- Making AI strategy into a well-functioning machine for processing information, generating action
- Analogy: brain, active inference, embodied cognition
- The Apollo problem: spreading information about risk
- How to design a better democracy (Weyl, Hanson, Caplan)
- Information flows on an international level, trust
- Flows between past, present and future self (Forte)
Reminds me of
- Zipf distribution
- Erlang distribution
- Pareto distribution
- Ben Golub
- Apollo, Mueller boxes
- Wiener – society as a Cybernetic organism
From Ideas to Action
Our traditional view of society was shaped in the era of the Enlightenment. Today’s societies differ. There are weaker ties and information flows are faster. Abstractions like Rational agents, classes and markets therefore don’t work as well.
Social Physics looks then stop at patterns of information close between people. It’s can be used to predict the productivity of groups.
Thought: This would be really useful for structuring the ecosystem of EA/AI Strategy
Example: You can apply the methods of social physics to measure the flows of information between traders. Pentland’s study found that the social influence of a few traders was too strong. It led to social herding, traders overreacted to each other.
Thought: Direct application to the x-risk space. Is the social influence of Paul Christiano or Eliezer optimal? If our machine is set up in a way that won’t reliably produce good outcomes, fix the machine.
Thought: Overreacting is almost certainly happening in EA. Traders have much better feedback from the real world. In its absence, I would expect people to resort more to social cues.
Confusion: how would you go about measuring?
Confusion: to what extent can you actually restructure groups to become better machines?
A Practical Science
(Does he mean engineering? Social engineering does not sound that great)
The origin of the term is from 18th century (who?). Pentland shares Marr’s ambition of a true behavioral science rooted in evolution and quantitative approaches.
The test is whether we can build practical systems with this framework.
Ambition is to translate into heuristics usable by non-mathematicians.
But then one broker agreed to let my MIT research laboratory try a social physics approach, drawing on our mathematical models of how ideas spread through social networks. By analyzing the millions of detailed messages among traders on a social network,
we discovered that the effects of social influence within the network were too strong,
causing the phenomenon of herding, in which the traders overreacted to each other, and so all tended to adopt the same trading strategy.
The mathematics of social physics indicated that the best approach to fixing this problem was to change the social network, in order to slow down the spread of new strategies within it. When we implemented these changes, it doubled the average return on investment,
leaving the standard economic approaches in the dust.
Slowing down the spread of ideas is not something that is found in a standard management handbook. And this result was no accident, because we had mathematical analyses based on millions of bits of data that made it possible for us to devise precise interventions and predict precisely what the outcome would be. Those equations are part of the mathematics of social physics, as I will begin to explain in Chapter 2.