ToKCast Ep 244: Deep learning is not "inductive".
43 snips
Aug 22, 2025 Dive into the nuances of reasoning with a fun exploration of deductive, inductive, and abductive approaches. Discover how AI and deep learning interact with these logical frameworks, questioning the true nature of predictive bias. The discussion sheds light on the limitations of merely extrapolating data, using relatable examples like boiling water to illustrate the need for deeper scientific explanations. Unpack the complexities of what artificial intelligence can really grasp and whether it has a true understanding or just patterns up its sleeve.
AI Snips
Chapters
Transcript
Episode notes
Deduction Is Formal And Reliable
- Deduction reasons from general rules to specific conclusions using logical forms like modus tollens.
- Deduction underpins mathematics and falsification in science as reliable, formal reasoning.
Induction Lacks Logical Justification
- Induction is commonly described as reasoning from particulars to general rules, but its logical justification is problematic.
- David Hume showed no amount of observation guarantees a universal conclusion, making induction philosophically shaky.
Abduction Labels Guessing, Not Explanation
- Abduction labels hypothesis formation or guessing but does not explain the mechanics of creative discovery.
- Creativity remains a mystery we can observe but not reduce to a known logical procedure.