Rational agents
In my previous blog post, I tried to give a short, concise introduction about AI, where we could grasp what is AI and why should we care about it. In this post, I would like to discuss about rational agents.
Humans have always been fascinated by the way they think and have formally defined intelligence as rationality — doing the “right” thing. According to Merriam-Webster dictionary, rationality means the state or the act of being reasonable. Rationality can also have specialized meaning many academic disciplines, such as philosophy, art, psychology, game theory, etc. With respect to AI, we should consider the idea of coming up with internal thought processes or algorithms that resemble the state of being rational. This means acting without involving human emotions or biases that can impeach judgement.
One of the first people that brought up the idea of thinking rational was Aristotle through his syllogisms. A syllogism represents a form of deductive reasoning consisting of a major premise, a minor premise, and a conclusion. For example, we have the following construction:
Major premise: All humans are mortal.
Minor premise: I am a human.
Conclusion: Therefore, I am mortal.
The advantage of such patterns is that it will always provide a correct conclusion given correct premises. The disadvantage is that, in most of the situations, we do not know for certain whether we have indeed correct or certain premises. Thus, we are basing ourselves on unknown territory. That is where an important field in mathematics can help us deal with this degree of uncertainty. This is called theory of probabilities.
Still, what is an agent? An agent is something that acts ( from the Latin “agere”). Russel and Norvig distinguish computational and rational agents. According to them, a computational agent should do the following:
- operate autonomously
- perceive its environment
- persist over a prolonged time period
- adapt to change
- create and pursue goals
The rational agent must be able to achieve the best outcome or the best expected outcome when dealing with uncertainty. This means making correct inferences, finding the best course of action, and acting upon it. There are, however, some situation where plain abstract thought does not work and where other skills are necessary, such as careful deliberation and slow course of action.
Biography
Russel S., Norvig P. — Artificial intelligence, A modern approach, 4th edition