본문 바로가기
A·I

What is AI and an Agent?

by 방구석 몽상가 2019. 4. 18.

What is AI?

According to Luger and Stubblefield, the AI is "The branch of computer science that is concerned with the automation of intelligent behavior". But AI cannot be simply defined.

There are largely four views of AI.

 

1. Thinking humanly

If we listen any music, we will feel diverse things as excitement, sadness and joy. Then how we know feeling of a music? It is very difficult to answer about that. So, thinking humanly requires scientific theories of internal activities of the brain.

Suppose that you somehow made such an AI. Then you must validate made AI. There are two ways to do this.

  1. Predicting and testing behavior of human subjects (top-down)
  2. Direct identification from neurological data (bottom-up)

But those are now distinct from AI. Those are approaches such as Cognitive Science and Cognitive Neuroscience.

 

2. Acting humanly

Representative example of testing 'Acting humanly' is the Turing test. The Turing test, developed by Alan Turing in 1950, is a test of a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human (Wikipedia). That test suggested major components of AI: knowledge, reasoning, language understanding, learning. But the Turing test is not reproducible, constructive, or amenable to mathematical analysis. In conclusion, validating 'Acting humanly' is not easy too.

 

3. Thinking rationally

It is not easy for people to think rationally. Furthermore, standard of thinking rationally is ambiguous. That is close to philosophy. Maybe, the validation standard will be defined about "What is the purpose of thinking?" and "What thoughts should I have?".

 

4. Acting rationally

The modern AI models are almost acting rationally. Rational behavior is to do the right thing which is expected to maximize goal achievement, given the available information.

 

Agent

An agent is an entity that perceives its environment through sensors and acts upon it through actuators. A rational agent chooses actions maximize the expected utility. But, computational limitations make perfect rationality unachievable. So, you must design best program for given machine resources.

When you design an agent, you should consider 4 things.

  • Percepts
  • Actions
  • Goals
  • Environment

For example, if you design Internet shopping agent.

  • Percepts : User demand
  • Actions : Get the list of products for user, Choose products
  • Goals : Choosing good products, Choosing the right products for your budget
  • Environment : Website, Consumers, Retailers

The environment type largely determines the agent design. Because agent functions varies greatly depending on the environment.

  • Accessible
  • Episodic
  • Static / Dynamic
  • Discrete / Continuous
  • Stochastic / Deterministic

 

Agent types

  • Simple Reflex agents make decisions based solely on the current environment .

    The agent perceives the current environment by sensors. Then, according to condition-action rules, the agent determines what action I should do now.

  • Reflex agents with state takes percepts as input and keeps internal state.

    Unlike simple reflex agents, the agent perceives current conditions through the current environment and internal states. And a memory is need to keep internal states.

  • Goal based agents is capable of thinking beyond the present moment to decide the best actions to take in order to achieve its goal.

  • Utility based agents is an agent that acts based not only on what the goal is, but the best way to reach that goal.

댓글