Goals: Are they for LLMs, Cats, and Thermostats?

We often hear that the biggest risk from super-intelligent AI is that their "goals" won't align with human goals. But what if we're asking the wrong question?

This video takes philosophical look into the concept of goals itself. We move beyond motivational speeches to explore the fundamental question: When we attribute a goal to a thermostat, a chess engine, or an LLM, are we talking about the same thing as a human goal?

We analyze prominent definitions of intelligence—like John McCarthy's view that intelligence is goal-directed computation—and use classic philosophical tools to dissect the problem, including:

• Propositional Attitudes
• Direction of Fit (Mind-to-World vs. World-to-Mind)
• Value Theory (Intrinsic vs. Extrinsic Goals, from Aristotle to Mill)

By examining goals in everything from hunger and cats to hammers and mushrooms, we develop a better framework for understanding the true nature of AI intelligence.

Do LLMs, Reinforcement Learning Agents, and Chess Engines truly have goals? Or do they have something else entirely?

Find out why the answer to this philosophical question is essential for tackling the biggest risk in modern AI: the Alignment Problem.