Rational agent; think/act like humans

Rational agent

A rational agent is an agent that, for every possible percept sequence, selects the action expected to maximise its performance measure, given the evidence from the percept sequence and its built-in knowledge. Example: a robot vacuum that cleans the maximum area with minimum battery.

Environment --percepts--> [ Agent ] --actions--> Environment

"Systems that think like humans" (cognitive modelling)

This approach tries to build programs whose reasoning process matches human thinking. First we must know how humans think, through introspection, psychological experiments and brain imaging. If the program's input-output behaviour and the timing/steps match human data, its mechanism may reflect the human mind.

• It is the field of cognitive science.
• Example: Newell and Simon's General Problem Solver (GPS) compared its reasoning steps with the thinking-aloud records of people solving the same problems. Artificial neural networks that model brain neurons and cognitive architectures like SOAR are other examples.
• Practical use: A tutoring system that models how a student solves algebra, to find where the student goes wrong.

"Systems that act like humans" (Turing-test approach)

Here the behaviour, not the inner process, must look human. The system passes if a human cannot tell whether it is a machine (Turing Test). It requires NLP, knowledge representation, reasoning, learning (and vision and robotics for the total test).

• Example: A customer-service chatbot or a voice assistant like Siri that understands speech and replies naturally. A humanoid robot like Sophia that talks, recognises faces and shows expressions.

Justification: Both are parts of AI because AI covers two dimensions: thought process vs behaviour, and human-like vs ideal (rational). Thinking humanly and acting humanly are the "human-like" side of these dimensions; thinking rationally and acting rationally are the "ideal" side. Together they make the four definitions of AI in Russell and Norvig.