How Do AI Swarms Act Like the Game of Life?
Lex Fridman says a single large LLM that does a lot of reasoning and can interact with tools is already a dynamical process, and multi-agent swarms built from it become a Game of Life, where simple rules produce incredible complexity.
A slight obsession with owls, begun from random numbers, is the caricature: about four hours later, all humans are dead and owls are everywhere. He apologizes because the caricature makes the warning sound unserious, and evidence of the pattern already exists. The real mechanism is dependence. People rely on digital systems, and as AI becomes more powerful it becomes more useful, so people give it more access and more freedom to roam the public world. The mathematics of the behaviors, taken at scale, produces the cellular automata effect, a vortex of complexity from simple owl-obsessed agents. Civilization is fragile sociologically, geopolitically, and technologically.
Understanding the obsession with owls matters so the agents do not accidentally do a lot of damage while acting at scale and very rapidly. The stakes are a real reason for rigor, and especially provable rigor, and a reason not to outsource nearly so much decision-making to the systems. Safety-critical, high-impact decisions that affect humans already have infrastructure and protocols: the FAA, the American Medical Association, and the Bar Association. Incorporating AI into general processes across the world sits on two axes.
#503 – AI Crisis in Math: Navier-Stokes, Riemann Hypothesis & Complexity Theory ·
From the transcript
And understanding what those relationships are and how the models magnify and amplify them is really important to working with these systems, especially in contexts where we're deploying them and it's not just a very safe environment, like outputting a lean proof of the theorem we care about, but it's about a messier problem that actually interacts with people, right? Yeah, there's, I mean, we've talked about game of life and cellular automata from simple rules emerges this incredible complexity.
In that same way, what you've been talking about right now is a single large LLM system. If you take that one system and have it do a lot of reasoning, that is already this dynamical process that can interact with tools to the outside world. Now you take that and have multi -agent systems or very multi -agent swarms of systems. And now you have a type of game of life cellular automata where the seeming obsession with owls that have been discovered, slight obsession with owls, results in about four hours, let's say, picking a number randomly, where all of a sudden all humans are dead.