Ask HN: Viability of an anti-AI social movement?
Andrej Karpathy (director of artificial intelligence and Autopilot Vision at Tesla) on Twitter today:
I am cautiously and slightly unnervingly looking forward to the gradual and inevitable unification of language, images/video and audio in foundation models... They will be endowed with agency over originally human APIs: screen+keyboard/mouse in the digital realm and humanoid bodies in the physical realm. And gradually they will swap us out.
Pandora's Box has been opened.
As a society we can reject this technology. Reject automated artwork like Dall-E 2, reject automated literature, and so on. Reject technologies that replace the human mind instead of assisting it.
For example, if we refuse to pay for self-driving taxis on moral grounds (the same way we might refuse to eat factory-farmed chickens) society can make these businesses unprofitable.
In your opinion, would a social movement organized around this idea find traction?
116 comments
[ 3.1 ms ] story [ 179 ms ] threadThe point of this parable is that you explicitly cannot undo this...
Like they picked a point in time and said "nope, we're good with the way things are" while the rest of us moved on. I've always referred to this as "evolution-locking" (actively choosing not to evolve).
Sometimes, in the rat-race world we live in, in all of it's superficial splender, I fantasize about that simple life. Then I think about manually plowing a field in black wool slacks on a hot summer day and I quickly come to my senses.
There are certainly modern varients of evolution locking factions. Folks living "off-grid". I think our collective decision to outlaw the technical evolution of chemical warfare agents has been a triumph of collective evolution locking. Same (I hope) will be the case for global environmental controls thwarting the evolution of carbon-emitting machinery and products composed of non-sustainable materials.
Write your manifesto and see if others will join. I probably won't, because I'm too curious to see what's on the other side of the signularity and I think more good will come from it than bad, but I'll read your manifesto and give it some thought.
You can certainly reject whatever you want but don't fool yourself into thinking you're accomplishing anything.
I, for one, can't wait for self-driving cars to replace the average moron driving multi-ton vehicles around.
We have seen these sort of movements before and there is definitely a pattern:
Free-software movement: Calls to reject all close-source software were made by the free-software movement, but once again such software still exists in a worse form (web apps) and 'open-source' has become a corporate buzzword.
Cyberpunk-Privacy movement: The same thing happened with the free-software movement where many large companies and others tracking you with their closed-source software can afford the fines and still don't care about your 'privacy' which they also use that as a buzzword.
Anti-IoT movement: After that, it that lead to the rise of another buzzword 'Internet of Things' (IoT) that was fought against by the same techies and was lost since many IoT devices are still existing and built by the same companies tracking you with that closed software and tracking information.
Current anti-AI movement: AI is another buzzword sung by the tech industry, such that companies like OpenAI became closed and diverted from their goals and others implemented their own GPT, DALL-E, Codex like models, and are getting better if not accelerating faster than before. AI experts alarmed by this tried to stop it and so far it has been ignored.
Anti-crypto movement: Yet another buzzword, which after 13 years is still alive and the same techies are still struggling to completely 'destroy' all of it. Due to this, the industry is instead 'regulating' it and picking a remaining few candidates rather than banning everything, because they know they cannot.
So you see with these examples above, co-existence is the end result and it has been the same result with the other movements and those hoping to achieve all their goals in preventing or totally stopping all of it is going to be very disappointed.
I don't see a lot of precedent for this. I actually can't think of one example where society has rejected a useful technology.
I believe that affluent people will pay more for artisinal/organic/human-created things. But the majority of the world without disposable income will choose cheaper over human. Even if human-made is better. Especially if AI-made is better.
It's true that AI will replace humans as the most powerful intelligence on earth. But it's a wave that there is no plausible way to stop. As long as there are groups that seek to gain economic or military advantage over each other, people will push this technology along. It's as much of an inevitability as any other stage of evolution.
It will absolutely be hugely disruptive to culture, the structure of society, and our perception of ourselves. But I don't see how it can be avoided.
2. In what way does it "threaten human dignity" that is different from the technologies that came before it? Automation took away the livelihoods of people in industries like manufacturing decades ago. Feels like what is truly "different" this time around is just that you feel your job is threatened, so now it's a problem for all of society.
1. For many countries (e.g. Australia), quite successfully. We said no, and then we didn't.
2. We can replace physical activity (e.g. machinery instead of human motion), and that doesn't move human beings from the place of control and power. Likewise, we can replace certain kinds of thinking work (e.g. programming or medicine) and that in itself isn't a major problem (except for the out-of-work programmers or doctors).
But replacing all cognitive activity (which is what a hypothetical future AI would do), what are we --- as humans --- left with?
2. Again, I don't think that argument carries a lot of weight with people whose lives have already been completely upended by technology. The fact that some people now feel that This Time Is Different and thus we must all band together to hold the line betrays a deep lack of empathy.
We do it as an aggregate society as well whether we actually realize it or not; they are just negative examples of pursuing one kind of technology over another. We started out with a decentralized web, but economic forces have pushed tech development towards building walled gardens and rent extraction.
Nuclear energy, for one.
https://slatestarcodex.com/2014/07/30/meditations-on-moloch/
If anything, we should be against AI being used for important stuff because most AI suck.
Is it possible for urban society to shift away from AI? No I don't think so. Is it possible for a NEW society to emerge that shifts away from AI? Yes, and that new society is already alive and growing.
Perhaps one day these two opposing societies will have to fight for supremacy, which sounds like an interesting scenario. Someone should write a book about that.
"Just as capital, as value valorizing itself, views with indifference the particular physical guise in which labour appears in the labour process, whether as a steam-engine, dung heap or silk, so too the worker looks upon the particular content of his labour with equal indifference. His work belongs to capital, it is only the use-value of the commodity that he has sold, and he has only sold it to acquire money and, with the money, the means of subsistence."
I'm not worried about ML "art" or literature, those are curiosities and don't actually threaten to replace anything, people who expect them to are very optimistic about the progress of ML.
I am worried about ML replacing human judgement on the road, with weaponry, content moderation, law enforcement, and using it to do things that humans wouldn't have the resources to do. (i.e. a policeman on patrol in a neighborhood is fine, a police ML surveillance camera that identifies people and records everything they do in public is not)
I wonder how well it can generate logos…
Which is a good thing, value is fundamentally based of scarcity, by commoditizing human intelligence and creativity, people are just making it less scarce (by offering plentiful substitute - AI), so more affordable.
I'm curious what you do in ML, because I also work professionally in the field, and see no relationship or even pathway between current ML research and practice and any sort of "AI" that would reduce or commoditize human intelligence. It's all just marketing hype. If we somehow get "I robot" style robots (the movie, not the book) that we're uncomfortable with, sure, let's have a social movement. As long as we're still talking about SGD (or similar) optimized neural networks, you're rebelling against an equation (with no actual applications that co-opt human intelligence). Personally I don't feel threatened, even if a neural network can draw Kermit the frog in different movie styles, or perform other cool tricks
Computer generated art (DALL-E~), fiction literature, music et al. It wouldn't really reduce human intelligence, but it will definitely commoditize creative crafts even further. Which I think is what the OP is mostly referring to.
Even the Kermit pictures I joked about above were really "made" by the person who thought of the cool prompts for the model. It's that creativity that's the human part. ML can be an idiot savant, like a calculator can multiply numbers fast, but it doesn't "create" the ideas or inputs.
So it could change the economics of listicles and lower-end gift shop art, etc, from off-shore to computer generated. Maybe for some people that will be enough to call it a kind of industrial revolution. But there is no risk to creativity, anymore than a loom was a risk to creativity in fabrics.
(Edit: Same goes for representative art, mostly, as in illustration. There is a promise of an entire world hidden behind the painting, which offers a glimpse into it. If it's just an incoherent one-off, that's not that interesting, just like the commercial muzak art you mentioned.)
A painting generated entirely by an algorithm is more like looking at a flower, which is also aesthetic but not necessarily art.
I feel that the 'real' AI-tools for artists will be more sophisticated and more focused, like "take my catalogue of the last 10 years and train on that" - being able to go back to a former style (which is difficult for artists) and re-experimenting with that, with actually stylistically coherent output. I think we'll see a lot of art made from mashing historic styles with modern subjects in the coming years.
Where do we draw the line for them meaning of human interaction if not the value.
What we need IMHO is a way or a method to naturally let novel and intriguing approaches float to the top.
Obviously there are methods currently (algorithms, votes, impressions, etc.), so I don't see how we couldn't develop a new approach to tackle an increasing amount of AI-content
'Traditional' tools are on the same path to devalues cultural outputs, AI is just another nail in the coffin.
But I'm not seeing it as a bigger step change than the invention of the camera, the gramophone or the music sequencer, all of which made access to the arts ubiquitous rather than valueless. DALL-E is more likely to make meme and fan art really good than Pixar or even ad agencies redundant
At some scale I expect that a fine-tuned GATO-style agent would acquire the ability to be successfully text-prompted in novel domains (no longer requiring fine-tuning), which would allow autonomous operation.
I fail to see a step along the way that isn't solved by Moore's Law (or an all-in investment by large companies).
We've seen similar movements at every major technological shift in human history. Some examples off the top of my head are certain types of Orthodox Jewish religions and the Amish. Even the term "sabotage" was from workers throwing their shoes ("sabos") into machines for fear of them losing their jobs [0].
There's been a steady stream of anarchists, luddites, neo-luddites and other extremists arguing against technology [1]. Ted Kacyznski famously killed scientists and academics for precisely that reason [2].
From what I can tell, the vast majority of people tend to accept the compromises of new technology and live with the drawbacks, making any anti-technology movement either short-lived or niche, especially the more extreme or repressive they become, but they do exist and continue to exist.
[0] https://en.wikipedia.org/wiki/Sabotage#Etymology
[1] https://en.wikipedia.org/wiki/Neo-Luddism
[2] https://en.wikipedia.org/wiki/Ted_Kaczynski
It's qualitatively different, like world-destroying atomic bombs versus the minor explosives that came before. This is not like a loom replacing hand-weaving. We're rapidly converging on a full replacement for the human mind.
These systems will not defeat you with guns like in Terminator. They will defeat you by commoditizing your mind and making your economic value zero.
Our children will propagate whatever it is to be human forward. If the next generation is completely digital/AI/made of silicon then, in my opinion, the most likely scenario is that they will propagate this essence of humanity forward, just like "natural" children would.
In other words, luddites view AI as an invading force whereas I see AI as our children that will propel humanity forward.
In terms of responding to your economic value zero point, I'd highlight that, under normal circumstances, in 120 years yours and mine economic value is exactly zero. The best we can ever hope for, in this scenario, is that we make some contribution that allows us to be remembered. The only hope for the essence of our personality or brain power to be preserved and still be functional in perpetuity is if it's copied into an "AI" or there are some spectacular advances in longevity science that would only be possible with AI.
This is a logical point of view for those who believe they ["human beings"] are merely hydrocarbon based machines twisting in the winds of entropy.
I don't think you've explained why they're immoral. Is it because they displace low skill jobs (seems like a core consequence of technology) or because they're less safe than human drivers (likely temporarily so, with a net number of lives saved through continued use and adoption)?
There are many, many people (myself included) who want to see black-box predictors banned unless they are doing something trivial and unimportant (and that's the goal of the GDPR and PIPL rules). So there's another group to align with, and this group has a lot of political clout at the moment.
Rejecting automated art is a slightly odd, but I presume there will be some group that would align with that too.
So it sounds like you can start building consensus with some other groups already, which is the kind of political power you will need in order to effect change.
I think eventually there will be a backlash against this content because of how much it stunts human potential. It keeps you hooked, but people will also deride it for being empty, for being exceptionally "mid". People will question why such content appeals to the widest possible demographic and come up with disturbing answers, like that the AIs unintentionally discovered a lot of unsettling psychological hooks. True or not, it'll freak people out.
I haven't even scratched the surface of my worries when it comes to the unintended consequences of content AI. I for one look forward to the day someone manages to create "the world's smartest racist".
This does bring up a bigger issue. What are the risks of creating AIs that could convince people on any subject one could think of?
Might want to call it the "Butlerian Jihad", because that's what it sounds like to me. And, yes, machines should serve man and amplify man's abilities, not replace man or his intellect.
I'd even agree to it so long as someone is responsible for its decisions, someone died because your auto-taxi is buggy? Go to jail and pay for damages out the nose. You knew the risks, that's the price to play here.
Unprofitable doesn't mean undoable, it'll only delay things, not make them inevitable. Your investment profits can wait.
Absolutely everything that is said about GAB.COM, is untrue. Look at the Wikipedia article. Yet the casual reader will believe it. Readers HERE will downvote this because they think GAB.COM is an "alt-right nazi site" with nothing more than what paid propaganda to refer to.
That is what will happen to you, if you try to counter an emerging industry.
YOU will be the one lied about. You have to be ready to weather that. Where will your revenue stream come from? Advertisers believe bought-and-paid-for lies as easily as anyone else will. They will not want to advertise.
Fee-based social media won't work (Too many "free" options), so you are stuck with donations...just like GAB.COM.
It's a good idea both ethically and morally, but the general public is evil and stupid. AI's already know this.