Metacognition
I learned a word recently: metacognition. Thinking about thinking. It landed harder than a new word usually does, and it took me a minute to figure out why. Then it clicked. I'd already written about this twice. I just didn't have the name for it.
Two Articles, One Muscle
I wrote one piece about approaching AI with an open mind — the cave explorer who can't see around the next bend and decides that's an opportunity instead of a problem. I wrote another about how a good guess is its own kind of intelligence.
Both were about the same thing. I didn't know it at the time. One was about staying loose enough to keep moving when you can't see ahead. The other was about what you actually do once you get there — you guess, and you try to guess well. Neither was really about AI. They were about how I was thinking. That's metacognition. I just didn't call it that.
Why Now
Because AI moves faster than anything I've worked on, and that speed is what forced the change.
The solution from 20 years ago isn't the solution from 10. Cloud, dynamic languages — you could watch those shifts coming over years. With AI, the right answer from last month can be wrong today. When the answer changes that fast, you can't lean on what you know. You have to watch how you're thinking, and be willing to throw out a belief the moment it stops paying off. That's a different skill than knowing things. Knowing things used to be enough. It isn't anymore.
Seeing Around Corners Is a Guess
Jensen Huang said something that stuck with me — that real intelligence is being able to see around corners. To sense a problem before it shows up. He says it comes from a mix of data, first principles, experience, wisdom, and reading the people in the room.
I'd add one word to that: it's a guess. A good one. Seeing around the corner isn't knowing what's there. It's forming the best guess about what's there from everything you've got. Which is exactly what I'd written about. He described my guessing article from the outside and called it intelligence. I think he's right.
The SAT Has It Backwards
In the guessing piece I wondered how any of this relates to standardized testing. Huang has an answer — he figures the person who sees around corners might do terribly on the SAT. I think it's the opposite, and that's the part worth sitting with.
A good guesser can land on the right multiple choice answer without knowing the material. Not by luck — by synthesis. The question itself, the four options, the way these tests are built — there's enough there to formulate a good guess, and a good guesser will find it. So the test doesn't measure what it thinks it measures. Huang sees the corner-seer failing it. I see the corner-seer beating it without the knowledge it's supposedly checking for. Either way the instrument is wrong. We just disagree about which way it breaks.
The Biology Underneath
Geoffrey Hinton brought biology into computing — modeling how we, and animals, think using neural networks. It clicks because it isn't a metaphor. It's closer to the blueprint. And once you're in that frame, other things start to map over.
Dreaming, for instance. I think of it as the brain reorganizing the day into better storage, indexing it so the lookup is faster later. Data needs the same thing. Building AI systems, you end up doing for your data what your head does for you overnight.
That's the part that forced the issue for me. You can't design how a machine organizes, recalls, and guesses without examining how you do it yourself. Thinking about thinking stops being a nice phrase and becomes the actual job.
Wrong to Be Right
First principles asks you to break a problem down to the things that can't be broken down any further. Sounds clean. In practice it means admitting that a lot of what you assumed is wrong — that you have to be wrong about plenty in order to be right about anything.
I keep coming back to this. It was in the guessing piece, and here it is again, which is probably the tell that it's true. It's irony at its finest. The path to a solid answer runs straight through admitting how much you had backwards.
Own Your Guesses
So here's where I've landed. The open mind, the cave, the guess, the corners, the dreaming data, the first principles — they aren't separate ideas. They're the same muscle. Watching how you think, and being willing to change it. There's a word for that, and I finally have it.
For anyone doing this work, I don't think it's optional. You can't manage a machine that guesses if you've never examined how you guess. Give it the right context and the right values, the same way you'd want them yourself. And then own the guess.
Guesses are good. Explainability is king. I just know the word for why now.