Home Reinforcement Learning and Artificial Intelligence (RLAI)

The experiential AI hypothesis

initial author: Rich Sutton

The ambition of this web page is to state, refine, clarify and, most of all, promote discussion of, the following hypothesis:

It is useful to think of experience, the temporal sequence of sensations and actions between agent and environment, as the data to the problem of AI.  The sequential nature of this data stream is the most important computational characteristic of the AI problem.

return to hypotheses

Definitions: AI stands for "artificial intelligence".

What about a-priori knowledge existing in the system BEFORE processing any data? For example, we could imagine that the brain structure itself, carefully crafted through Genetic Algorithm, represents some kind of a priori knowledge. This structure favors certain answers over others, affects decisions, and determines in a lot of ways what kind of rewards an animal receives through its neurotransmitters (hunger, pain, sex drive, etc).

It might therefore be the case that the only system that has no a-priori knowledge is simply a vacuum, and therefore any system in order to do some kind of computation must have some intrinsic knowledge in one way or another.  

Perhaps the comment above is really directed towards the empirical knowledge hypothesis?  The hypothesis here is about what the data is; it does not speak to what is happening in the agent other than the processing of the data.

Extend this Page   How to edit   Style   Subscribe   Notify   Suggest   Help   This open web page hosted at the University of Alberta.   Terms of use  1896/0