Rationality

What do you infer from a Rational Agent? Books would say that the agent that takes the best action in a sitaution is rational. But we can agree that this defination though true, is a bit vague. What does best action means? Different readers will have different definations what the best action is according to them. Or this best action can change from time to time. So, at this point it becomes need of the hour to define the notion of performance measure. A performance measure with respect to whom, agent or reader? An agent w.r.t itself can achieve perfect rationality by deluding itself that it's performance was perfect. So, the performance measure or more importantly the rationality of an agent depends on the reader/user who wants the agent to work at it's best for the situations it is made.

Performance measures

Vacuum World

To highlight the concept of rationality and performance meaure we understand the case study of two agents for the vacuum world.

It is simple enviroment with 2 floors that get dirty and have to be cleaned by our vacuum agents. We have two type of agents that can be used for vacuum world.

  • Agent1 that cleans continously but has slower reflexes.
  • Agent2 has faster reflexes but needs to shutdown at regular intervals

So, according to you which agent is better? If the answer you are thinking is ," well it depends on situation and our needs", then you are surely understanding the concept of rationality. What if we are more concious about the cleaniness of the environment, than agent1 sounds better to rely on, but if are more worried about the power consumption, then agent2 might serve our purpose.

The performance would be measures by the score of the agent through time. The performance measure of the agents would be a sum total of scores evaluted from:-

  • The amount of time the agent is working( 10 points would be deducted from the total score for each move)
  • The cleaniness of the environment( 50 points would be awarded for for cleaning the environment and 5 points would be deducted for each time interval the environment remains dirty)

There are many parameters at play that may affect the performance of our agents, which are left for you to play around and find the best agent in different situtaions. Readers can surely argue why these factors are important. Following control the agent's performance in a given environment.

Performance Factors
dirt-frequency speed-agent1 speed-agent2 interval-agent2

Agent1

Agent2

Agent's Performance Plot

Conclusion

The rationality of the agents are defined w.r.t to the reader and can vary from situation to situation and from time to time. Perfect rationality might not exist. The aim we should have is not to become perfect, but to work for achieving perfection and make are agents more efficient.