Last week I gave a guest lecture at an Oxford seminar on AI and Philosophy.

I had never been to Oxford before. What an extraordinary experience!

The title of my talk was Surfing the Silver Swarm.

It wasn't until I pulled the talk together that I realized that most of the bottom-up strategies I apply are intuitively about orchestrating the energy of a swarm and then surfing it.

The talk covers the asymmetric power of swarms, why they are so hard to control, and why if you let go of the need for control you can surf them to miraculous and surprisingly coherent outcomes.

Swarms are like flood-fills of possibility space, a search algorithm that scales linearly with investment.

With a single individual plan, the effectiveness typically scales sub-linearly.

The overall motion of the swarm is the sum of all of the motion vectors of its sub-entities.

By default, a random swarm has brownian motion, which means the net vector of the swarm is nothing.

But if you can induce a small asymmetry consistently across each agent, you can get significant movement of the overall swarm.

Interestingly, each individual agent in the swarm will (correctly!) report only a very small influence on them by that asymmetry.

But when you average all of the agents together, that small but consistent asymmetry adds up to something that strongly stands out from the noise.

It's also possible to use a kind of macro-scale Maxwell's Demon effect.

Pick the subset of entities in the swarm that are already going in the direction you want to go in, and select them into a subset.

The subset will now have a significant amount of velocity in roughly the direction you want to go in.

Now, the normal gravity-well dynamics of a successful swarm kicks off, and the energy becomes self-accelerating.

Others nearby who are mostly aligned will choose to join in, upping the momentum.

You're effectively creating free energy by just applying some discerning judgment on what to allow in!