Artificial Agents

Solutions are to Problems, as rational agents are to task environments

---

What is an Agent?

• An agent is simply something that does or acts
• The word agent is derived from the Latin word agere, meaning to do.
• An agent is anything that can be viewed perceiving its environment via its sensors, and acting out upon its environment via its actuators.

So, what types of agents are there?...

1. Theoretical Agents
   • Omniscient Agenet
2. Stateless Agents
   • Simple Reflex (Stateless) Agents (AKA Reactive Agents)
3. Model-Based (Stateful) Agents
   • Model-Based Reflex Agents
   • Goal-Based Agents
     • Problem-Solving Agents
   • Utility-Based Agents
   • Q-Learning Agents
4. Planning Agents

• Knowledge-Based Agents
• Rational Agents
• Logical Agents
• Learning Agents

---

Honda's Osimo

Intelligent Agents

• An agent is said to be rational if it does the right thing, based on what is knows.
• An entity that is capable of perception, action and goal directed behavior.
• Such an agent might be a robot or an embedded real time software system.
• It is deemed intelligent if it interacts with its environment in a manner that would normally be regarded as intelligent if that interaction were carried out by a human being.
• To further break these definitions down, we have 4 important concepts...
  1. The performance measure.
  2. The agent's prior knowledge of the environment.
  3. The actions that the agent is capable of.
  4. The agent's percept sequence to date (this accounts for the percept limitations of the agent too).

For each possible percept sequence, a rational agent should select an action that is expected to maximise its performance measure (1), given the evidence provided by the percept sequence (4) and whatever built-in/prior knowledge (2) the agent has.

• To sum up...
  • Rational ≠ Omniscient (Percepts may not necessarily supply all relevant information (e.g. person being run-over by a truck)
  • Rational ≠ Clairvoyant (Action outcomes may not be as expected (e.g. imperfection; George Bush becoming the president of any country)
  • ∴ Rational ≠ Successful
  • ∴ Rational ⇒ Exploration, Learning, Autonomy

---

A self-repairing robot, Cornell University - Given some starting state, the robot tries to reform to produce two standing equal-sized towers, which it does so beautifully.

Agent Architecture

The job of AI is to design an agent program that implements the agent function mapping percepts to actions.

agent = architecture + program 

Agent Function

• An agent function is an abstract mathematical description.
• Describes what the agent does in all circumstances purely based on the condition-action rule

  if codition then action rule
  

Agent Program

• An agent program is a concrete implementation of the agent function which runs on the agent architecture.

Reactive agents choose their actions by what they currently see. Model-Based agents also choose their actions by what they see, but in addition, can use information gathered from the past]. Planning agents can use search techniques to further add to this by planning several steps ahead in order to achieve its goal(s).

---

References

•  Russell, S. and  Norvig, P. (2003).  Artificial Intelligence, A Modern Approach (2nd ed.). New Jersey, Pearson Education International

Related

Ancstors ☣ ..
Other ☣ PEAS