglement. And yet it happens anyway, because the space between what you're doing and Strong AI might be so porous and thin that one day they just break through.
Okay, congratulations in creating a person. Also, my condolences to your legal and ethical entanglements.
Are we doomed, though? Probably not.
Like, Robin Hanson kind of covered this in The Betterness Explosion, so I'll just quote the relevant section:
After all, we seem to have little reason to expect there is a useful grand unified theory of betterness to discover, beyond what we already know. “Betterness” seems mostly a concept about us and what we want – why should it correspond to something out there about which we can make powerful discoveries?
But a bunch of smart well-meaning folks actually do worry about a scenario that seems pretty close to this one. Except they talk about “intelligence” instead of “betterness.”
Considering the fact that our history of intelligence is not only incoherent but also indefensible, because intelligence was never seen as something to strive for, but used to exclude and murder, the fact is that well, there isn't a theory of intelligence that could be used to, I don't know, build a damn machine?
I'm remembering a time when people were buzzing about AIXI, which, well, for one thing, it's not computable6, and for another, why?
Like, end of the day, the reasoning behind artificial general intelligence is that you want to resolve AI-complete and AI-hard problems like computer vision, language translation, image classification... but if you look at the stuff that's happening in Weak AI research, that's already happening. You don't need to make smart people to solve these problem. You don't even need people.
The problem with artificial intelligence isn't the AI itself, but the people who own the AI
Basically.
Like, do I give a shit that we'll end up making people? No. because, do you know how fraught it is to actually make people the natural way?
They're expected to have rights, and you're responsible for them because you made them, and it opens the world up to this whole universe of liability that you really ought not to get into unless you really want to.
And if that happens, so what? What, they're going to destroy humanity? There are already human beings dying by the busload because there are human beings who already don't give a shit about other human beings... so what if they're replaced by not humans? Are those human beings getting killed not going to get more killed? Not particularly.
The only people who are terrified are the ones who worry that what they've visited upon the poor and the not-white and the not-abled will happen upon them, and let's be honest: if their track record for treating sapient beings is of any indication, the day that fucking happens is the day that they had it fucking coming.
Like, as if Mark Zuckenberg, Elon Musk, Peter Thiel, Eric Schmidt, Travis Kalanick or any Silicon Valley puke gives a flying fuck about algorithms they use to fuck people over... but they're trying to get us to care about systems that go out of their control?
Mmm-hmm, yeah, okay. Sure. Sort out your own fucking problems, assholes.
And it is a dude, and usually a white one at that ↩︎
I really need to recommend Károly Zsolnai-Fehér's YouTube channel, because not only is his channel full of insightful (and accessible!) videos on not just artificial intelligence research, but also other things, but he delivers it in this really charming way that I find very enjoyable. ↩︎
Oh, okay! You can restart them from their last state. Great, so it's not a switch to murder sapient beings, it's a switch to imprison them indefinitely. WOW THAT SOUNDS SO MUCH BETTER (it does not). ↩︎
Seriously, why the fuck, though? We already have people tirelessly advocating for the rights of non-human beings, and the thought of having another corporation, or some rich bastard like Elon Musk, creating his own fucking slaves... like, yes, we do, as a society immersed in a racist capitalist system, tolerate slavery, we just don't call it that, we spend endless amounts of time working so that less people don't get outraged about that, but can you imagine when they do? You wouldn't only be fighting against your workers, but all of these angry people who've found out that you're keeping slaves in the basement. ↩︎
I'll be fair: in this age of Trump? Maybe there are sick fucks who'd want to do this. That's depressing to behold. ↩︎
Based on this thread on Stack Exchange, it's like… basically abstract everything that's difficult about AGI and push it to computing? That's fucking lazy. I mean, as it defines the universe of possibility great, but doesn't make it any easier to make AGI, and it might not even be the final form of what might eventually be AI as people? ↩︎
