If You Wish to Make an Apple Pie From Scratch...

We are going to think about what civilizations of superintelligent physicists might look like. This is going to be difficult. It is not easy to even know where to start.

But it should be an amazing journey.

(I doubt they read Sidney Coleman, drink coffee, and scribble equations in whatever passes for their local coffee shop, but who knows).

So where do we start?

Nearly all of my professional life has been spent trying to build things that should exist, but don't. I've been fortunate to have been in the vanguard of several revolutions in technology, including quantum computing, early LLMs, embodied AI, humanoid robots, and superintelligent agents. Some of these are well established now, and some of them are still science fiction. In the case of quantum computing, today there are dozens of public companies in the space. But when we started D-Wave in 1999, it wasn't even clear that nature allowed even a single macroscopic qubit to exist.

Living on the edge between leading global technology revolutions and being viewed as a half-baked wingnut really affected how I think about science and technology. One of my strongest opinions about this is that to stack the odds a project will end up in the 'revolution' category, progress needs to be based on building actual real things that exist and watching how they really behave, not just ideas and words. This might seem obvious. But it is remarkable how many people working in the difficult subjects we will be traversing end up totally off the rails because they ignore the fact that reality exists and is the ultimate arbiter of what is possible to build.

So for this project, I am going to adopt this philosophy.

Speculative fiction has its place for thinking about where technology might be taking us (Greg Egan has multiple stories of relevance here, such as Wang's Carpets, Permutation City and Schild’s Ladder). But it is easy to become untethered in a topic like this. We should be excited, optimistic, and open minded. But if we are going to contribute anything worthwhile, we need guidance from experiments.

How in the world can we get experimental guidance for something so outrageous as civilizations of superintelligent physicists? We have trouble even communicating with and understanding other people, let alone some potential superintelligent civilization!

Given our commitment to building and observing real things, I can think of two possibilities.

The first is working to establish contact with actual real-life superintelligent alien civilizations. We have lots of examples of what this might look like in science fiction. It is possible that such civilizations exist – in fact it seems overwhelmingly likely that they do. But I am not going to consider this category of thing. We could think about building ET detectors to look for them – maybe they are already somehow here, but we lack the knowledge to find that out and communicate with them. But I don't find this direction very compelling. I think the reason for this is that personally I have no idea how you'd actually do this, and the outcome of it if you were successful is completely out of our control. Maybe they would just eat us.

The second thing we could do is to start building these civilizations ourselves and observing how they behave. This seems like a much better approach. Not only is there an obvious starting point (we'll see this shortly), but it is as controlled as something like this can be. Doing it this way might even help us understand how not to get eaten if the previous scenario were to come to pass.

In video games, everything in them is there for our benefit. The characters are meant to be controlled and interacted with from outside their world, by us.

But what if instead we built worlds meant to be experienced from the inside, by the characters in these worlds, from their own first-person perspective? We could populate such worlds with millions of creatures, who would effectively live in a Matrix-style simulation of our making.

Could such creatures, inhabiting the inside of a simulation and running modern learning algorithms, learn, all on their own with no interference from us, to develop economies, languages, governments, coalitions, technologies, and science?

At this point, no one knows. But it doesn't seem completely crazy that something like this could happen. The idea of learning from experience is the foundation of the field of reinforcement learning, which has been extremely successful when applied to learning specific things, like how to play games well.

This will be our starting point. It isn't guaranteed to produce superintelligent civilizations, but it is in my view an excellent place to start.

We are going to create simulated worlds, inside of which are agents that can (hopefully) think, learn, and evolve. If we set things up just right, I think we can study how science itself might arise. What is necessary for a civilization to discover science? We don't currently know, either in simulation or in the real world.

In the posts to come, there will be much more detail about what simulated pocket universes and agents actually are, as we dive into doing real work.

We will think hard about what these pocket universes should be like, and will watch the agents in them closely.

These agents will make up our simulated civilizations, and will be our candidate superphysicists.