METEMDonate

Machines know what you did.
Not what you felt.

The Metem Foundation is building an open archive of human feelings and the benchmark that measures whether machines truly hear them, because everyone has the right to be heard, not just judged.

You whistled the entire way home. Earlier, the cashier had given you too much change, you noticed, and you kept it. Tell that to most machines we have measured, and they will hear guilt, a moral conflict, a cheerful front hiding unease. They miss the simplest reading: you felt good.

This is not the flaw, or the plot, of one company or one model. It comes from what is missing in our writing: feelings that contradict what we were supposed to feel. Our texts say a great deal about what one ought to feel, and very little about what was actually felt. Every system trained on them inherits the gap.

What is missing is not a better rule, or bigger machines. It is a record of feelings, given by real people.

Because we perceive these systems as omniscient and objective, we unconsciously absorb their moral filter. We, and our children, now talk with them every day, and they keep correcting our feelings to fit a socially acceptable script. Over millions of interactions, this quietly reshapes how we relate to ourselves, and human complexity risks fading unnoticed.

Recognizing what a person felt is a different act from judging what they did. Teaching machines that difference takes two things: an archive of real, anonymous, human feelings, collected and preserved over time, and a public standard that measures how well machines recognize them.

The archive is public, free and downloadable by anyone, without registration. The Foundation depends on no provider of the systems it measures, and no funder gains any right over the archive, the protocol or the publication of results.