Unintelligent AI Matters

TikTok's recommender doesn't reason.

It doesn't read the videos. It doesn't understand the audio. It has no language model, no chain of thought, no opinion about whether anything is interesting.

It is, on every axis a philosopher might use, profoundly dumb.

It also decides what 1 billion humans see today.

Two hours per user per day. 730 billion hours of human attention per year, routed by something that scores videos on click-to-view ratio and watch time.

If the threshold for "this AI matters" is "it can think" — TikTok doesn't qualify. The threshold is wrong.

Otter is a speech-to-text engine. It transcribes a meeting. The underlying technology — sequence models with acoustic features — has existed since the early 2010s.

None of it is novel. None of it would impress a researcher. It's commodity infrastructure.

It also restructured 50 million meetings per day.

The role of "person who takes notes in the meeting" no longer exists, almost anywhere. Institutional memory — the thing the note-taker carried in their head — is now in a searchable transcript.

The decisions that flow from those meetings come from a different place now. Not because the AI is smart, but because it's there.

The third one isn't even AI in the modern sense. Applicant tracking systems — the software that screens resumes for Fortune 500 jobs — is mostly keyword matching. Regex. Boolean queries on a database.

It would not pass any reasonable AI benchmark. It is not even, technically, learning anything.

It also rejects roughly 75% of resumes before a human ever sees them. Three out of four people who apply for a job are filtered by something that wouldn't qualify as AI by any meaningful definition.

And it decides — at industrial scale — who gets to enter the workforce, in which fields, at which companies.

The intelligence is irrelevant. The placement is everything.

The systems that reshape your day are mostly not smart. They don't need to be. They just need to be in the path.

— From Reshuffle, Chapter 1

The debate about "is AI conscious / sentient / actually intelligent" is the wrong frame. The vast majority of consequential AI in your life right now isn't intelligent in any interesting sense. It's just standing between you and a decision.

If you're waiting for "real" AI before you take the reshuffle seriously, you'll be the person waiting at the gate that's already open.

Sangeet on this in Chapter 1 ↗