Definitions of intelligence

Summary

The topic “Definitions of intelligence” in the philosophy of AI examines how intelligence can be defined and measured, noting that answers depend on what is meant by “intelligence” and which kinds of machines are considered. In AI research, intelligence is often treated as a question about machine behavior rather than about whether machines truly “think.” A classic proposal is Turing’s test: if a machine’s conversational behavior is indistinguishable from a human’s, then it can be called intelligent. However, critics argue it may measure “humanness” rather than intelligence itself. Another major approach defines intelligence as goal-directed performance. Here, an intelligent agent is one that perceives and acts in an environment to maximize success according to a performance measure; the more problems it solves well, the more intelligent it is. This approach avoids testing for superficial human traits (like typing mistakes), but it can blur the line between genuinely intelligent systems and simpler goal-like devices (e.g., a thermostat). The content also contrasts arguments that machines can show general intelligence—such as the idea that brains can be simulated and that human thinking is symbol processing—with arguments against purely symbolic accounts, including Gödelian anti-mechanist claims and Dreyfus’s view that human expertise relies heavily on implicit, intuitive skills rather than explicit step-by-step rule manipulation.

Key Points

• 1

  Turing’s test defines intelligence via indistinguishable human-like behavior in conversation, though it may conflate “humanness” with intelligence.

• 2

  Goal-based agent definitions treat intelligence as maximizing success in an environment, but can make very simple systems count as intelligent.

• 3

  Arguments for machine general intelligence include brain simulation and symbol-processing theories, while critiques include Gödelian anti-mechanism and Dreyfus’s emphasis on implicit skills.