How I see things

Freshman year of college I became very interested in the mathematics of intelligent systems in hopes that it might have something insightful to say about AI alignment. The belief was that if we have a principled way to talk about all forms of intelligence, whether that be biological, artificial, etc., we would better understand how to align AIs and how they fit into the multi-polar complex system that is society. I was drawn to the idea that these kinds of systems can be described in terms of concepts like energy landscapes, entropy, phase transitions, etc. I am still of the opinion that this is largely the right line of reasoning. For example, there is a growing body of work establishing the connection between statistical mechanics and machine learning.

But statistical mechanics alone is not the full story. One core tenet I have is that all forms of intelligence, without exception, are collective intelligence. One can always decompose an intelligent system into smaller computational parts sending and receiving information to and from one another. This is true of everything from layers in a neural network to ants in a colony, and is largely in line with the ideas of Levin, Fields, and Friston. To me, this idea is most eloquently phrased by Jean Petitot in Cognitive Morphodynamics,

"According to [the connectionist] perspective, entities possessing a semantics are, on the "micro" subsymbolic level, global and complex patterns of activation of elementary local units mutually interconnected and computing in parallel. Their semantics is an emergent holistic property. The discrete and sequential symbolic structures of the "macro" symbolic level (symbols, expressions, rules, inferences, etc.) are qualitative, stable and invariant structures, emerging from the subsymbolic level through a cooperative process of aggregation. Here again, there is a key analogy with phase transitions. If we now introduce the Lyapunov functions of the attractors [...] we are naturally brought back to Thom's morphodynamical models."

That is to say, embodied computational processes are deeply related to the global structure of its underlying dynamical system. And what constitutes a computational process is incredibly general; everything from neurons minimizing free energy to a company deciding what action to take. This might sound like a tautology, and to some extent I can't even argue with that assertion. At the same time, this perspective lends itself to both the formal structure of mathematics and extreme generality, which I find very attractive. The more unified one's set of ontologies are, the more free they are to analogize and abstract; the easier it becomes to identify common properties that are useful for solving problems.

I believe complex systems theory is a very useful way of looking at things. Everything is a complex system. Everything from the international web of geopolitical relationships down to the cells in your body. Insofar as we can axiomatize a theory of intelligent systems, I am confident it will have incredible consequences for how we think about the world. I want to take this understanding and apply it to steer the world towards more desirable attractor states during this societal phase transition that is the development of AI. To me, this is not just a technical problem, but a social one as well. As cliché as it sounds, we do in fact live in a society. So long as humans are in charge, one's ability to enact change beyond themselves comes down to working with others. As such, I am ultimately interested in leveraging this set of useful ideas in whatever manner is most effective, whether that be technical alignment research or running an AI safety organization.