Can AI predict the future?
Summary
Host Matt Bennett explores the predictive capabilities of AI by feeding frozen models a simulated battle plan for the 2026 Iran conflict, finding they accurately forecast military outcomes better than human experts. He contrasts this success with the models' poor performance in predicting sports results, financial markets, and political leadership changes, highlighting their reliance on recency bias. The episode ultimately argues that while AI excels at analyzing established military doctrine, it lacks the intuition to predict human decisions driven by ego or political will.
Topics discussed
Introduction and show branding
The AI memory gap and missing news events
How AI models are trained and the 'frozen in time' limitation
The concept: using frozen AI to predict the future
Iranian traffic cameras and Israeli intelligence surveillance
Netanyahu's pitch and Trump's approval of Operation Epic Fury
Setting up the experiment with three AI models
AI predictions on immediate war objectives
Long-term predictions and the Strait of Hormuz outcome
Testing AI accuracy on sports predictions
Testing AI accuracy on financial markets
Testing AI accuracy on UK political leadership
Why AI excels at military doctrine but fails at human decisions
The irony of US AI investment vs. Trump's gut instinct
AI response to Trump's tariff policy and episode credits
Listen ad-free on Castria