More Than Meets the AI

In 1987, Transformers introduced a line of toys called Headmasters — giant alien robots, already sentient, already powerful, who needed a smaller human-equivalent partner to become their head and direct them. It's a strange premise if you stop to think about it. Why would something already capable need a second, smaller mind layered on top of its own?

This essay is about that question, applied to something much closer to home: what it actually looks like when a human and an AI collaborate well, versus when that partnership fails. It covers IBM Watson's expensive, public collapse in oncology, why medical AI has gotten remarkably good at the easy 90% of diagnosis but still leans on human judgment for the hard part, why AI's lack of self-interest makes it a different kind of advisor than an institution protecting itself, and a real moment from a difficult professional situation where staying in a long-running AI partnership paid off in an unexpected way.

Also featuring: a toy Scorponok purchase thirty-three years in the making, and the part of his story that most people skip past — that the bond runs both ways.

More than meets the eye, as the old tagline went. Or, these days: more than meets the AI.

Update, June 23rd: A couple days after this episode went up, a widely-discussed MIT study on AI and essay writing crossed my feed — EEG data showing that students who let ChatGPT do too much of the thinking ended up with weaker brain engagement, worse memory of their own work, and less sense of ownership over what they'd written. It's good, careful research, and it lines up closely with what this episode already argued: the failure mode isn't AI itself, it's deferring to it completely instead of staying in the loop. Nice to see the argument holding up against data that came out after the fact.