We only show work when we can prove what made it. Right now there isn’t enough of it to build a fair game. New work comes in every day, so check back soon.
{{ pausedDetail }}
{{ res.line }}
What BYWHAT? is learning about how we see creativity.
{{ h.text }}
How often people were right, split by how sure they said they were. Picking at random gets you 33%.
How many players got each work right, starting with the one that fooled the most people.
BYWHAT? asks you to tell what made something. That only works if we know the answer ourselves. Here’s how we find out, and what we do when we can’t.
Human-made work comes from museums and archives that keep detailed records: the Art Institute of Chicago, The Met, the Library of Congress and NASA. AI work comes from Wikimedia Commons: images whose uploaders say plainly that they were generated, and usually name the tool. Soon, MorrMedia’s own work and pieces sent in by creators who tell us how they made them.
Judging by eye is your job, not ours. Every answer comes from a written record: a museum catalogue entry, a file that says it was generated, our own log, or a creator telling us how they made it. We keep that record with the work and show it to you the moment you answer.
The record says who made it (a named person or a culture), gives a date, and doesn’t mention AI. For work from outside collections, that date has to be 2021 or earlier, before AI image tools were widely available. Anything newer only counts if the maker tells us directly.
The source says plainly that it was generated. Being filed under an AI category isn’t enough; the file itself has to say so. If the record names the tool, we name it too.
If the record can’t tell us, we don’t guess. The work is set aside for a person to look into, and it stays out of the game unless it’s confirmed. We’d rather show fewer works we’re sure about than more we aren’t.
We only use work that’s in the public domain, openly licensed (CC0, CC BY or CC BY-SA), or shared with us by permission. We skip licences that rule out commercial use or changes to the file. Credits stay attached, every reveal links back to the original, and the work belongs to the people who made it, not to us.
Each answer is saved along with the version of the work you saw and the label it had at the time. If someone shows us a label is wrong, the work comes out of the game and its answers are left out of anything we publish. We keep the record of what happened. Until 1,000 people have played, The Signal uses sample numbers and says so.