Notes on 2040 hedge fund AI's inside MPC circuits
I think that autonomous AI’s will develop primarily as a matter of comparative advantage but applied to intelligence.
They will sit on problems for days, weeks, and months in “system 2” mode (deepthink). Unlike humans, who require NDA’s to prevent leakage of proprietary information, these corporate AI’s can be run as pure algorithms inside encrypted secure multi-party computation circuits.
Just as hedge funds develop highly-sophisticated trading strategies which make markets more efficient, thus producing more accurate signals for the market supply-demand process, so will there be firms developing highly-sophisticated AI models which will be used by firms to improve their economy too.
AI over MPC is fascinating to me because it allows parties to not only private compute over information, but privately ideate over information.
Two hedge funds could trade alpha using MPC, but they would not be able to ideate on what they’ve learnt without encoding it in an algorithm. Using AI they plausibly can.
I’m excited how this applies to biology. The human genome and DNA sequence is the equivalent of 10,000 years of evolutionary tech debt. Perhaps by collectively analysing all our genomes, we will be able to separate enough noise from the signal. With enough compute, we might discover the model which deterministically predicts how people evolve.