I talk a lot here about how to use AI. Why you should learn about it and how to use it. Why it’s not scary. Why, instead, it’s the key to your future as a worker.
But here are 4 times you shouldn’t use AI (these are taken directly from Ethan Mollick’s One Useful Thing newsletter with some editorial comments from me):
- When you need to know, learn, and deeply understand something. AI can (and should) be a helper for you here, but for you to deeply learn something, you have to do the work, reading, and thinking yourself. You can use AI to help you learn the mechanics of reading the sheet music, but you’ll never be able to play the piano without sitting at the keyboard yourself.Â
- When high accuracy is required. AI errors and hallucinations can be difficult to spot, especially if you’re not an expert on the topic. AI really looks like it knows what its talking about. But you should treat it a bit like the blowhard know-it-all neighbor down the street. He’s often right, but he’d rather make shit up than admit he doesn’t know. And there’s always some backstory that makes it plausible.Â
- When the struggle for true creative knowledge is required. Current AI is generative only in the sense that it can discover patterns and create new patterns from all of the knowledge it has. But what about new knowledge? If all AI knew about the rainbow was OYGBI (instead of ROYGBIV), then it could never generate the R or the V or any intermediary that requires those. The struggle is required for the discovery of R and V.Â
- When AI is bad at something. Yes, AI stinks at some particular tasks or situations. For example, it sucks at sarcasm. Just think of it as being on the spectrum.Â
AI is great and you’re gonna need to use it, but it’s not great at everything.Â
