Criteria for Success

How do we decide whether a system is "intelligent"? AI proposes several concrete, testable criteria rather than relying on intuition alone.

The Turing Test

Proposed by Alan Turing in 1950 as the "Imitation Game." A human interrogator communicates via text with two hidden parties — one human, one machine — and must decide which is which. If the interrogator cannot reliably tell them apart, the machine is judged to have passed.

Capabilities the Turing Test implicitly requires:

The Total Turing Test extends this with a video signal (to test perceptual ability) and a slot to pass physical objects (to test robotic manipulation) — adding computer vision and robotics as further requirements.

Criticisms of the Turing Test

• It rewards deception (tricking a judge) rather than genuine reasoning
• It says nothing about internal thought process — only observable behaviour ("black box" criticism, cf. Searle's Chinese Room argument)
• It is anthropocentric — a very intelligent but non-human-like system could fail it

Alternative / Complementary Criteria

Modern AI evaluation leans heavily on the rational agent view: success is measured by how well an agent's actions achieve its goals given its percepts, knowledge, and constraints — a more measurable standard than asking whether it merely "thinks" or "acts" like a human.