Random chat room: a first session, and the question about bots
You think a random chat room is mostly scripts. So did I, for a while. This is an essay about what changed my mind, written slowly and in the order a first session actually goes.
It isn't a rebuttal. Some of that certainty is correct and I'll say where. What I'd argue with is the part that treats the question as settled before anyone's opened the thing.
The first three minutes, and what you arrive expecting
A first session opens, badly, by confirming exactly what you already believe: a permission dialogue, a rectangle that stays black for two seconds, then somebody who's gone before the picture sharpens. Three of those in a row and the conclusion writes itself.
So here's the honest thing to say up front: the first three minutes are the worst evidence there is, and they're also the only evidence most people ever collect.
- The permission prompt takes your first pairing, which never happened as far as the other side is concerned.
- The next two are usually people skipping at the same speed you are, because the queue hands fast movers to fast movers.
- And the first second is when the picture looks worst, which is exactly the second everybody judges on.
None of that proves anyone's real. It does mean a verdict formed in three minutes is a verdict about the interface, and video chat with strangers deserves twenty minutes before you sentence it.
A modest claim, that. I'd rather not make a bigger one than I can hold up.
Osaka, a few minutes after midnight, and a corridor light
An open-plan office with most of the lights off, one person at a desk by the window, and a minute or so into the conversation he leaned back and the corridor behind him lit up on its own. A motion sensor, finding him from ten metres away, then letting the light die again while he sat still.
That corridor light answers the bots question by itself, and I was slow — embarrassingly slow — to notice what I'd actually been deciding on. Not the face, not the reply, not any test I could set. The room had responded to him.
He was printing a letter. He explained that without being asked, and it's the sort of detail no test would ever have produced.
Three in the morning, when you are the good connection
Here's something technical that nobody mentions. Late at night your own side of the line is the best it gets all day: the home connection sits idle, nothing else in the building is streaming, and the contention that makes evenings stuttery has gone.
Which flips the usual asymmetry. For once you're not the one who freezes, and a conversation where you're the stable end runs differently: you interrupt less, you wait less, and the small social penalty of a bad line lands on somebody else.
Where the performance drops, which is the interesting moment
Everybody performs for the first thirty seconds or so. I used to think the point of this format was that nobody bothered — that's wrong. They bother, then it stops, and the stopping is the part worth watching for.
- Wait the opening out rather than trying to break it — it's short and cheap to outlast.
- Watch for the first unscripted thing — a cough, a cat crossing, a word in the wrong language — because that is where it breaks.
- Say something ordinary immediately afterwards. The moment the performance drops is the only reliable opening a mobile video chat gives you.
Past that point the random chat room conversation has either started or it never will, and I've not once seen one recover from being missed.
What the sceptic gets right, which is more than I expected
Four sections of arguing with you, and here's the part where you're right. It's not a small part either.
- The proportion of scripts is real, it's higher at some hours than others, and nobody running these services publishes a number worth anything.
- Any front page claiming moderation is describing an automatic first pass, with a person appearing only after something is reported.
- And a random video chat app on a phone store is a different business from the browser version, with different incentives and a different quantity of automation.
What I wouldn't concede is the conclusion. A format where some share of pairings are machines isn't the same thing as a format that is machines, and that difference gets settled by watching one room for thirty seconds rather than by arguing about percentages.