I read a Medium article via a post on Mastodon a little while ago regarding the fact that it is possible to hold two conflicting viewpoints about something within one mind — Justin Cox’s Stop Choosing Sides On AI (unfortunately this is a “subscribers only” link for the Writers’ Collective - if anyone has a free link to the full version, HMU on Mastodon). I’ve also written a bit about this before in AI And Purity Culture when Cory Doctorow ruffled some feathers by espousing this same sort of dichotomy. Furthermore, in the time it has taken me to realize that I started this post but never actually posted it to PreemChrome, I’ve seen this argument ramp up even more in posts by my chooms in regards to things like art (the Bandcamp debacle and backlash where they laid off a bunch of devs, ostensibly because “AI”) and software folks that have hobby projects being vilified for having the audacity to use AI tools in limited capacity to help them on their personal, unpaid projects.
It seems that when it comes to AI technology as a whole, there’s absolutely no middle ground or living “in-between” with it — one can either love it (and thus becomes a “collaborator”) or hate it (being part of the “principled human resistance”). The thrust of the article I read was that this black & white, absolute oppositional view is untenable and yes, contrary to popular opinion, a person can hold two strongly contradictory opinions in their head at the same time and also actually live with both of them in various contexts. I have always wanted to see expert systems and machine learning come to be for practical and beneficial use cases, but certainly what we have been given in this time line is not that. Not really, anyway.
Ever since humanity invented the computer — and likely even before that — AI, machine intelligence, and artificial autonomous robots have been a dream of science fiction fans and authors and computer scientists alike. I remember taking AI courses and AI-related courses like cognitive psychology during college in the early ’90s, and that was just built on work started by Turing and others that had been going on for decades. It was always the vision and, I believe, is part of humanity’s inborn idea to create and even create things in our own image. Humanity has always wanted to create things to make our lives easier and more reliable, and the base idea of AI falls right into line with that. We have always wanted tools that make our lives easier; what could be more helpful than having a gigantic mind that can collate hundreds or thousands of information sources at 100 times the speed of the human brain and provide guidance, recommendations, and answers based on the distillation of all that information?
And in that respect, I feel like this explosion of model development at such a rapid pace is a good thing because now that these projects are funded, some good (and specific) models can (and have) come out of it that will be able to satisfy those narrow and beneficial-to-humanity use cases. Without this headlong rush pell-mell into furiously competitive “frontier model” development, these AI tools would still be essentially undiscovered and locked away in the niche data science arena. Having been forced to use Claude Code at work (while also going through the gamut of other coding assistant models), I can genuinely say that while the underlying code for that product might be inefficient shit, the products of that model are actually quite good in the hands of a moderately experienced software engineer.
It’s In The Way That You Use It
Eric Clapton
It’s in the way that you use it
It comes and it goes
And don’t you ever abuse it
Boy, don’tcha know
And yet, I’m in danger of losing my job by either being replaced by a coding assistant tool or because I’m not as efficient at using it as the person next to me or because for some ludicrous reason, the number of tokens one uses is somehow indicative of some sort of efficiency metric or some stupid shit like that. How now is the carpenter in danger of being replaced by his hammer?
I believe that the problem with what passes for “AI” currently is 100% solely due to the way that it was rolled out and brought to market in a capitalist, monetization-fixated society. And, perhaps in the most ironic twist in the story, it’s also a human problem. My short list of why I personally hate “AI”, all of which are primarily driven by capitalist and (more precisely) human greed:
- Humans were lied to about the capabilities of the models and those models were applied across the board, in many places where they should not be
- The models are, almost in their entirety, built on plagiarism and theft
- In order to get these models to live up to the lie their creators marketed, the climate, economy, waste, ecology, and humanity are suffering across the board in horrific ways to keep these models making money
- These models were not built to solve specific problems but rather to engage (and keep engaged) the largest number of users possible purely for monetization and profit
- Big Business saw — as with so many “labor saving technologies” — a way to get more work out of a single worker, while paying them less or not paying them at all (firing), all in order to keep more profits in their pockets for the few at the top
- Business has always wanted slave labor
- These models were imbued with false personhood and anthropomorphic intelligence that they do not - nor cannot - possess purely for “engagement” and monetization purposes
- Generic models produce generic output, leading to a collective lowest common denominator and the dumbing-down of the users
- These tools are inextricably tied to politics and governmental bids for supremacy and control
And of course, there’s the humans that use the damn things without understanding what they can and cannot do, how they came to be, the ongoing damage that they’re doing, and the simple maxim that just because you can see the moon from your backyard doesn’t make you an astronaut. And of course the worst classification of these users are the people that just don’t care about any of that. There’s this notion of “democratization” of these tools that the snake-oil salesmen like to crow about, but just because you can get a machine to generate whatever silly picture or song or book or application that pops into your head doesn’t make you “creative” or an artist. The machine can’t create, at best it can only synthesize from (usually stolen) sources. Here’s another maxim that more people would do well to remember: just because you can, doesn’t mean you should.
The danger of “labor-saving devices” has always been laziness and stagnation - when humans rely on the machines to do everything for them, we’re starting to get into Wall-E territory here. The kicker is, the capitalists will sell you on that “utopian” vision of living a complete life of leisure without ever having to lift a finger so you can do more “creative” things, but the hidden, insidious truth is that they only want that vision for themselves - “leisure and security for me, but not for thee” thanks to the magic of capitalistic pillaging and a new form of slave labor.
It’s a house of cards built with lies on a foundation of half-truths and omissions. Knowing all of this — and more — makes me personally feel dirty every time I have to interact with some LLM or coding assistant model because I know that whatever “Fiddle-faddling…” I initiate is one more straw on the climate and societal camel’s back. But if I want to keep my job, I need to use it as much as everybody else; but them principles won’t pay my bills. It becomes a lot harder to justify those principles when you’re facing the loss of your livelihood and living income.
Where to Go From Here
Doctorow constantly goes on about “seizing the means of computation”, interoperability, and the dangers of becoming reverse centaurs. Now is the time for humans to separate the good, useful models and tools and make them work for us — not the capitalists and the TESCREAL crowd. The hacker ethos has long since been mainly about “breaking systems that break people” and taking those systems and making them work in unintended ways.
In this particular case, I feel like one of those actions could be putting those models in the hands of the people and making them work for individuals without all the capitalist and profit-driven bullshit that comes with them. Figuring out how to run those models on one’s own hardware, on one’s own data, and for one’s own purposes. Most of those damn things have to consume so much energy because they’re trying to collate the entire damned Library of Alexandria (and all the seedy underbelly of misinformation and worthless input from the cesspools of the Internet as well) - imagine what those same models could do with a much smaller, more focused, and personalized set of training data? How much resources would they require to be as effective?
I see people like Cory Doctorow and various hackers and techies in my Mastodon feed or personal circles getting good and useful results from models that they’re running completely locally where all they had to do was take the time to train them on specific data sets. They’re taking what the big companies have developed and constraining it to their own hardware; they haven’t bought into the hype that “it takes a global cloud to reply to a simple prompt”. I’ve seen it personally as well through having been coerced into using Claude Code at work - when sufficiently constrained and bounded, I can get it to make a decent starting skeleton of a multi-service application in 30-60 minutes where it would take me probably 3 times that long to just write that basic stuff myself. Yes, there was a lot of prep work up front to write the constraints sufficiently scoped narrowly enough to be useful but once that was done, creating a new service becomes that much easier on a subsequent basis.
The capitalist money whores have given us the models; now it seems like it’s time for the hackers and techies out there to seize those models and figure out the ease-of-use to run those models locally and share that information with the rest of the world. Take the computing power out of the hands of the rent-seekers and put it in the hands of the residents. These models only need all these ridiculous and wasteful resources because they’re trying to be all things to all people in all situations; focusing and constraining them not only makes them more useful, it makes it much more possible to run on one’s own hardware - and it teaches the “common man” how to effectively use ML models for their own, personalized, every-day existence while keeping all their data within their own network.
And hell, if we really feel that we have to share computing power to get faster results or distribute the load, I say we bring back Beowulf clusters (remember those?) based on community sharing, not rent-seeking cloud providers. Smaller, shared loads just mean less and less need for overpowered and world-burning GPUs because a reduction in scale means a reduction in resources. It might not be as fast, but I can almost guaran-damn-TEE you that the answers you get will be better, more consistent, and higher quality.
I daresay that if we focus those models and make them runnable on local resources without having to “phone home” at all that will then eliminate a lot of the resource waste and societal damage that cause most people to hate AI in the first place. The effectiveness of a given model will be directly affected by what the user puts into it and makes them personally responsible for the resources they use. I’m pretty sure that would cut out a lot of the wasteful uses of these models pretty damn quick. But since capitalism will never let that happen without a fight, it’s up to the hackers to liberate these models and break these systems that break people to realize the true potential of what they can offer us so that we don’t have to live between these two diametrically opposed viewpoints.
