Charlie Kennel and Giovanni Paternostro have spoken with David Victor. David is a professor of innovation and public policy at the School of Global Policy and Strategy at UC San Diego, where he holds the Center for Global Transformation Endowed Chair in Innovation and Public Policy. David is also co-director of the campus-wide Deep Decarbonization Initiative, and he is a professor (adjunct) in Climate, Atmospheric Science & Physical Oceanography at the Scripps Institution of Oceanography.
Dear David,
What could be achieved if there was a public or nonprofit AI effort with the same scale and level of funding as the current large private efforts? What would be the benefits for society?
David: The line between what's public and what's private is very fluid these days. There's a lot of for-profit AI that is burning a massive amount of capital, and that is not yet achieving a profit. It could be enormously profitable, or it could be that, incidentally, some of the biggest impacts of AI will be on the public sector.
For example, there's a lot of discussion these days about how you could you use AI to improve health care and to improve education, including universal basic education. All of these are things where innovations in AI that might be driven by private capital in Western countries could end up spilling over and having a much bigger impact in the rest of the world. I am doing some work on those kinds of questions with my UCSD colleagues Elizabeth Lyons and David Danks.
In my day job, I mostly work on energy and climate. AI could have a really big impact on energy and climate. An example is in the data centers needs for electricity, which are significant. When the high market cap tech companies go out and procure power for their data centers, they mostly want to procure green power in some form or another, which is not just renewables, but it is increasingly nuclear power and geothermal. I don't know if it's public or private, but they, for public mission purposes, and for brand concerns, are already jumpstarting a significant investment in clean energy, clean electric power in particular.
One of the most important questions about the future of the energy system is how do we electrify as much as possible, and how do we make sure that the electric power system is highly reliable? And that is a task that could be made radically easier by highly sophisticated AI systems. That's because electric grids that include a lot of renewable power face a big challenge of managing that grid for stability. Electric grids that have a lot of decentralized power generators, I think rooftop solar or small geothermal units or batteries, represent a decentralized control problem where the current strategies of controlling grids, where you centralize all information and decisions, are not going to work. If you had an adaptive, self-healing, decentralized grid control system, that could be very powerful.
Some of these challenges looks amenable to improvement through AI. So those are the kinds of things where AI could be very transformative in terms of the demand for clean energy, but also in terms of the way we organize the energy system.
We are talking to scientists in different fields. Some biologists and biomedical scientists told to us about the complexity of modeling the cell or the human body. One of their main concerns is that the lack of transparency for some AI tools which are needed for these tasks might reduce trust in science. What about planetary models, including both geophysics and ecosystem biology? You mentioned the potential benefits of AI for climate modeling in one of your recent papers (1). Maybe the increasing amount of new data available and AI could together help us to have better planetary models?
David: One of the things we've learned about AI is that the line between raw data and processed data it's very fuzzy. There might not be raw data in the future because everything is processed in some form or another and the idea of raw data is an anachronism. And I think that's important for us to keep in mind here.
The challenge in climate science is not unlike the challenge in biology. There are applications where we're seeing big advances in biology from AI, in understanding, for example, the folding of proteins and the behavior of viruses and vaccines. That's a combination of data with the processing and use of the data. And it's quite analogous to what we face in climate.
There's a lot of data coming in. Charlie and I have for more than a decade worked on how to improve the kinds of data being collected. If it were possible to understand, at a very granular level and in a high-speed way, which data sets were most valuable under what circumstances, it would improve the way we understand the potential physical impacts of climate change, just like it has improved the way that people do weather forecasting.
It is also very important to understand the limits of what might realistically be achieved.
We have another question.
We are encouraging researchers at different career stages to share ideas about complex science problems that could benefit from a large-scale AI effort. We found that motivation and recognition could be provided if you and other well-known scientists were willing to talk to people that suggest the best ideas. You would be the judge and decide if any idea is for you deserving of attention. Any scientist selected might receive advice but could also be a potential collaborator. Many ideas will be produced, and society will take notice. Would you be willing to talk to any of these scientists?
David: If there's something I can do to be helpful, I'd be happy to. I have thought about and worked on the importance of scientific collaborations.
When I was a graduate student, the Cold War was underway, and one of the first topics I worked on as a student, and later leading a research group, was to provide systems analysis, which was set up during the Cold War as a place for American scientists and Soviet scientists to work together on things that weren't political.
And today, I'm spending a lot of my time on the analogous challenge of how we keep the United States and the Chinese scientific communities connected to each other. There is increasing suspicion, and the consequences for the productivity of the scientific community and, frankly, the consequences for our two countries if we don't have scientists who know each other and can deal with each other when things get bad, are mostly adverse. We've been demonstrating for a long time the value of collaboration, notably across the United States and Chinese scientific communities, but we have not always been able to convince politicians.
Another point relevant for our reflections about spill overs from current AI efforts is demonstrated by Sal Kahn, the founder of the Kahn Academy, a non-profit that is well-known in the field of education. Sal has a new book out making the argument that AI makes it possible to have a universal basic education that is not just watching some YouTube videos, but an interactive, Socratic kind of system, where the AI asks you some questions. You answer those questions, and the subsequent questions are adjusted based on what you say. The AI helps educate you the way the Oxford or Cambridge method of tutors would work. But it's infinitely scalable, whereas the Oxford or Cambridge method is not infinitely scalable because there aren't enough tutors out there for the entire planet. This is an example of spill over to the public benefit from the for-profit work on LLMs.
There are a few comments we received that are related to your last point. A graduate student suggested that AI might train researchers to understand how best to use it in their specific area of interest. Other scientists have highlighted the role of AI as a provider of inspiration and a source of questions, rather than as a place where to find responses. The Socratic role of AI could help scientists at any level.
Thanks a lot for finding the time for talking to us.
David: This is a really interesting set of topics and conversations, so I'd be happy to be helpful.
REFERENCES
1 - David G. Victor "How artificial intelligence will affect the future of energy and climate"
Brookings, 2019.