There's a lot going on this year but to me, the most exciting things are all happening in the lab. Last year, we started a project to figure out how cognitive effort works - our lab was far from alone in this project, so there was a ton of really interesting reading to do. Clay Holroyd's Controllosphere concept is essentially the answer to the question of what the experience of effort in a cognitive context involves. With this insight, and with Hugo Fleming's effortful but not difficult number switching task, I was able to create an experimental sandbox that we'll be using next year to drill deeply into the subject and test specific predictions.

Web3 continued to be a deeply troubled nascent technology field. I'm not a bitcoiner and never really was, but seeing Ethereum fail to outrally bitcoin in general has been a bit of a wake-up call: 2021 isn't coming back, and a mature crypto market is going to be a lot different from the standpoint of trading tokens. I'll be sure to let you know if I find another trading angle that consistently works, but to me it's looking more and more like you're going to need to be the developer of the successful token if you're really intent upon making money from the space. For now, I think I'm okay just having a little bit of fun and watching. I have some (non-public) builds in the works that are pretty exciting and I suppose there's a sustainable level of hope there.

There's some really remarkable stuff around the Free Energy Principle - for example, the Good Regulator Theorem postulates that a controller must contain a model of the system being controlled; but if you look at an individual body as it minimizes free energy to remain a distinguishable system (continues living, in other words), you'll find that in all cases for the free energy to be effectively minimized, the environment must be controlled. In other words, it's a logical requirement of life that the world external to the living thing be modeled inside of that thing.

This is all interesting from a cognition standpoint, but it's also very interesting from a Web3 standpoint - in creating decentralized networks that permissionlessly move data around with varying degrees of privacy, we're building extraordinarily complex models of the world. And the financial side, the currency side, of what's getting built out, is the part that carries the most information. The value that everyone has on-network needs to be safeguarded, and this common goal shapes the behavior of people on these networks.

In individual bodies, the interesting place to push these days has to do with dynamics that can be accessed by modeling Markov blankets as they relate to behavior. The Markov blanket is a self-maintained boundary between a living organism and its environment - they work for whole bodies, as in mine or yours, and also for parts of them - cells or organs inside us. It strikes me as very interesting to see that dynamics onchain take up certain properties we see at all levels of scale in cognition in individual bodies - as dynamics emerge at higher levels of cognition, what new patterns might we observe?

For a Markov blanket to hold, internal states and external states need to maintain conditional independence from one another. In a body, this just means keeping temperature within the healthy range and maintaining metabolism, avoiding predators, etc. In the onchain environment, there is a security problem that could be interpreted as a Markov blanket in the sense that the blockchain's rules prevent users from simply writing whatever they want to the onchain ledger. The internal states (onchain record) are conditionally independent from events in the environment because they can only be updated if a user follows specific rules.

Now, the chain's environment also includes its users, each of whom is actively modeling the chain as well, yielding a massively complex recursive thought process.

I'm not going to pretend to understand that part yet, but 2025 did lead to a lot of progress toward a model that may one day be able to help us understand cognitive behavior at new levels of scale.

If the next year is a good one, perhaps we'll be modeling cognitive processes in new ways by the end of it. I definitely have at least 2 experiments getting ready to run. And there's a new robust core of theory underneath them that, if it stands up, will probably be a great place to build a career. The going has been tough but the summit is in sight.

There's a lot going on this year but to me, the most exciting things are all happening in the lab. Last year, we started a project to figure out how cognitive effort works - our lab was far from alone in this project, so there was a ton of really interesting reading to do. Clay Holroyd's Controllosphere concept is essentially the answer to the question of what the experience of effort in a cognitive context involves. With this insight, and with Hugo Fleming's effortful but not difficult number switching task, I was able to create an experimental sandbox that we'll be using next year to drill deeply into the subject and test specific predictions.

Web3 continued to be a deeply troubled nascent technology field. I'm not a bitcoiner and never really was, but seeing Ethereum fail to outrally bitcoin in general has been a bit of a wake-up call: 2021 isn't coming back, and a mature crypto market is going to be a lot different from the standpoint of trading tokens. I'll be sure to let you know if I find another trading angle that consistently works, but to me it's looking more and more like you're going to need to be the developer of the successful token if you're really intent upon making money from the space. For now, I think I'm okay just having a little bit of fun and watching. I have some (non-public) builds in the works that are pretty exciting and I suppose there's a sustainable level of hope there.

There's some really remarkable stuff around the Free Energy Principle - for example, the Good Regulator Theorem postulates that a controller must contain a model of the system being controlled; but if you look at an individual body as it minimizes free energy to remain a distinguishable system (continues living, in other words), you'll find that in all cases for the free energy to be effectively minimized, the environment must be controlled. In other words, it's a logical requirement of life that the world external to the living thing be modeled inside of that thing.

This is all interesting from a cognition standpoint, but it's also very interesting from a Web3 standpoint - in creating decentralized networks that permissionlessly move data around with varying degrees of privacy, we're building extraordinarily complex models of the world. And the financial side, the currency side, of what's getting built out, is the part that carries the most information. The value that everyone has on-network needs to be safeguarded, and this common goal shapes the behavior of people on these networks.

In individual bodies, the interesting place to push these days has to do with dynamics that can be accessed by modeling Markov blankets as they relate to behavior. The Markov blanket is a self-maintained boundary between a living organism and its environment - they work for whole bodies, as in mine or yours, and also for parts of them - cells or organs inside us. It strikes me as very interesting to see that dynamics onchain take up certain properties we see at all levels of scale in cognition in individual bodies - as dynamics emerge at higher levels of cognition, what new patterns might we observe?

For a Markov blanket to hold, internal states and external states need to maintain conditional independence from one another. In a body, this just means keeping temperature within the healthy range and maintaining metabolism, avoiding predators, etc. In the onchain environment, there is a security problem that could be interpreted as a Markov blanket in the sense that the blockchain's rules prevent users from simply writing whatever they want to the onchain ledger. The internal states (onchain record) are conditionally independent from events in the environment because they can only be updated if a user follows specific rules.

Now, the chain's environment also includes its users, each of whom is actively modeling the chain as well, yielding a massively complex recursive thought process.

I'm not going to pretend to understand that part yet, but 2025 did lead to a lot of progress toward a model that may one day be able to help us understand cognitive behavior at new levels of scale.

If the next year is a good one, perhaps we'll be modeling cognitive processes in new ways by the end of it. I definitely have at least 2 experiments getting ready to run. And there's a new robust core of theory underneath them that, if it stands up, will probably be a great place to build a career. The going has been tough but the summit is in sight.