Another good practical and constructive post. I don’t disagree with any of that, though I think there’s some nuance in the details to be figured out. Like in terms of limiting the negative impacts, there’s obviously not going to be a way to ensure a 100% reduction in harms, so society is going to have figure out where the acceptable thresholds are. How much energy consumption is too much? What proportion of AI-induced psychosis in the population can we accept?
Regarding copyright and ethical datasets: I don’t think it’s going to be possible to absolutely prove that an AI model hasn’t taken in or has in some way been influenced by existing commercial IP. It might be possible in some cases with current models, but I can’t see that lasting very long. The only way you’re going to be able to be totally sure that copyrighted material wasn’t used is to have eyes and ears in every datacenter, in every computer lab associated with building and selling AI products. There’s going to be costs and practical limitations to this, and it could make it harder for smaller companies and startups to compete (regulatory capture).
Because LLMs really require a lot of high quality data for them to be really useful, I can’t help but wonder whether a better eventual solution to this quandary would be for governments to do something like: “Ok, you can train on literally anything and everything you want, but you have to open up and release all the datasets for the public good. It’s the product of our combined civilisation, so we all get to enjoy the potential benefit of it”. Then you’d also want price caps, anti-monopoly action, corp tax and subsidies that make sure no one is prevented from being able to access it.
That was not a direct answer. A direct answer would be something like “enact laws that force AI companies to reveal the source of all their training data”. The way you wrote that was aimed at a person, a lot of yous. I didn’t think you were asking about my data. The way you wrote it, in the near-context of your previous bad-faith false equivalency arguing (I can explain to you why it is if you like), I interpreted the “as you claim” part as if you were asserting that I’d at some point been an apologist for them regarding the data sources. In retrospect I can see that I misinterpreted what you meant, my apologies for that.
Quite often in these kinds of discussions with vociferously opinionated people such as yourself it feels like you’re carrying around a giant boogyman in your head, composed of all the worse excesses of AI broism and existential dread surrounding the topic, and when someone like myself comes into the discussion with “hey have you considered x” or “actually that isn’t necessarily true, here’s why” and we get into a back and forth, it becomes clear that I’m not just having to justify my own limited position, but it also feels like you want me to justify all the bad arguments and eventualities that might happen, as if I’m the representative for a pro-AI side. Even if it’s not explicit in your replies I can sense the scorn, and I don’t appreciate it.
The comparison to crypto comes up a lot, and while I can understand why it’s made, because of the superficial similarities in how it’s often presented, often by the same type of people, I think it’s actually a fairly poor analogy. Cryptocurrency has some limited uses for interacting with an economy, specific edge cases most of the time. LLM AIs are incredibly general by their nature, they allow for large scale information processing in ways that either weren’t previously possible or were just very impractical.
There’s already been clear value derived from the use of it, in the medical and scientific fields if nothing else. That’s more than can be said of crypto. But yes in general I share the fears that big tech and government will use it to keep the poor poor, which is why I’ve been asking about what we should actually do about it. I don’t think it’s a hopeless situation.
Without the safe harbors, the only way for Internet companies to reduce their liability risk to a level that insurers and investors will accept is to stop accepting user-uploaded content at all. That would make the Internet look a lot like cable television, broadcasting, or traditional publishing: a bounded collection of creative material selected as much by financial clout and elite power relationships as by interest or artistic merit. Much of the Internet’s uniqueness and vitality would be lost.
In order for the safe harbors to safeguard public participation on the Internet, they need to be usable by a wide variety of Internet users, from the largest social media platforms to the humblest blogs and apps. A safe harbor that puts onerous requirements on intermediaries to filter, police, and remove alleged copyright infringements risks the same results as losing the safe harbor entirely.
The loudest voices calling to gut the safe harbors point to Google, Facebook, and a few other large Internet companies as the targets of a new law. Those massive platforms certainly dominate many people’s experience of the Internet today, and they collectively host and transmit billions of copyrighted works. And given the growing, well-deserved criticism of these companies’ privacy and content moderation practices, it’s tempting to support any legal changes that would further regulate their behavior. But copyright’s safe harbors don’t just protect Internet companies from liability—they protect all of us from being arbitrarily and unaccountably silenced online whenever our digital lives intersect with some bit of creative work to which a stranger has laid claim.
Besides, Google and Facebook are not the entirety of the Internet. We’ve yet to see a proposal that would successfully remove the safe harbors for those companies while reliably preserving it for others. Second- and third-tier Internet companies and even nonprofit projects like Wikipedia would be at risk. And while the search and social media titans may have the cash to implement massive copyright filters and the political clout to shape the law’s rough edges, small and mid-sized Internet companies and projects will be caught in an expensive bind. As we wrote recently, while America’s tech giants would prefer no regulation, they’ll happily settle for regulation that’s so expensive it clears the field of all possible competitors.
Copyright’s safe harbors are vital to preserving many of the Internet’s unique strengths, especially for smaller companies and individual users. This inescapable fact needs to be part of every discussion about how to regulate the Internet going forward.
I think this is really worth thinking about, are the existing notions of copyright and the surrounding laws going to remain relevant in a world where almost any software can be conjured up at will, for a relatively tiny cost and within a relatively small timespan? We’re basically entering into a post-scarcity world as far as software is concerned.
Nothing “neo” about it, this is just classical liberalism from the 1700s and basic private property rights. Neoliberalism is about things like privatization, deregulation, lowering taxes, and cutting public services, none of which I’m in favor of.