Last year, the Future of Life Institute organized the "AI Worldbuilding Contest", where they challenged competitors to envision a future where humanity has safely navigated the development of artificial general intelligence, avoiding major wars and catastrophes. The winning entries were imaginative and diverse; I profiled a few of my favorite entries here.
Now, FLI is following up on the project with "Imagine A World"! This new podcast series digs deeper into the eight winning entries, exploring the different ideas and solutions that each team used to build their distinct, aspirational futures.
Here's a little bit about each of the first four episodes; I'll cover the next four once they all come out!
"What if new governance mechanisms helped us coordinate?"
This is the first episode of FLI's series, and it's also my own team's entry! Here are some of my favorite moments from the podcast interview:
At the 11:15 mark, we introduce the overall historical arc of our scenario, then move on to talk about how new institutions like prediction markets and charter cities could improve humanity's ability to make wise decisions and coordinate on global problems.
At 27:12, talking about how a rapidly-changing landscape of AI capabilities will create a lot of uncertainty about who the ultimate "winners" vs "losers" will be. This uncertainty might actually make international coordination easier, creating a situation like the "veil of ignorance" thought experiment from philosopher John Rawls.
Around 38:30, Holly and I gush over some of sci-fi stories that we took inspiration from -- books like Ada Palmer's Too Like The Lightning, Ursula LeGuin's The Dispossessed, and Yudkowsky's Inadequate Equilibria, that take improvements in "social technology" just as seriously as advancements in science and engineering.
"What if narrow AI fractured our shared reality?"
(Apple Podcasts, Soundcloud, Spotify, and here is the original worldbuilding entry) This world, from Michael Vassar, brings a strong dose of Robin-Hanson-esque pessimism about human motivations and realpolitik about the structure of societies and institutions. In this world, cutting-edge AGI hasn't yet been rolled out to the world, instead being cautiously tested and tweaked for alignment. But meanwhile, despite advancing living standards, society continues to go off the rails in the sanity department -- nations' state capacity is eroding, society is more politically fractured, and individual people have a more confused view about what's really important.
At 7:25, Vassar talks about his conception of how the world might break apart into isolated "media bubbles" where AI-influenced media leads people to become less and less connected to a shared sense of reality. Plus a wacky analogy between AI and the movie "Who Framed Rodger Rabbit"!
At 17:36, we explore some of the social & economic details of Vassar's scenario, seeing how the world has gone off the rails in various ways. Ultimately, this pessimistic take on human institutions seems aimed at investigating the same question that I tried to grapple with in my own scenario -- how can our current level of civilization, which feels so inadequate in terms of its capability for wise, coordinated action, ever hope to deal with the challenge of aligning and safely deploying a superintelligent AI?
At 39:07, some interesting thoughts about optimism vs pessimism, and why people are often biased towards being irrationally pessimistic. This continues at 45:45, with discussion about a special kind of nihilism that Vassar sees emerging in today's social media culture.
"What if we had digital nations untethered to geography?"
This team is based in Kenya, and their scenario is a deep dive into the possibilities for virtual communities and network states! In their world, deep-learning-based AI techniques like large language models start to stall out in the near future; instead, AGI is ultimately achieved via "digital person" algorithms designed more closely to mimic the architecture of the human brain.
Here are my top highlights!
At 7:45, Tracy Kamande talks about how trying to imagine realistic futures -- and in the process, taking emerging technologies and other changes more seriously -- can help you live more intentionally, learning more and getting a more grounded view of what's going on in the world.
At 35:30, the concept of "digital nations", or "network states" as described in the book by Balajis Srinavasan. Interestingly, over the past year, the pacific-island nation of Tuvalu has taken steps towards becoming a true network state, which it hopes will help it maintain sovereignty and provide services to its citizens even as rising sea levels erode the physical integrity of its territory.
At 43:00, a description of how they envision the details of "digital people", which need to be raised almost like human children (reminding me of the Ted Chiang story "The Lifecycle of Software Objects"). I especially appreciate the thought they put into constraining the nature of "digital people" (like the idea that digital people might be intrinsically averse to recursive self-improvement, or might only be able to process information at some fixed maximum speed) in order to avoid scenarios that would lead to "runaway computation explosion". These constraints are needed for the story to work, since (like Holden Karnofsky outlines in this 80K interview) human-level AGI based on today's deep-learning architectures might still lead to explosive growth even if we never got to superintelligence.
"What if we designed and built AI in an inclusive way?"
This episode features Elaine Czech and Vanessa Hanschke, both academics at the University of Bristol. This team put a lot of thought into how new technologies might integrate into people's daily lives, and has a left-wing political perspective that emphasizes the importance of listening to marginalized voices in order to build a bright future for everyone. Highlights from this interview:
Around 9:18 -- In this scenario the most advanced AGI models are large, costly, and unwieldy, so interaction mostly takes place in defined settings with an "oracle" consulting-like atmosphere. True AGIs are heavily regulated so they are not commonly interacted with by the public.
At 20:39 -- This world features technological unemployment where most manual labor jobs disappear. Despite this, human nature is slow to change. One imaginative way that society addresses this is by creating a virtual high-stakes war/competition space so countries have an outlet to continue fighting.
Future Episodes
I've only caught up to about halfway through the series, but they've currently released seven out of eight total episodes! You can subscribe to the Future of Life Institute podcast here -- besides these "Imagine A World" interviews, they also host talks with all sorts of fascinating intellectual figures on new ideas in philosophy, AI alignment, and other important topics.
Finally, the Future of Life Institute would like to note that they're not endorsing any one idea with this AI Worldbuilding Contest. Rather, they hope to grow the conversation about what futures people get excited about.
(Personally, I'm not an FLI employee or anything, and I would like to note that I'm totally endorsing one idea, namely my own! As I tried to showcase in my contest entry, I think that experimenting with improved institution designs like futarchy, network states, liquid democracy, Georgism, quadratic funding, etc, is an underrated approach within EA and has a lot of potential to create a safer, wiser, more flourishing civilization. But that's a topic for another post!)