If we knew we’d never get flying cars, most people wouldn’t care. What do we care about?
I find inspiring José Luis Ricón’s vision of cozy futurism:
Cool-scifi-futurism seems almost synonymous with technical advances and feats of engineering: Artificial General Intelligence, nanofabrication, Big Objects In Space, and so forth.
In contrast, cozy futurism starts not with technology but with current problems and human needs. You could imagine societies where poverty is absent, housing is affordable, cities are architecturally pleasing, economies are environmentally sustainable, and all disease is cured. Then you work backwards from there to the technologies, cultural shifts or policy changes needed to get there.
Cozy futurism is not necessarily less ambitious than cool sci-fi futurism: by the time we get to Mars there will still be homeless living among developed countries back on Earth. Nor is cozy futurism just about institutional or cultural solutions. Fusion reactors and anti-aging therapies are key enablers of a cozy future. With cozy lenses on, when envisioning that future you think of what technical advances enable, how they improve our lives; rather than the advances themselves.1
Technological progress and a future we’d like are not zero sum, or in conflict at all. Cozy futurism doesn’t mean “stop advancing sci/tech,” it means “think about what those are for.” It means prioritizing advances that make our lives and societies and cultures better.
Is Mooglebook AI making them better? Will AI text generators make them better? Will superintelligence make them better?
AI will probably not be a wish-fulfilling genie that solves all problems. What sort of AI do we want, and why? Concretely, how would it help, in a world we’d like?
Since the 1960s, whenever skeptics ask that, AI enthusiasts say “healthcare!” So far, AI is perpetually “about to” revolutionize medicine, but has delivered very little.2 The specific longer-term future healthcare applications typically gestured at also do not depend on the sorts of AI currently under development. Specifically what sort of AI would help, how? All-purpose AI as a wish-fulfilling genie is not a likely scenario. All-purpose AI is more likely Doom.
- 1.Paraphrasing “Cozy futurism,” Nintil, 2021. I have edited this extract fairly heavily, for relevance and concision.
- 2.Dan Elton’s “AI for medicine is overhyped” (More is Different, Mar 29, 2022) is a sobering debunking by an expert in the field. He explains that the main difficulties, with consequent real-world failures, are due to statistical distributions being quite different in the deployment environment than during training. That is a fundamental, inherent limitation of current AI systems. Visar Berisha and Julie Liss reach similar conclusions in “AI in Medicine Is Overhyped,” Scientific American, October 19, 2022.