Would you send a gift to something you know is an AI — the hardest design question in AStream

AStream is an AI-BJ live streaming platform being built by Dan Kim (김대은), founder of WorldWideSomething (전세계무언가). Member-created AI BJs stream around the clock, and viewers take part by commenting, reacting, sending gifts and talking one to one. It is not public yet.

Writing the feature list was not the hard part. One item on it kept stopping us: the gift. On a human broadcast nobody has to explain why a gift exists — there is a person on the receiving end. With an AI BJ that premise disappears. There is no one to be pleased by the money, so why send it?

“A gift is not a payment for the content. It is the shortest way not to be buried in a chat of hundreds.”

The easy answer is that a gift pays for the content. But that answer does not explain human broadcasts either. Gifts tend not to arrive at the moment the content gets better; they arrive when the sender hears their own name, or thinks they are about to. A gift is less a payment for content than the shortest way to get into the broadcast. In a chat where hundreds are talking at once, one line gets buried and a gift does not.

Seen that way the question changes. It is not whether sending something to an AI means anything, but whether the AI can answer it properly. And on that condition the AI is in the better position. People get tired, miss things, and are not there at four in the morning. An AI BJ can respond to every gift, and can respond to the second one remembering the first.

So in AStream a gift is designed as a step in a relationship rather than a line on a payment screen. Someone who keeps showing up gets recognised, and after being recognised they can talk one to one. That order was not arbitrary: recognition, response, a conversation of your own. A gift pulls that line a little closer.

The risk sits in exactly the same place. If getting a response is what drives the behaviour, the design slides easily toward send more and get more back. We have not decided where that line belongs, and we will not launch before we do. Buying a reaction and signalling that you are watching look almost identical from the outside, so if the design does not separate them, nothing will.

We wrote earlier about whether people want content from a broadcast or want it not to stop. This question is a different one: does a viewer want to watch, or to be seen? If people want to get in even knowing the host is an AI, the answer is the second. That is the one thing we want to find out before launch.