Funding AI Public Goods and Keeping Them Public - Berkeley 2024

[Music] we've been talking about a lot of other public goods today and uh I think the the AI the AI influence on a wide range of goods is an amplifying one so uh I'm delighted to have both of you here to chat a bit about what it means to support AI public goods and how we might think about the what's required to maintain them and to keep things that might feel public now available in the public interest in this context public ey means Public Access some degree of public support and some degree of public accountability and in particular uh in the w3c and the ietf standard it has to be accessible and publicly funded with some accountability it should be maintained in the public interest but that's hard and that's what that's what we're talking about and it may be Prov provisioned and developed in a range of ways it could be publicly inspectable it could have free and open components there are groups that are building public interest tools that think that that's not possible for them they're still supporting the public interest how do you think about the where we are in the State of the State of the knowledge and and and public comments today I I think that the the when being going way back in the internet and being involved in artificial intelligence was still in the steam age and getting my doctorate in that area and seeing what's evolved over a period of time as we move into an area that that where you're getting Ai and you're now getting the appropriation and collection of data from Wikipedia and other sources and it's being was what it's a commment is being captured and appropriated into proprietary models there's a huge challenge there and in my own View and I and I take this from front my experience and being interested in burkman doing the issues around how do you get people to have control their identity and data and recognizing that that is something that should not be appropriated that it is something know it's a public good it's a private good as well it's something that the people should have how do you how do you prevent that capture and I think a lot of the regulatory thinking and and the problem is is that they they think of governance as an externality we have to think of governance as being implicit into the technology it's not an externality I think that's the key issue I think the other key issue is there has to be Financial incentives uh for protecting that good it can't it can't just be something that is a donation an afterthought you you sort of you you create all these negative externalities and then you say oh let's let's make up for the mess and and and put some public funds into it it really has to be viable in its own terms and so you need to have some kind of economic model that actually rewards the positive function of being able to create a positive social good and I think that that is really sort of the key element this is where I I think that the next generation tokenomics are to come out they're going to be could be very significant and how do you create not just a common a governance of a common but a a a means by which it can sustain itself and resist the predation of of of external Capital it would be a different investment in a desert ecosystem than in a rainforest ecosystem so I think analogously for different kinds of cognitive ecosystems it might be a different rate limiting step or different Horizons that come into play uh a major piece that transposes across and also Within These information ecosystems would be like the reliability of the information and the information supply chain knowing that there doesn't have to be like an extra Loop or cycle whether it's a a person with the consequences on their time and well-being or whether it's like an AI agent or however we want to think about that taking like extra Cycles to consider information how that can be grounded and then not just tracing that back to kind of the seal but then to go even further past onto basically The Meta governance and the distributed standards and the ways that within and across the organizations there are norms these are all some starting places who is the public for whom these shared goods are for if you imagine an ecosystem that inv has a lot of AI agents what are the kinds of public goods that they would need in order to maintain a healthy ecosystem but how do you take some of those ideas that were sort ofly pre AI but what what's the concept of being able to uh what are the means of oversite that makes sure that the homeostasis of the G the commons stays in place and I think that you have with this notion of the codification of the scientific method and they call free energy principle you have something as scale free and it's domain free that is an independent metric that you can allocate to being able to see whether a set of outcomes are are consistent or have a better greater predictability or have a certain uh level of risk with them so you you're you're a subject what you're trying to do is ground a community in a in a measure of reality regarding their own goal of keeping their community alive it's it's like what does it mean what do we have to do to keep keep this Community is Biore region alive what what what are the what are their inter dependencies there's certain Goods they have to achieve there's certain kinds of efficiencies there's certain kinds of metabolic closures they have to take as a whole in order to make that valid and so you can start to measure that and that becomes a principle of the governance to the extent to which how close or how far you're away from that and they can come and revise that they can reset that in their ways of participation and allocating rewards based on that but it I I think this is this is it really it's not an externality it's it the governance is something it's intrinsic into what it is for that living thing to stay alive a little bit at that okay so the question who is the public and as publics and Commons and spaces include not just like a broader range of humans and non-humans what what are we going to be doing so a few pieces on that one would be recognizing and making explicit and legible multiple time scales so here's the high frequency time scale here's a slower Rhythm and nested rhythms that allow for like slower stewardships as well and then as to interactions even amongst humans with a similar cultural background and a similar spoken language still there's a need to have Clarity on shared semantics that becomes especially important as there's more variability and uncertainty all these things so being able to coordinate on systems that have a narrative or ontology different ways to think about what that reference frame is but enabling human accessible and computationally usable reference frames is definitely critical otherwise there's always going to be this lost and translation Gap going both ways um and then drawing from active inference a community at multiple scales being able to say what is of pragmatic value and what is of epistemic value what is of utility what is going to align the world with what we want to see what do we actually prefer and then what would learning look like and what are we willing to learn we're at time but it's an important conversation to continue uh we're going to be here outside after the talk and I know going [Music]