So, lately I have been seeing AI bros always using the “It democratizes coding/game dev/modding/drawing/music creation/video editing/insert a hobby that requires actual skills” argument. And it always just irritates me.
Like, democratizing a skill means giving everyone an equal opportunity, which is what free (specially FOSS) tools do, think Krita, Blender, Godot, etc… You get the tools that you need for said hobby free of charge thanks to the GOATed people that maintain them. And then you learn it.
But AI, it just works on the basis of who has more disposable income to feed the capitalist machine. Like where is the “democratization” when the person with more money gets the better experience. Isn’t that the era of paid software all over again? isn’t that literally the opposite?
Yeah, so that is the end of my rant, I just wanted to vent out to people who will understand that sort of thing (talking about a fine echo chamber ;) )
And I hope people point out that stupid argument when they see meatproxies using it :)
I think there are two arguments that you’re making.
One is the anti-capitalist angle, corporate ownership of AI only serves to amplify the growing income inequality. It’s build on the destruction of much of the personal computer industry, the labor market, the environments around the datacenters, etc and all we get for this incredibly high cost is the chance to forever pay a rent seeking corporation who stole the collective sum of human knowledge. I super agree with this.
The other is the technology angle. The AI run on the massive datacenters are more capable than the open weight models available to the rest of us so anyone paying into the corporate AI machine is able to have an advantage over anybody else. Agree with this too.
I don’t see AI as replacing actual skilled people who’ve dedicated their time to a hobby. An AI isn’t a better programmer than a person with years experience and it does not create images or music than an artist who’s dedicated years to their craft.
But, it does raise the floor for what a non-rich person with zero training or skill can produce. Most people can’t afford to hire a programmer with 30 years of experience or an artist who has decades of experience but they can produce poor quality but functional code with AI.
It’s kind of like how libraries democratized knowledge. Sure, if you were rich you could afford to buy books, hire private tutors or attend college and that gives you an advantage but libraries gave regular people access to knowledge without the monetary barrier. The rich person still has every advantage but a library creates a path for working people to access knowledge.
Definitely fuck the corporate AI, but the people in the FOSS community who’re working to pry AI for corporate hands and make it accessible for everyone don’t deserve to be lumped in with Sam Altman.
Running a model locally still comes with considerable costs in terms of hardware and electricity. So until it can run on an office laptop with only an integrated GPU, the financial barrier is still quite a bit higher than just having access to a PC, software and learning material.
Also, even FOSS locally run AI models are in practical terms “owned” in a way that most other open source software is not. To make a comparison: if vim were to somehow prevent me from specifically writing the long-awaited sequel Communist Manifesto 2: Communist Harder, I could look in the source code and see why this is happening. Of course this requires skills that not everybody has, but there are enough people with that skill set that it’s unlikely things will stay that way without consequences.
But if an LLM has some kind of slant that causes it to favour certain output, what do I do? Do a big data analysis on the training data, then redo the training with the biases removed from the training data, all without the funding from people who probably want the slant to be there? You have less ability to hold the developer accountable to the users, but at the same time, the high costs involved in making these means you don’t have many real alternatives to choose between.
Lastly, this:
This is no more democratising than a carnival shooting gallery handing its customers an AK-47 is. Yes, someone can ask an AI to create a program that does a thing and then verify whether the program produces the result they expect, but they have no way of verifying whether it’s doing what it should be doing and only that. Is it hammering the server it’s interacting with? Is it making network shares accessible over the internet without credentials? They have no way of knowing, because they don’t have the necessary skills. Increasing people’s ability to accidentally and unknowingly cause harm to themselves and others isn’t empowering them.