This topic contains 3 replies, has 1 voice, and was last updated by 
josh March 25, 2021 at 6:19 pm.
-
Posts
-
joshRelated:
AI A1 has gathered a bunch of experience with Game X. Now A1 is asked to provide some bit limited communication about X to help A2, a presumably less experienced player. What should A1 say to be maximally helpful? It depends on established convention. Consider the idea that the convention includes, but is not limited to, proposing heuristics that attain elevated priority in the queue of hypotheses to consider & experiments to try.
AI A2 is not sure whether A1 is truly a great ally, a random blatherer, or an arch enemy. Estimates of those probabilities also affect the scenario.
Can player societies benefit from using heuristic languages to communicate & estimate sincerity? Seems likely.
-
joshThis Wikipedia classification of Reinforcement Learning Algorithms doesn’t do much to clarify the relation between the algorithms & different classes of given learning problems. For game AI learning, the learner is in a more advantageous position if it can simulate many forward paths in the game, as it compares different strategies. Some algorithms may want to add computational complexity cost or other internal structuring heuristics in the evaluation process.
-
josh“Communications of the Association for Computing the Information Content of Communications on the Associations of Incompletely Defined Structures” ///
I’m joking about the serious idea that it is relevant to try & describe some set-theoretic foundations experiences & dialogs about experiences in order to talk about measure (probability) on that. The exact details of the most productive models are currently unclear. Formal model building – where it helps to isolate abstractions that are worth continued analysis – is helpful. We’re currently in an exploratory phase where the empirical scientific foundations are still being developed at the same time.
You must be logged in to reply to this topic.