Topic 6a – What Is AI?

Agents and Behavior

1. This is reflex behavior. The system responds directly and immediately to a sensor reading (lane position) with a fixed action (steering correction) without evaluating multiple possible goals or outcomes. There is no internal reasoning about where the driver wants to go or why — just a stimulus-response mapping. If the system also adjusted its sensitivity over time based on feedback, it would begin to exhibit learning behavior, but as described it is purely reactive.
2. This is goal-directed behavior. The program does not simply react to the current board position; it evaluates many possible future sequences of moves, maintains a representation of the goal state (winning), and selects the action most likely to achieve that goal. What distinguishes it from reflex behavior is the internal search and deliberation — the system is reasoning about potential futures, not just mapping an input to a fixed output.
3. As an agent: the spam filter is the agent. Its sensors are the mechanisms that read incoming email content and user feedback (the "mark as spam" action). Its actuators are the classification decisions it delivers (move this message to spam; leave it in inbox). It exhibits learning behavior at the point where it modifies its internal classification rules based on user feedback — when its behavior on future emails changes as a result of past experience, rather than only reflecting its original rules.
4. Procedural knowledge is knowing how to do something — knowledge embedded in skill and practice that is often difficult to articulate explicitly. Declarative knowledge is knowing that something is true — factual knowledge that can be stated in words. Teacher examples will vary; a sample pair: procedural = knowing how to manage classroom transitions smoothly (you do it without consciously listing the steps); declarative = knowing that a particular student has an IEP and what its accommodations are.

Strong AI, Weak AI, and the Turing Test

5. Weak AI (also called narrow AI) is AI designed to perform a specific task well, with no general understanding or awareness beyond that task. Strong AI (also called artificial general intelligence or AGI) would be an AI with general, flexible intelligence comparable to human cognition — capable of reasoning across domains, learning entirely new tasks, and having genuine understanding. Both a spam filter and a chess-playing program are weak AI: they are excellent at one narrow task and completely incapable of anything outside it.
6. The Turing Test proposes that a machine should be considered intelligent if a human interrogator, conversing with it via text, cannot reliably distinguish it from a human. Turing argued that the question "can machines think?" is too philosophically vague to answer directly, and that behavioral indistinguishability from a human is a practical, measurable substitute. If a machine behaves intelligently, he reasoned, the philosophical question of whether it "really" thinks becomes less important.
7. ELIZA used simple pattern matching: it scanned the user's input for keywords and responded with scripted phrases or turned the user's words back into questions (e.g., "You mentioned your mother — tell me more about your family"). It had no understanding of meaning, no memory of previous exchanges, and no internal model of the world. People believed it understood them because human minds are wired to interpret language as meaningful communication, especially in a conversational context. The ELIZA effect reveals that the Turing Test measures social believability, not intelligence — a system can pass it by exploiting human psychological tendencies rather than by genuinely thinking.
8. Producing fluent, coherent conversation is exactly what ELIZA demonstrated is possible without any genuine understanding. A modern language model's output feels intelligent because it is statistically consistent with patterns of intelligent human writing — not because it reasons, knows, or understands in any deep sense. The student is making the same mistake that ELIZA's users made: inferring intelligence from convincing output. The question is not whether the output is indistinguishable from human output but whether there is any genuine cognition producing it. This remains an open and genuinely hard philosophical question.