1
$\begingroup$

In Artificial Intelligence: A Modern Approach Edition 3, Page of 43, At Single Vs. Multi-agent section's last line, Writer says,

In some competitive environments, randomized behavior is rational because it avoids the pitfalls of predictability.

Can someone help me understanding this line? Specially, what does he mean by pitfalls of predictability ?

$\endgroup$
2
$\begingroup$

Think of playing the game rock-paper-scissors. If you were using a deterministic algorithm to choose which of rock, paper, or scissors to chose, you'd also choose the same one. Thus, if I played you, I could quickly figure out how to learn every time, because you are predictable. So, it is a better strategy for you to randomize your choice: choose uniformly at random among the three options every time. That makes you unpredictable, and increases your chances of winning the game.

| cite | improve this answer | |
$\endgroup$
2
$\begingroup$

Someone who owns (an abstract of) the book might be able to give you a better contextualized answer, but in general models and simulations are subject to certain 'laws' that the systems they represent usually are not. In the case of multi-agent systems, one of these is the fact that some agent gets to act first on a given time step, even though the designer often assumes and intends for them to act in parallel (this is the result of computers being incapable of truly parallel computation). If the same agent were to consistently act first, that could introduce significant bias into the system. This is probably what the author means with the pitfall of predictability and why randomization is important; so that we can prevent (to some extent) these sorts of issues.

| cite | improve this answer | |
$\endgroup$

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.