When AI can do almost anything, what's left that's yours?
Dear Magicians,
On Monday I asked whether a machine could feel Einstein’s happiest thought. This week I put that question, and a lot more, to Ethan Mollick, who runs the Generative AI Lab at Wharton. His new book, Co-Existence: The Next Phase of AI, is out this week. He builds with these tools every day, and he still draws firm lines about where they don’t belong.
The moment
Ethan told me he never lets AI take the first pass at anything that matters: not his writing, not his peer reviews, not his replies to students. His reason comes from his research on innovation. Brainstorming was invented in 1955, and within a couple of years researchers knew it backfired, because once you hear the group’s ideas, yours start to fit the group. AI does the same thing. It writes a seductive draft you can’t get out of your head.
He also named four human edges for working with AI: deep knowledge, wide knowledge, taste and agency.
“The enemy of slop is taste.” Ethan Mollick |
Take this with you
Before you ask AI anything this week, write your own first draft. Rough is fine. Then use AI as the second opinion, never the first. You’ll keep your own voice, and you’ll be able to tell when the machine gets it wrong.
Until next time,
Brian
Missed Monday? Can a machine feel the happiest thought of Einstein’s life? Read this week’s musing