Week Two: What is Intelligence?
Could a machine ever truly think, or does it only appear to?
1. A Mind Map
Intelligence: the ability to solve, learn, reason, adapt, or achieve goals Understanding: grasping what symbols, ideas, or situations mean Consciousness: subjective experience — what it is like to be a system; the felt quality of seeing red, tasting chocolate, or feeling pain Intentionality: the "aboutness" of mental states — thoughts being about objects, facts, or situations in the world Agency: the ability to act toward goals in an environment Substance Dualism
Descartes posited that mind is separate from body. How does the mind interact with the body? Physicalism (Materialism)
Dominant view in modern philosophy Mental states depend entirely on physical processes Identity theory: mental states = brain states Brain damage reliably alters personality; Drugs alter experience. Functionalism
mental states are defined by their causal and functional roles Pain is defined by its functional role (attention, avoid...) any system that functions like that could be considered to be experiencing pain multiple realizability: a mental state can be realized in different physical substrates If functionalism is correct, then the substrate does not matter.
2. The Chinese Room
John Searle wrote "Minds, Brains, and Programs" in 1980. Person in a locked room receives Chinese pictographs, looks up and sends back correct responses. Person does not understand Chinese but folks outside the locked room may think he does. Syntax (rules) is not semantics (meaning).The Objections Searle Addresses
The Systems Reply
Even if the person in the room memorized all of the rules, they still wouldn't understand Chinese.The Robot Reply
Even a robot with sensors that interacted with the real world would not understand Chinese. Simply connecting more inputs and outputs to symbols does not result in understanding.The Brain Simulator Reply
Even a neural net that simulated a fluent Chinese speakers brain would not understand. Simulation is not duplication.
3. Dennett's Response: Why Intuitions Can Lead Us Astray
3 ways to predict and explain a system
Dennett says we can't comprehend human understanding, therefore we should treat AI as intentional.Physical examine the physical properties of the system (atoms, forces) Design examine the system's design and purpose Intentional treat the system as a rational agent with beliefs, desires, and goals.
4. Where Does This Leave Us on AI?
View What matters most? AI implication Turing-style behavioral criterion Indistinguishable intelligent behavior If it behaves intelligently enough, that may be sufficient evidence of intelligence Functionalism Functional/causal organization AI could think if it has the right causal/ functional structure, regardless of substrate Searle Biological causal powers and semantics Programs alone cannot generate understanding; formal syntax cannot produce meaning Dennett Predictive/explanatory intentional patterns If the intentional stance works deeply and reliably, the "really" question may be misplaced Stochastic parrot Statistical pattern production without grounding LLM fluency should not be mistaken for understanding or meaning The "Stochastic Parrot" View
Stochastic (random, probablistic) LLMs predict likely next tokens based on training data without any grasp of meaning. Explains hallucinations. The Functionalist Counter
LLMs improve with scale. They can do multi-step planning and reasoning. They can access external tools and lookup more info. Emergent? Not clear if modern LLMs are Chinese rooms. The Question You Cannot Avoid
No theory of consciousness lets us verify whether a system has experience
5. Why These Aren't Just Academic Questions
Trust calibration – LLMs require more oversight than humans Governance and accountability – laws depend on whether AIs understand what they're doing Moral status and human relationships – if AI do experience feelings, then they should be treated morally
Reading: Searle, "Minds, Brains, and Programs" (1980)
Which of the objections Searle considers do you find most compelling? Does his response to it succeed? Is AI strong (a mind like ours) or weak (just a tool) Does the Searle argument apply to neural networks that learn their representations from data rather than from explicit rules?