I'm cautiously in favor of further AI development when done responsibly (which isn't happening right now), but any generated value needs to be fully socialized and not restricted, captured or managed by capital, nonprofits or other private interests. Regardless of your view on maximalist socialism, the AI case should be a no-brainer or at least highly compelling given the risks and benefits and especially the source of the data they're using to train their models. https://mastodon.social/@whitequark/112426645665489274
And even if all AI training data was legally in the public domain it wouldn't change any of this for me. If anything it would make the point more strongly about why we have a notion of public domain. It's to serve and benefit the public. I think getting lost in debates on authorial consent and licensing is missing the bigger picture; that said, I do support the class action lawsuits by Butterick, etc, as necessary and possibly useful within our current framework. Just my two cents.
If you currently do a search on "AI socialism" it means something very different: it's almost exclusively about AI-managed, AI-supported socialist planning. Incidentally, that does kind of have an interesting real-world history. Most of modern optimization theory was developed in the USSR to support economic state planning. A lot of the research didn't make it beyond the Iron Curtain for decades; Stephen Boyd's convex optimization lectures has a lot of jokes about Russians inventing everything.
This is not exactly a ringing endorsement of modern optimization theory but I don't think the well-known shortcomings of USSR economic planning can be blamed so much on the optimization methods or the limits of computation at the time. :)
@pervognsen a counter argument to that is the whole issue of central planning is there's no push for innovation. ML based stuff can only choose known solutions.
@morten_skaaning Of course you can allocate resources for research and development. What do you think a company does? How does public sector research happen? Anyway, I'm not advocating for central planning; that was just an aside on the two possible meanings of "AI socialism". (I also don't agree with the premise that ML-based methods can only choose known solutions because that isn't even true for solving a linear equation. I assume you're thinking about some notion of novelty beyond that.)
Add comment