Artificial Intelligence/Knowledge Representation

From Dev Wiki
< Artificial Intelligence
Revision as of 07:28, 14 December 2021 by Brodriguez (talk | contribs) (Expand page)
Jump to navigation Jump to search

For all but the most basic AI, there will need to be some representation of the surrounding environment and world.

See also: Discrete Math - Propositional Logic


Terminology

  • Intelligent Agent - an entity that perceives some value from the environment, and makes rational choices accordingly. The agent attempts to do some action to or within said environment.
  • Knowledge-Based Agents - Takes the "Intelligent Agent" a step further. They try to form internal representations of the world, and use inference to update these representations to the best of their ability.
  • Knowledge Base (KB) - The set of information maintained by the Agent.
  • (KB) World - The environment in which a KB Agent exists within and gathers knowledge about.
  • (KB) Sentence - A language representation for some assertion about the world.
  • (KB) Model - A "possible world", comprised of a set of specific state information.
  • Entailment - With logic, when one thing naturally follows from another.
  • Inference - The derivation of new sentences, from existing sentences.
  • Valid Sentence - A sentence which is true in all models (aka, all possible worlds).
  • Satisfiable Sentence - A sentence that is true in some models (aka, only certain possible worlds).
  • Unsatisfiable Sentence - A sentence that is true in no models (aka, no possible worlds).
  • Completeness - A KB in which "not a" is False also implies that "a" is True}.
  • Soundness - A KB in which "a" is True also implies that "not a" is False}.


Knowledge Bases

The KB generally contains domain-specific content. Often, domain-independent algorithms act as the "inference engine", working to update the knowledge base and help the agent decide what action to take next.

If information available in the KB is both correct and complete enough, then the conclusion is guaranteed to be correct.

KB Representation

Often, the KB is represented in the form of "sentences":

  • Each "sentence" is represented in the form of a formal "knowledge representation language" with a consistent syntax.
  • Each "sentence" also represents some assertion about the surrounding world.
  • Sentences should be defined such that conclusions can be drawn from them.
  • Generally speaking, each new percept updates the KB with new information.