According to the definition of a fully observable environment in Russell & Norvig, AIMA (2nd ed), pages 41-44, an environment is only fully observable if it requires zero memory for an agent to perform optimally, that is, all relevant information is immediately available from sensing the environment.

From this definition and from the definition of an "episodic" environment in the same book, it is implied that all fully observable environments are, in fact, episodic or can be treated as episodic, which doesn't seem intuitive, but logically follows from the definitions. Also, no stochastic environment can be fully observable, even if the entire state space at a given point in time can be observed because rational action may depend on the previous observation that must be remembered.

Am I wrong?


Before answering the question directly, we must research the cited book “Russell & Norvig, AIMA”. It is part of the hacker library and stays on one level with the “Red book”, “Purple book” and the “Dragon book”. The problem with the AIMA-book is, that it was overrated in the past. In contains only of 1200 pages, which is equal to 50 normal papers which are uploaded to Arxiv. And no, it is not enough for an Artificial Intelligence reference.

But let us go into the details. Russel&Norwig have in their book a chapter called “intelligent agents” which is 25 pages long. This is the part, the OP is directing us. The problem with the Agent-chapter is, that it has no references to other papers and the explanation what an “episodic task environment” should be remains superficial. Also, Russel&Norwig is wrong, if the see episodic tasks in contrast to sequential tasks, because chess can be played with an episodic memory in the opening book.

I wouldn't call a game (aka an environment) episodic vs sequential. Because implementing the game in software is not the problem. Instead the agent strategy has to fulfil constraints, that means the agent can use prior experience or not.

  • $\begingroup$ It's refreshing to know other people think this book is overrated, and chapter two is especially flawed. I would disagree that 1200 pages is not enough for a high level overview of most of AI though. Papers and text should not be directly comparable by page count. Of course for a deep understanding of topics and understanding latext work one must read articles. $\endgroup$ – Francis M. Bacon Mar 18 '18 at 6:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.