Welcome to the cognitive science & AI architectures track. Before studying any particular mind or machine, you need the master tool for keeping descriptions of them from talking past each other.
The vision scientist David Marr argued that any information-processing system must be understood at three levels, each answering a different question:
- Computational level — What problem is being solved, and why? What are the inputs, outputs, and constraints that any solution must respect? (Marr’s example: a cash register’s computational theory is arithmetic — the properties of addition exist regardless of the machine.)
- Algorithmic level — How, abstractly? What representations are used, and what processes manipulate them? (Addition could be done in decimal or binary, by lookup table or by carrying.)
- Implementational level — In what physical stuff? Neurons, transistors, gears — the substrate that realizes the algorithm.
The levels are partly independent: one computational problem admits many algorithms; one algorithm runs on many substrates. A claim pitched at one level neither confirms nor refutes a claim at another — and an enormous amount of confused arguing (about brains and about AI systems) comes from not noticing which level a claim lives at.
Try it on something mundane. Four true claims about a spam filter appear in the question above. Exactly one of them is an algorithmic-level claim — it commits to representations and processes, not just to the problem, and not to the hardware.