CODEWITHSUNDEEP
machine-learningartificial-intelligence
user3094317
user3094317

Reputation: 15

What type of AI makes choices without considering the future?

In other words, what type of AI makes a decision based solely on what the current state of the game is? Thanks in advance!

Upvotes: 0

Views: 138

Answers (2)

Novak
Novak

Reputation: 4779

In the strictest sense of the word and the question, a pure reflex agent might satisfy the conditions. A reflex agent chooses an action solely based on the previous percept or on its current model of the world, so depending on what you consider "considering" to be, that might fit the bill.

However, that really stretches the point far past breaking, in my opinion.

An agent in an MDP setting, running a policy for that setting, might be said to be making its decisions only based on the state without consideration of the future. But that only begs the question: Where did that policy come from? It came, obviously, from a careful probabilistic consideration of the future rewards for taking particular actions from particular states. Or in other words, the algorithm of the agent might be "run this policy," but the algorithm of determining the policy and the future-orientation of the policy itself are where the actual intelligence is. We consider the agent to be intelligent by virtue of having determined that policy.

(And conceptually, the notion of reflex agents is really getting at the agent's treatment and consideration of the past, not the future. What I say above is true in a technical sense, not an important philosophical sense.)

All AI in the planning domain is future-oriented, by definition. If you're not at some level considering the results of your actions into the future, you're not an intelligent agent.

Upvotes: 3

ziggystar
ziggystar

Reputation: 28686

If you are interested in decision problems under uncertainty, you can have a look at influence diagrams.

Otherwise the problem appears to reduce to an optimization problem, of which there are many different kinds:

  • linear optimization
  • constraint satisfcation
  • convex optimization
  • non-convex (or general) optimization
  • ...

Upvotes: 3

Related Questions