
Nov 22, 2026 · 30 min
AI’s fluency revives the question of machine understanding
The Global Story: AI can imitate a human. But can it think like one?
The episode shows why tests of intelligence—from IQ exams to chatbots—can measure performance without settling what understanding is.
- 1Intelligence tests capture particular abilities, but their history shows how easily narrow measures become misleading.
- 2Modern language models can imitate human reasoning through pattern processing without demonstrating consciousness or genuine understanding.
- 3AI’s future may depend as much on limits of data and embodied experience as on continued technical progress.
Don't miss
The discussion uses skateboarding to show why describing an embodied activity is not the same as performing or experiencing it.
The brief
Thomas Germain traces the search for machine intelligence from IQ testing and Alan Turing’s imitation game to today’s large language models.
IQ tests were designed to measure particular forms of reasoning, then misused by eugenicists; the episode treats that history as a warning about benchmarks.
ELIZA persuaded some users that a simple program understood them, while Deep Blue and Watson showed how machines could master tasks once seen as distinctly human.
ChatGPT and Claude may have made the Turing test less useful: they produce fluent language through pattern processing, but fluency alone does not establish understanding or consciousness.
The discussion tests AI against coding, household tasks, films and embodied activities such as skateboarding, asking whether text can supply experience or genuinely novel ideas.
The episode ends with a cautious view of AI’s trajectory and a refusal to define human intelligence through mystery alone, focusing instead on capabilities and social consequences.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.