Steven Pinker: AI in the Age of Reason | Lex Fridman Podcast #3

Lex Fridman interviews cognitive scientist Steven Pinker about the nature of intelligence, consciousness, and the meaning of life. Pinker argues that fears of AI as an existential threat, the takeover scenario and the paperclip value-alignment scenario, are incoherent because intelligence does not imply a will to power and engineers don't deploy untested all-powerful systems. He champions the safety-oriented culture of engineering and the humanitarian promise of AI, such as autonomous vehicles preventing tens of thousands of highway deaths. The conversation closes with Pinker reflecting on the books that shaped his thinking and the psychology of negativity bias.

• Pinker says the only truly mysterious thing about human neural networks is subjective first-person conscious experience, which he doubts we can ever know exists in machines.
  00:04:13
• He calls both major AI existential-threat arguments incoherent: the AI-takeover fear confuses intelligence with a will to power.
  00:11:32
• Pinker states building nuclear weapons was a massive mistake and possibly avoidable if not for World War Two.
  00:12:34
• He argues the paperclip/value-alignment scenario is self-defeating: a system smart enough to cure cancer would not be too stupid to know we don't mean killing everyone.
  00:15:13
• Pinker contends genuinely intelligent systems by definition weigh multiple constraints, so single-minded goal pursuit is 'artificial stupidity,' not AGI.
  00:18:50
• Pinker reveals he will appear on Joe Rogan's podcast in a couple of weeks, and notes Rogan is concerned about AI existential threat.
  00:30:21

• Humans are unusual among animals in acquiring knowledge to make tools, strike agreements via language, and extract poisons, letting us occupy every niche.
  00:02:41
• Pinker says current deep learning systems lack a semantic level, extracting statistical regularities without understanding who did what to whom.
  00:05:49
• He compares safe AI design to building cars with brakes: engineers trade off across multiple goals, not maximize one blindly.
  00:18:50
• About 40,000 people die yearly on US highways, vastly more than terrorism deaths, yet a trillion dollars was spent on the war on terror.
  00:18:50
• Pinker argues jobs automated by AI are often soul-deadening and dangerous, so eliminating them is a boon if income is redistributed.
  00:20:26
• A tenet of cognitive psychology: perception of risk and fear is driven by imaginability, not data, causing massive misallocation of resources.
  00:28:48
• Pinker reflects on negativity bias: as satirist Tom Lehrer said, 'always predict the worst and you'll be hailed as a prophet.'
  00:33:31
• George Miller, the psychologist, originated the famous idea that human memory holds seven plus or minus two chunks.
  00:36:38