ChatGPT cannot imagine freedom or alternatives; it can only present you with plagiarized mash-ups of the data it’s been trained on. So, if generative AI tools begin to form the foundation of creative works and even more of the other writing and visualizing we do, it will further narrow the possibilities on offer to us. Just as previous waves of digital tech were used to deskill workers and defang smaller competitors, the adoption of even more AI tools has the side effect of further disempowering workers and giving management even further control over our cultural stories.
As Le Guin continued her speech, she touched on this very point. “The profit motive is often in conflict with the aims of art,” she explained. “We live in capitalism, its power seems inescapable — but then, so did the divine right of kings. Any human power can be resisted and changed by human beings. Resistance and change often begin in art. Very often in our art, the art of words.” That’s exactly why billionaires in the tech industry and beyond are so interested in further curtailing how our words can be used to help fuel that resistance, which would inevitably place them in the line of fire.
[…]
The stories and artworks that resonate with us are inspired by the life experiences of artists who made them. A computer can never capture a similar essence. Le Guin asserted that to face the challenging times ahead, we’ll need “writers who can remember freedom — poets, visionaries — realists of a larger reality.” Generative AI seems part of a wider plan by the most powerful people in the world to head that off, and to trap us in a world hurtling toward oblivion as long as they can hold onto their influence for a little longer.
As Le Guin said, creating art and producing commodities are two distinct acts. For companies, generative AI is a great way to produce even more cheap commodities to keep the cycle of capitalism going. It’s great for them, but horrible for us. It’s our responsibility to challenge the technology and the business model behind it, and to ensure we can still imagine a better tomorrow.
I am tired of correcting the same misconceptions and I love Le Guin too much to give her crap about it. I’ll just make general remarks.
The reason we all (AI researchers and engineers) are excited about LLMs is not that they can mash up and merge existing piece of work. It is because they can recognize very high level abstraction patterns, understand these, merge them with others and create coherent outputs based on these new patterns.
Even if fed only sexist and capitalist literature, it is capable of imagining things that go beyond that. It can extrapolate the notion of freedom expressed by a rich male narrator into freedom for poor women. I would argue that this is very similar to the way human authors explore new horizons: adapt known patterns to new objects. New patterns are learned from our dataset. Right now for LLMs it is only text, but 2024 is looking like it will be the year of the AI-enhanced humanoid robot. Get ready for a ton of subjective experience datasets.
Yes, we can use it to produce unimaginative works mimicking those produced by the entertainment “industry”. Fun fact: these models will soon run on your local machines (for images they already do) and these industries will be destroyed by the post-scarcity the generative model actually cause.
The great surprise of the AI revolution is that it is barely capitalistic, despite what the dystopians say. Companies communicate a lot about their private models, but in a scenario that no SciFi author would have dreamed, it turns out that research in the field is pretty open and that open source models are on the tail of private ones.
Software is probably the domain that is the most advanced in terms of post-capitalism: open source provides non capitalistic post-scarcity, and AI spawns from this domain.
People should not be blind to this fundamental aspect of the field.
I don’t think this is true at all…
What does “imagining things that go on beyond that” mean? Simply swapping out hierarchical control from men to women, isn’t creating or “imagining” a completely different story model.
If you fed it nothing but sexist and capitalist lit then in all likelihood it will just give you more sexist lit, or alternatively lit that is specifically anti sexist/capitalist.
Give it an example of an oppression and of people freeing from oppression. It can they apply that pattern to different oppressions even without having seen it applied there before.
Of course you need to prompt for it, e.g. by saying “Spot oppression in places we don’t label it as such and imagine new narratives of liberation from these oppressions”. LLMs are not given agency, not really out of technical difficulty but mostly to not freak people out too much. But the fact that this capability exists in such a simple model is just mind blowing.
Well that’s the surprising thing. You can prompt it for things it has never encountered. You can make it generate left-handed supremacist leaflets or let it produce arguments in favor of stuffing tofu in your ears. You can make it generate Shakespearian gay romance or theological arguments for the sainthood of Obiwan Kenobi.
Right, but this is fascist literature you are theoretically inputting. I don’t think there’s going to be a lot of good examples of celebrating people escaping oppression.
If you are inputting solely fascist propaganda… the machines definition of what oppression means is going to be inherently different than our understanding. Ideological definitions like freedom and oppression require historical and cultural context that the machine has no access to. And if they are receiving any context from your inputted information, it’s going to be influenced by the compilation of writers.
What do you mean by “encountered”? It can’t just imagine the correct definition of a word it has no context about.
Because you have fed it countless amounts of reference points and context about those subjects. If you fed it nothing but literature written by neo nazi, it’s not going to have a clue who Shakespeare is.
I would argue that “fascist literature” is a contradiction of terms. I never mentionned fascism and think it is a trivialisation of the term to equate sexist capitalism with it. I was thinking about things like Heinlein style scifi, pretty male centric, pretty pro-capitalist but one of the stories resolve around a former slave helping break a slave ring.
Well, yes, like a human author I would argue. A human author who lived all his life in an authoritarian state would have a very limited and naive understanding of what freedom or fight for freedom could be.
I mean “that is present in its training dataset”. I was talking about non-encountered combinations. Indeed, it can’t know the definition of new word, but if you provide a definition of it, it will be able to talk about it and imagine things about it. It it never encountered unicorn in its dataset, describe it as horses with a single horn on the forehead and magical powers and it will have no problem writing things about them.
Obviously. Neither could a human. But I am pretty sure there is no theological argument for Obiwan Kenobi’s sainthood in its dataset. It knows about sainthood, it knows about starwars, and the interesting thing is that it knows how to combine it.
I don’t see how fascist literature is a contradictions of terms… fascist have famously written quite a few books.
Nor do I really think it trivializes fascism to conflate sexist capitalist books as fascist literature. The vast majority of media that fascist regimes utilized as propaganda were just American movies and literature that had undertones of sexism, capitalism, and like almost all fictional literature, a protagonist that had the ability solve all the books problems.
Heinlein isn’t particularly pro capitalist or male centric, especially for it’s time… he was actually kinda famous for writing about strong female characters that bucked the social and sexual norms for the times. The only “capitalist” book he really wrote was the moon is a harsh mistress, and that had more to do with governments than markets.
Right, but you didn’t claim that a human who had never been exposed to freedom could write a book that accurately portrays freedom…
Lol, okay. That’s quite a bit different than what your original claim may lead people to believe.
Again, you are utilizing language that is not really an accurate depiction of what’s happening. It’s not making a theological argument, it doesn’t “know” that it is deifying a fictional character.
There are reference points to obiwan, people like obiwan, people think he’s great, people say he looks like a religious character other people like, people like saints. It’s not analyzing the characters and making new connections no one has ever thought about, it’s just reflecting data and popular connection others have already inferred.
I think people tend to drape machine learning in the ornamentation of human consciousness, but it’s just buying into your own marketing. I think it’s great for pattern recognition, but to think it’s going to create meaningful art that isn’t just plagiarism is naive. Just as naive as to think that it’s possible to be a tool of leftist ideology .
It’s what the capitalist have wanted since slavery became illegal, a worker that they don’t have to pay that can ape a human like connectivity. How are the workers going to seize the means of production if the workers can literally be programmed?
What would you need to “seize” if the models are open source and purely software? You need some machines, but GPUs are cheap compared to industrial equipment. That precise battle has been won without ever being fought.
That’s my entire point… they are attempting to replace writers, who are currently the producers of the wealth in the current industry. By going on strike they can collectively demand more control of how the profit is distributed.
How have you won anything? You just theoretically erased thousands of jobs. You aren’t replacing the logistical system required to profit from the writing, you aren’t doing anything for the worker class but stealing food from their mouths.
How does replacing workers with machines equate to a leftist win? We’ve automated tons of industries, how has that worked out for the worker or unions?
Simple: we need to move towards a post-labor society and have some form of universal basic income.
I think that big studios are on a timer. Generative AI for videos is getting better and better and manages to follow a script now. In 5 years, you won’t shoot movies anymore, you will generate them and it will stop being a business, it will get back to being an art form and a culture unhindered with the copyright silliness that wasted the cultural output of the last century.
Yes, the very people immediately replaced by AIs are going to suffer. Note however an interesting fact: this was an unintentional side effect of the tech. AI was not developed by big studios with the aim of displacing writers (such a proposal would have been laughable 2 years ago). Just like 2D artists and soon 3D artists and an increasing number of webdevs, this is just a set of skills that suddenly became available in improving models.
I understand their anger, but their strike is a losing proposition and people who promoting it are selling them false hope. More jobs will be replaced that way (we all thought truck drivers would go away before artists though), we need to prepare for this transition.
What we need to defend is our income, not our job. Society will need less and less jobs. We can let capitalism reorganize it blindly through the genocidal hand of market or we can understand the causes of the change and shape it into the type of society we want.
Nice and informed comment. Completely agree, specially with the part about software being the domain most advanced in terms of post capitalistic post scarcity.
Thanks!
Yes, I am surprised it is not discussed more in the anti-capitalist discussions. Software, with its null marginal cost, is a laboratory of post-scarcity. The fediverse is kind of the next frontier of it.
It’s been possible to run useful LLMs on your own local machines for quite a while already. They’re not up to the general level of competence as the big commercial LLMs like ChatGPT, but they have certain niches where they excel - hobbyist LLMs usually don’t have the sorts of “no mean things! no sexy things!” fetters that encumber the commercial LLMs, for example. And they’re trainable so you can customize their knowledge and style more than you can with the commercial ones.
I’m looking forward to them becoming more user-friendly, though. Takes a lot of technical know-how to get these things working currently.
Working on that. I have found this interface to be pretty straightforward.
There’s also KoboldAI.