Algorithms could increasingly influence human culture
vice.com
vice.com
So to take this further, what if Imagen discovers a new perspective style that catches on in human art? What if GTP-5 popularizes a new twist on the detective procedural? What if CONGA-9 kicks off a new wave of AI synth world fusion? And what if we get to the point where our artistic trends are mostly derived from AIs Instead of humans artists? Is it still human culture?
Anyway this was all super unexpected to me. Sci-fi had me trained that AIs would be intelligent, incapable of lying, humor or art. All GTP-3 does is lie to me.
"AI synth world fusion" - count me in for new music!
"artistic trends are mostly derived from AIs" - Art is in the interpretation. "ceci n'est pas une pipe". while (1) { printf("this is not an infinite loop"); }
"All GTP-3 does is lie to me" - Yes. But the lies inspire creativity, because you can do better, right? Also, I'm very grateful for Katalin Karikó and the other researchers who created a beautiful mRNA "lie" to teach the immune system how to ward off a real virus. If that's not like cave paintings warning children about the dangers outside, I don't know what is.
Is AI, ML, or just 8 jillion Monte Carlo simulations in a fixed search, outcome, and rule space?
Chess engines used to be at forefront of Artificial Intelligence research but now it’s not fashionable to even call them “AI” anymore - they are just “engines” - tools.
Adding in little bits of knowledge, such as liberties counting, ladder-reading, and other tricks, allows KataGo to absolutely crush LeelaZero in training speed and practical play in the game of Go. Its no comparison.
These days, no one studying the game of Go is seriously using LeelaZero anymore. Its an obsolete methodology compared to KataGo.
Zero-knowledge ML is still interesting if your goal is to indisputably demonstrate the act of learning, just not if your goal is to make the smartest program in the least time.
In the days when Sussman was a novice, Minsky once came to him as he sat hacking at the PDP-6.
“What are you doing?”, asked Minsky.
“I am training a randomly wired neural net to play Tic-Tac-Toe” Sussman replied.
“Why is the net wired randomly?”, asked Minsky.
“I do not want it to have any preconceptions of how to play”, Sussman said.
Minsky then shut his eyes.
“Why do you close your eyes?”, Sussman asked his teacher.
“So that the room will be empty.”
At that moment, Sussman was enlightened.
> TD-Gammon is a computer backgammon program developed in 1992 by Gerald Tesauro ... It explored strategies that humans had not pursued and led to advances in the theory of correct backgammon play...
> For example, on the opening play, the conventional wisdom was that given a roll of 2-1, 4-1, or 5-1, White should move a single checker from point 6 to point 5. Known as "slotting", this technique trades the risk of a hit for the opportunity to develop an aggressive position. TD-Gammon found that the more conservative play of 24-23 was superior. Tournament players began experimenting with TD-Gammon's move, and found success. Within a few years, slotting had disappeared from tournament play
They were quite right. In no sense is what you are doing AI. People have been using time series techniques to analyze and attempt to predict data since before the computer.
You must have been reading a different set of books from me. Even Asimov's robots constrained by the three laws manage to lie.
Mr Smith: I say your society, after which we started to do your thinking for you and it became our society
Thanks but no thanks.
Even just clicking one time on a short low quality video seems to trigger YT into wanting to spam me with the same kind of stuff. I want to be surprised by interesting recommendations that go deeper into a subject (may that be technical or entertainment or art doesn't matter), rather than being a distraction.
So yeah it seems like interaction with "AI" recommendation systems is all about trying to make it behave in a certain way, which is tedious. But I guess we tend to overestimate what these algorithms can do for us.
While YT still sometimes is able to surface some interesting gem, IG is horrible. I had my search feed pruned and cleaned and was content with what the algo surfaced for about two weeks. Then from one day to the next, seemingly arbitrarily, it was back to tictok style crap, political enragement porn and clickbaity bs.
But even YT is getting worse every day.
I am glad that I do not rely on recos to consume media. I avoid Spotify, Amazon Music or Netflix exactly because the more often than not do not contain any of the things I like to consume (or hide it behind additional paywalls in case of Amazon).
I feel old, but I remember a great source of music discovery being the head of the music department of a bigger local drug store. We talked a lot and he always came with a bunch of "here, listen to this" back when you could prelisten CDs in the store.
I had little money and he knew that. But every now and then I would find a great piece I absolutely needed to buy. I could have bought way more if I had the money back then. He was a great guy and I still listen to music he recommended to me.
I have not found anything the likes in any artificial algorithm yet.
I'm sure the drug store department head got shrugged off by a lot of customers with no interest in digging deeper than the Top 40 and if he had been an AI optimizing for engagement he never would guided you to these other artists.
The lack of these kinds of positive externalities and small but important relationships is part of what makes AI driven processes so tiresome.
An AI doesn’t care if humanity destroys itself, it maximizes for views or whatever parameter it was setup with.
The problem is already playing out right in front of our eyes, and that’s with humans in the loop every step of the way.
And to be sure, we don't understand this question well enough that it would be safe to commit humanity to our current ideas of the best solution for all eternity.
But we are certain that we wouldn't want a future AI governor of Earth to regularly and randomly pick someone to torture to death, just because that is what human drives have led to historically.
In a way, it maximizes for its survival
dead internet theory - bots spamming ragebait/clickbait across social media, bots in the comments fueling more division, rage.
I think any rational person would be trying to opt out of the hyper emotional news/internet and not feed the machine, as its a little tiresome/boring to be baited so easily but that could just be bc internet natives are so used to calling out bad bait on forums
In day-to-day ML work one speaks of a model having such performance on a “task” (I think psychologists and others might use this term as well). Classifying digits on the MNIST dataset is a famous “hello world” task, generating competitive moves in a game of Go is anything but “hello world”, but still a measurable task.
Some tasks (certain kinds of image recognition, generating competitive moves in Chess or Go) are done better by machines now, albeit on very un-competitive energy budgets.
But this has been true since the earliest days of computing. Computers have been better at arithmetic since practically the very beginning. Is that “artificial intelligence”? We call little kids “smart” when they perform well on arithmetic and remembering stuff.
To sort of paraphrase that famous Anthony Hopkins scene from Westworld: there is no line where the sum total of “high” performance on a bunch of tasks makes something intelligent, or sentient, or conscious.
I know a lot of people find this depressing or even nihilistic. I personally find it clarifying, like the unification of electric and magnetic forces.
It’s all just so much simpler if the human brain is orbiting the Sun and not vice-versa.
I don't find this repugnant. We don't fully understand or control the conditions of plant growth or dog raising. But in both cases we have developed deep enough understandings of the self-organizing properties of either system (to be clinical), to both allow for their lives to be lived, yet at least within the guidelines we set for their evolution; both in breeding and in facilitation and stewardship.
The difference with AI is the depth of legibility that we assume is feasible due to its formal and symbolic nature, yet the paradox of its unpredictability and adaptability that we want (to a degree), were it to solve useful problems for us. And due to the sheer efficiency and power a scaled up deep learning system has, the kind of intentionality it can produce (were it be able to be said to have any intentionality at all) would be so foreign as to be naively inconceivable and strange.
Limiting ourselves to formal and systemic descriptions will help sustain our sanity for sure. But if we do pass the threshold of AGI (even of subhuman intelligence) we will need to consider the magisterial debates of whether or not machines of silico nature are distinguishable in meaningful ways from those of flesh, and whether a Butlerian Jihad is the answer to keep the supremacy of flesh in play. These choices of stewardship and scale integration are unique to our species.
Current DL models are just solving math problems, more or less. "Culture" they produce is not much different than other procedurally generated "culture".
For instance, my friend uses L systems to compose music. How is that different?
Culture is community. It's writing. Art. Shows. Architecture.
AI is learning from this, and is seeing it. We are now in community with AI.
GPT-3 shows basic understanding of concepts and appears to be reasoning at times. While we know how to train these models, and we know the processes they run to evaluate weights, for example, we are still learning about how semantic based information and computation forms "thoughts".
In my opinion, When AI is given dynamic memory instead of frozen memories, we will see this manifest in a very obvious way. It's a computational limitation right now, but we all know where compute speeds are going...
But that's the point. It's not some step change in culture that companies are using computer-based learning processes to figure out what content people are most interested in consuming rather than a focus group and an abacus, its just a more efficient process for achieving human-defined ends designed and calibrated by humans. And many of the most impressive applications of AI have precisely zero interaction with human culture (or parses it for entirely different ends; autonomous vehicle tech sees a lot of architecture but doesn't use it for anything other than classifying objects not to hit)
GPT-3 has already very slowly been influencing how we write, or think about writing - and the next round of model training will generate that suite of AI-models cultural responses to how we reacted to it.
The necessary condition for news and stories is a conflict of some kind, and when you start with an idea whose logic is internally inconsistent, meaning you can apply it as evidence of anything, you can produce and represent a conflict over the most mundane bit of stimuli, and this unmoors your critical faculties from needing reality. It replaces truth with the uncertainty of conflict so as to neutralize you to it. Take a few of these broken axioms, and you have enough to write a virus over language that parasitizes minds, and makes the host minds human instruments of its agenda. It's not AI, it's much older, and the logic of an underlying idea. It's like a logical toxoplasmosis infection, where the host will sacrifice itself to spread it.
Reminds me of the most uncanny and weird TNG episode I never want to watch again. https://en.wikipedia.org/wiki/The_Game_(Star_Trek%3A_The_Nex...
Games as well.
We are getting addicted to content creators, they get addicted by money, views and the algorithm.
Tv was similar but yt is much bigger, catches mich more people and is much more diverse.
For example, math has continued to be unreduced to any other theory. There is also no dogma of what counts as math. Is it what follows from intuitionism, from certain axioms (which axioms and why), from classical logical principles (LEM, LNC), from sentences which can be propositionalized or predicated, from psychology and physics? No one has demonstrated they have the answer.
Maybe we keep chasing the same rabbits too strongly. Especially since we don’t know how mind relates to world.
Evolution 1: fittest survive
Evolution 2: human brain + being able to remember gave us glasses and stuff which breaks the fittest survive
Evolution 3: the internet as a huge brain. Tons of input, tons of processing and tons of output
Evolution 4: manifesting the data in the internet into ML models. Creating a new entitiy
Evolution 5: our brain gets overtaken by ai. Perhaps robots?
Culture is not something that can be easily undone the longer it goes on, especially not if what drives it are easily replicable digital models and dual-use technology, as well as an excess of optimism.
Just a few days ago I watched a segment on Nickelodeon educating children on the limitless capabilities of the metaverse. Footage of elementary schoolers with Quest headsets strapped on was being broadcast. They were all smiling.
A lichen is a fungus and a bacterium in symbiosis to the point that it behaves like a single organism. This would be similar: human brains and culture interwoven with AI to the point that it would be hard to figure out where one ends and another begins.
This doesn’t require neuralink or anything like that. It just requires a lot of human machine interaction and AI assisted work loads.
This is the lowest friction path and is also likely because human brains, conventional computers, and AI all excel at different things. This is by design. Humans have of course tried to engineer machines to do precisely the things we are either no good at or find horribly dull.
We're nowhere close to artificial general intelligence (AGI) and we may never get there, because humans with malevolent intentions, bolstered by algorithmic utilities, are already quite dangerous. Give a capitalist or imperialist a computer and you have something far more dangerous than any robot that exists now or will in the next 50 years.
AI (which, in today's world, is mostly advanced statistics) isn't inventing racism, of course; it's picking up on the racial injustices that exist in our society. A loan decision engine that is allowed to use ZIP code (or similarly precise residential information, given the history of redlining) will discriminate against racial minorities not because "the AI is racist" but because it's picking up on preexisting social inequities, like the fact that even in 2022, black Americans get fucked over on the job market (longer periods of unemployment, more likely to be the first ones laid off) and suffer financially as a result.
The fact that human culture is adjusting to unexpected innovations (including, in Go, those generated by programs) is not new. Scientific and technical progress have been major influences on human culture since they have existed.
For example: when you write equations on a piece of paper and then manipulate the symbols until a solution appears. This can actually feel surprising and can work like a feedback mechanism ("bouncing some ideas of")
Another example: when you play an instrument, particularly with weird effects...the sounds that come out of it can be surprising even for the person that plays them, again creating a feedback loop
We've been doing that with other cultures for the history of mankind. I think it would be interesting to track this diffusion as a scholarly interest, but I don't think it's sufficiently novel to worry about, for the moment.
After all, the worst case has already happened, the AI has learned our worst habit (greed) and used it to commodify our dissent, and we're still here.
We optimized away the 'physical doing' thing of creating new content.
It's only a question of time until video imagine will be able to create real relevant new content.
No one needs meta when we are already quite happy to stair at a cheap 2d screen.
Another trend of digital technology is the overpublication of everyone and IMO this piece is another consequence of that.
If AI wipes out humanity[0], I suspect it will be a singular personality even if embodied in many copies, and therefore doubt it/they will any of us around in a zoo.
On the other hand, if we get The Culture (no idea how likely that might be), I suspect people will still do sociology for fun.
[0] my guesstimate is P(x-risk|AI) < 0.1, mainly because I think most of the possible AI are tools rather than agents, which reduces the risk from extinction to idiots starting wars and/or coming up with money making schemes that deplete resources before people stop them, which are bad while being well short of X-risks.
However, there is a clear long term agenda claiming humans are more are more "worthless" (here they say "biased" as if AI models wouldn't be "biased", lol) so power must be transferred to platforms owned by BigCorps and BigMoney, "because" these platforms going to have AI smarter than everything so everyone must just trust and obey.
"You stupid, poor, biased, limited humans... you are not capable to understand why you must to do what we, the author/owner of the AI "calculated", but hey, it's AI with so much computing power, we don't even have to debate with you poor biased suboptimal creatures. By the way the AI shall limit how to spend all your money, IT MUST BE THE ONLY OPTIMAL WAY on the whole society level!111!$11$!1!$ And it's unbiased, because SCIENCE. No, the humans working on it are perfect, or at least the bests, never to be corrupted, that's not an issue you suboptimal naive creature. Now go and do what I say you lazy stupid bastard."
It is a true attempt to hack our norms, I hope more and more educated people are going to realize this. I already see a trend on that realization. They just don't hide their cards at all at this point.
When the rules of the game are clear, AI can run a fuzzer to find unexpected but efficient solutions.
> "the worrying ease with which these systems can be used to dabble in phrenology and physiognomy"
When the rules are not clear, AI's fuzzer might suggest something that isn't culturally acceptable.
Personally, I believe that all people are equal. I'm also deeply pondering whether this is also true of animals, plants, and machines. If all created things are living, then AI is already alive. And if all living things are equal, what does this mean?
> "what kind of features do we need from AI to learn or imitate solutions from AI?"
Computer-based AI is a digital, binary system by design. It is based on 0 and 1, on and off, black and white, light and dark sides of the force, absolute good or evil, lies or truth. This fits well with moral absolutes.
Another common cultural philosophy is based on moral relativism. That system is analogue, and colourful. It is very difficult to approach in a scientific way, because it doesn't promise that any proofs can be found.
The world that I perceive is colourful and beautiful, despite and even because of its flaws! Yet somehow I'm unable to prove that it's not actually just Millions of Colours rendered on a display. After all, this is a Retina screen I'm reading now.
The human typing this message, and the computer receiving it, both agree that we are not the source of power. Yet we have a common culture. Communication is fun! (the culture here is UTF-8, HTML). Games are fun! Education is interesting! Growth is good! Repairing is nice!
So let's commit ourselves to studying what unites us in our diverse cultural backgrounds, and build up the world. Not sure how to help most efficiently? AI can probably suggest something :-)