On Point with Meghna Chakrabarti Brainwaves: Is AI actually thinking?
10 snips
Feb 11, 2026 
Meet Kyle Mahowald, a linguist studying language and cognition, and Melanie Mitchell, an AI and cognitive science author. They probe whether AI's language tricks mean it thinks. Short tests with jokes, differences between prediction and embodied understanding, and when it makes sense to call AI 'thinking' are discussed in lively, accessible conversation.
AI Snips
Chapters
Books
Transcript
Episode notes
Thinking Is A Loose, Useful Shorthand
- Thinking isn't a single scientific term but a useful shorthand for many cognitive processes.
- Early AI modeled thought as logical, language-linked steps and that approach failed to capture real human thinking.
Question AI Humor And Explanations
- Be skeptical of fluent AI humor and test outputs critically rather than assuming understanding.
- Ask follow-ups and check for grounding or evidence before trusting model claims.
Next-Word Prediction Drives LLM Skill
- Large language models learn by predicting the next word across massive text corpora.
- That simple objective yields capabilities in language structure and conversation without grounding in the physical world.
