Idea dump
1. Use might and van Horn's abstraction strategy (introduce a store and perform structural abstraction on the store) to the standard pathfinding, puzzle-solving, equation solving AI tasks that have used ad-hoc abstractions in the past.
2. Interpret a fair scheduler such as the recent linux scheduler in economic terms, as if the tasks are buying and selling the cpu.
3. Agre and Hammond are interactionists - focusing on stigmergy, similar to Shaw's "unreasonable man". They focus on stabilization, but stabilization is only one side of the coin. Possibly the slogan should be something like "Stabilize success, disrupt failure". In a genetic algorithm, underperforming members of the population get "killed" by being replaced with a random new individual. Possibly we can understand interactionist AI as if the agent is breeding "environmental memes" - a population of postit notes.
4. Brian C. Smith is interested in representation and intentionality. A subject (an agent) tracking an external object, sometimes directly in causal contact with that external object, and sometimes separated. This is what business programmers do over and over again, of course, building "entities" that model users, orders, other businesses, and so on. Benjamin C. Pierce has studied "lenses" which are (if I understand correctly) a sophisticated kind of synchronization. Can we combine B.C.Smith with B.C.Pierce to achieve extremes of representation and/or intentionality?
5. It's easy to write a context free grammar, such that strings in the grammar specify two-dimensional pictures - include productions for horizontal and vertical cuts, fills, and so on. If we apply Might and Van Horn's trick to this inductive type, maybe we could find a data structure representing sets of two-dimensional pictures that could capture the stability that we see in certain patterns - spacefillers, rakes, catacryst, or bitwise-serial delay-loop arrangements like the glider hotel patterns. It would be awesome to get a real abstract interpretation of 2d CA computation.
2. Interpret a fair scheduler such as the recent linux scheduler in economic terms, as if the tasks are buying and selling the cpu.
3. Agre and Hammond are interactionists - focusing on stigmergy, similar to Shaw's "unreasonable man". They focus on stabilization, but stabilization is only one side of the coin. Possibly the slogan should be something like "Stabilize success, disrupt failure". In a genetic algorithm, underperforming members of the population get "killed" by being replaced with a random new individual. Possibly we can understand interactionist AI as if the agent is breeding "environmental memes" - a population of postit notes.
4. Brian C. Smith is interested in representation and intentionality. A subject (an agent) tracking an external object, sometimes directly in causal contact with that external object, and sometimes separated. This is what business programmers do over and over again, of course, building "entities" that model users, orders, other businesses, and so on. Benjamin C. Pierce has studied "lenses" which are (if I understand correctly) a sophisticated kind of synchronization. Can we combine B.C.Smith with B.C.Pierce to achieve extremes of representation and/or intentionality?
5. It's easy to write a context free grammar, such that strings in the grammar specify two-dimensional pictures - include productions for horizontal and vertical cuts, fills, and so on. If we apply Might and Van Horn's trick to this inductive type, maybe we could find a data structure representing sets of two-dimensional pictures that could capture the stability that we see in certain patterns - spacefillers, rakes, catacryst, or bitwise-serial delay-loop arrangements like the glider hotel patterns. It would be awesome to get a real abstract interpretation of 2d CA computation.