Theses on Cognition

(The following is a list of theses/claims/beliefs that I believe in with >50% credence and that I intend to investigate more closely/think about over the next few months, that is, roughly, until the end of 2025.)

• All cognition is, fundamentally¹ "squishy" at the core. Cognition requires interfacing with a messy multidetailed reality which in turn necessitates that cognition acquires messiness and tunes its messiness so that it properly operates with messy reality. This doesn't mean it's not Lawful™ but its Lawfulness is messy, a complex system that doesn't admit a neat formal description.
  • Still, the messy complexity of cognition does not make cognition impossible to understand. Messy complexity is fully compatible with there beign a way to describe, track, control (?) a system, ensuring that some of its relevant/key "summary statistics"/"high-level characteristics" remain within certain desirable bounds.²
  • Cognition can adopt non-squishy/tractable/formal structure for the purpose of increasing its own stability, reliability, in addition to adaptivity of its well-tuned messiness. (Stability-adaptivity dichotomy is a very provisional handle here.) Examples:
    • A human can do some math to get better informed about a thing, incorporate that new information about the thing, and then go with the gut, that gut being better informed (better attuned) than if they counterfactually didn't any math at all.
    • A mathematician uses formal systems to construct proofs of mathematical propositions, asserting their truth or falsehood with conditionally full confidence. However, a proof or a formal system is not the "essence" of mathematics. A proof is more like a receipt, a confirmation that you did your mathematical-truth-seeking job well.³
    • (One should be wary of drawing analogies between adaptivity in general and cognition in particular but …) Most of flesh is squishy but some kind of skeleton (bones or shells or wood) helps squishy flesh do its job better.
• The Great Reductionist Project of reducing everything to math+physics is another attempt to salvage the same dream that e.g. logical positivism was trying to salvage.⁴
  • There is no way derive an abstraction from scratch from atoms without imposing on it our subjectivity/embededness/value-directedness. This also holds for "morally relevant" abstractions, e.g. those that people tend to summarize using words such as "sentience" or "moral patience" or "moral agency". Often, we are faced with a genuinely open choice which forces us to genuinely decide and commit to a certain ontology, possibly including a decision what to value.
  • For structurally similar reasons, the attempts to reduce all value to terminal value (or being in service of it) are doomed to fail, at least as long as we actually want to meaningfully preserve our current normativity (so that we don't get more utility via redefining utility).
• Compression models of cognition postulate stuff like: "An agent is something that can be usefully modeled using the intentional stance." or "To believe X is to be disposed to act as if taking that X is the case". Essentially, they are trying to reduce some phenomenon to a not-necessarily-existent referent of an instrumentally useful mental element (/abstractum).
  • The key problem with compression models is that they are largely dodging the questions they are supposed to answer instead of actually answering them. The response to the intentional stance would be: "Why is it that some phenomena can be sensibly modeled using the intentional stance whereas others cannot?". The response to dispositionalism about beliefs is: "What makes it so that certain systems are sometimes well-modeled in such and such way and this is parametrized by what we call 'beliefs' and there is a certain logic behind those beliefs changing?".