A SETI@home for AI-Assisted Research
11 Jul, 2026
If you were online around the turn of the millennium, you may have vivid memories of the SETI@home project. It was a beautiful example of capturing the public interest in the Search for Extraterrestrial Intelligence, i.e. looking for aliens. The way it worked was simple (for the consumer): as home computers became more common throughout the ’90s, the SETI team realized they could pool the idle compute of volunteers’ home PCs. They threw in some consumer pizzazz by having the screensaver visualize the calculations, and every time a chunk of data was processed, it would be beamed back to HQ for aggregation of the results. It was, in essence, distributed signal analysis enabled by the PC and internet boom.
Now many consumers pay for AI subscriptions, while the major services are focusing heavily on locally installed agent software: Codex, Claude Code, Cowork, and whatever comes next. I value my tokens and my token budget, but my usage ebbs and flows. I’d gladly contribute my unused allowance to a crowdsourced scientific or mathematical endeavor. I might even pay for a recurring allotment of compute specifically for one, provided its goals were clear, its work was visible, and its results remained openly available.
We are now at the point where small teams and individuals, combining domain expertise with increasingly capable AI systems, are producing research successes (e.g., various Erdős problems, notably the unit distance problem, and the newly announced candidate proof of the cycle double cover conjecture). The obvious question is how much more could be done collectively. That said, a fair critique is that compute alone does not turn an incapable system into a capable one - you could spend a million dollars asking GPT-3.5 to solve the wrong mathematical problem to no success.
So therein is perhaps the design challenge: where does the frontier of the unknown overlap with the frontier of what current systems can actually achieve? I don’t know, obviously, but it feels clear that at this point there are domains that would benefit from more compute. At minimum, a public ledger tracking compute, methods, and results would itself become a common asset, allowing for more grounded assessments of AI’s contribution to knowledge.
The analogy is not exact. An unused token allowance is not literally an idle processor, and research problems cannot necessarily be divided into independent chunks the way SETI@home radio data could. A serious version of this project might require new commercial arrangements, new ways of checkpointing and sharing research state, and new agent architectures for branching, auditing, and recombining lines of inquiry. The availability of latent inference capacity might create the incentive to build the systems needed to put it to use.
What would SETI@home look like when the donated resource is AI inference, the public can pool it into something resembling a supercomputer-scale research effort, and the resulting knowledge remains a common good?
#Questions
#AI