Korva is a scroll of the internet’s best ideas.
Humans were society's edge compute
Humans feeding society
Each person started by learning pieces of what society already knew, which is education. Then they walked around, saw slightly different things, chewed on a tiny bit of a bigger problem, and mixed that unique experience with what they had learned. The result was slightly novel. They reported it back through scientific papers, PowerPoint presentations, office conversations, and memos, and society reassimilated it and grew a little wiser.
AI as the new center
AI works as a new world-mind of the same kind. It takes the accumulated findings of individual humans, robots, and other sensors and integrates them into a general picture of the world. Each time someone reports a new scientific finding or a new perspective, AI's understanding grows, so every scientist and businessperson is writing for the AIs now, whether they intend to or not.
Before the 2008 crash, most financial derivatives were traded over the counter, with people calling other people and selling them products. After Dodd-Frank, most derivatives trading moved to central clearinghouses. The shift from human organizations to AI as the coordinating intelligence resembles that move from over-the-counter exchange to a clearinghouse.
The bottleneck disappears
AI's grasp of complex ideas is not limited by what one human brain can take in or communicate, so the information bottleneck is gone, or will be soon. For now, AI still needs people as edge compute to go out, discover new facts, and bring back new perspectives.
Why process data near where it is collected?
Sending every raw measurement to a distant data center uses bandwidth and adds delay. A device that filters or summarizes data locally can send only what matters and can react at once. A car's braking system, for instance, cannot wait on a round trip over the internet. Light in optical fiber travels about 200 kilometers per millisecond, so distance alone sets a floor on that delay.
How did Dodd-Frank move derivatives to clearinghouses?
Congress passed the Dodd-Frank Act in 2010, after the 2008 financial crisis showed how losses could spread through private contracts that no one could see in full. Its Title VII required many standardized swaps to be cleared through central counterparties. A clearinghouse becomes the buyer to every seller and the seller to every buyer, so each trader faces one well-capitalized party instead of a web of individual counterparties. Clearinghouses also collect margin from their members every day to cover potential losses. President Obama signed the Act in July 2010.
How does published research reach AI systems?
Large language models are trained on very large collections of text, and open repositories of research papers are one possible source of that text. The arXiv preprint server hosts millions of papers in physics, mathematics, and computer science, all free to read. Papers there can be downloaded in bulk by machines as well as by people. Physicist Paul Ginsparg started arXiv in August 1991 at Los Alamos as an email server.
More from Noah Smith
