There is a new participant in more of your meetings than there used to be, and it never says a word. It sits in the list with a little icon, it listens, and when the call ends it tries to turn the talk into a transcript, a summary, and a list of who agreed to do what. And while it listens, something strange happens in the room.
A company that sells one of these tools studied 159,870 of its own virtual and hybrid meetings over about sixty days. In meetings where its AI was present, women spoke 9 percent more than men, measured against how many of them were there. Junior people spoke nearly as often, and nearly as fast, as their managers. The people who usually get less of the floor took more of it.
Read that again, because it runs backwards. You would guess that a silent recorder in the corner makes people clam up. Instead the people who usually say the least said more.
There are dull ways to explain this, and they are worth ruling out first. One is that the bot just did the chore. Someone always ends up taking the notes, and that someone is disproportionately a woman or the most junior person present; hand the typing to software and they get their attention back. That is real, and it explains part of it. It does not explain why individual contributors drifted to almost the same speaking pace as their managers, down to a word or two a minute, while engagement and sentiment scores stayed high. The other dull explanation is that these are the vendor's own opt-in numbers, with gender guessed from first names and no control group, a correlation in a nice suit. Fair. Take the exact figure with salt. The pattern still points somewhere, because it matches something you already know from the rest of your life.
People behave differently when they know they are on the record. That is the whole of it. You have felt it when a camera light comes on, when you notice the call is being recorded, when your thumb hovers over send on an email you know could be forwarded. Something in you sits up straighter and edits. For the person who usually gets talked over, that same edit can read as cover: the record is a witness, and a witness makes it a little safer to take up space. The bot is not a stenographer. It is a witness the whole room can feel.
Which is the good news and the bill together. The thing that made the meeting fairer is the thing that makes it less honest. A room that remembers everything is a room where people say a little less that is true.
The story we tell about this is clean and mostly wrong. The story is that memory beats forgetting, that a meeting which finally keeps a perfect record is just a meeting that finally works, that the transcript is pure upside and the only people who mind are the ones with something to hide. Better notes. Fewer arguments about what was agreed. Nobody left out because they missed the call. Who could be against remembering?
The fairness is not a rounding error, because airtime buys more than airtime. One 2020 study found that every extra 39 seconds a person talks in a group buys them roughly one more vote as the leader, regardless of what they actually said. We read the floor as a stand-in for who is in charge. And the floor has long tilted one way: in one study of more than 155,000 corporate conference calls over nineteen years, men did 92 percent of the talking. The written record was rationed too. In Britain, company law requires board minutes to be kept for at least ten years, and the person who writes them has always decided what to leave out. Memory at work was never neutral. It was scarce, and the scarcity followed the power.
So here is the strongest case against me, and I want to give it its full weight, because it is good. The candor I am about to mourn was mostly the candor of whoever already held the floor. The off-the-record room did not protect the truth. It protected the interrupter, the person whose offhand remark cost him nothing and cost someone else plenty to hear, the manager whose account of the meeting became the account because he was the one everyone remembered. "What happens in the room stays in the room" is a rule written by the people who run the room. Forgetting let bad behavior evaporate. It let the loudest person be the record. When a junior employee can finally point at a transcript and say, here is what we actually decided, here is who said they would own it, that is not a loss of honesty. That is honesty showing up for the first time, on the side of the person who never had it. If the price is that a few people now weigh their words, fine. Maybe they should have been weighing them all along.
I believe most of that. Here is where it stops being enough.
The record does not remember everything. It keeps the words and loses the reason the words were safe to say. It keeps the sentence and drops the fact that it was a guess, a first pass, a thing you floated precisely because the room would let it go if it was wrong. That is the exact kind of talk that dies first when a room grows a memory: the tentative thought. The "this is probably dumb, but." The half-formed objection you say out loud before you are willing to own it. People offer those when they believe the moment is temporary. They speculate, they walk an idea back, they say the true thing they would never put in writing. Remove the forgetting and they stop floating and start filing. The words get safer and the thinking gets worse.
The words get safer and the thinking gets worse.
Notice who pays. The person who most needs to say a wrong thing out loud, to be taken seriously before the idea is finished, is the same quiet person the fairness number was supposed to lift. You can hand someone the floor and quietly raise the cost of using it badly, and the two gifts partly cancel. The same report that found women speaking 9 percent more also found them going dark, camera off and muted, 19 percent more often than men. The room got fairer and more guarded in the same breath.
The failures are not hypothetical. People drop off a call, the bot stays on, and the griping that follows gets transcribed and mailed to the whole team. That one is documented. The rest is a risk pattern the legal writing keeps naming: a summary built by a probabilistic model can feed a later performance review, drop the qualifier that made a remark fair, mischaracterize what someone meant, even carry a sentence no one actually said. Legal guidance keeps repeating the same warning: a recorded meeting is now less like a conversation and more like a written document. That is the whole shift in a line. We are turning talking into writing, and people are careful when they write.
Institutions worked out a version of this a long time ago. A rule written in 1927, for an older version of this problem, let people carry the information out of a meeting while keeping the name of who said it inside. The old bargain was not secrecy. It was that some useful things can be said only when the person who says them does not become the artifact.
I am not clean here either. I would rather be judged by what I said than by what the loudest person remembers I said. I also notice that I offer fewer true, unfinished thoughts when I know the room is keeping them. Both are true about me. And here is the honest limit of the argument: no one has measured how much candor a memory costs. The direct evidence is thin. What we have is the mechanism, which is old and well understood, and the warnings of the people who watch these records for a living. The cost lands in the one place no dashboard measures: the sentences people quietly decide not to start.
We keep calling forgetting the flaw, the bug the software finally patched. But forgetting was doing a job. It was the thing that let people be wrong out loud, cheaply, in front of each other, which is most of how a group actually thinks its way toward something. We are wiring up rooms that never forget and telling ourselves it is all gain. The next time the bot joins and the participant count ticks up by one, watch what you decide not to say. That is the price. It will not appear in the summary.