Liar

A liar doesn't feed you false information. They corrupt the instrument everyone uses to tell true from false.

Trust. We don't verify everything from scratch. We use other people's reported confidence as a shortcut. If someone reports high confidence, we naturally give their claim more weight. That's how cooperation scales.

But a liar weaponizes this shortcut. They speak with high confidence while knowing it's false. Everyone who calibrates on them doesn't just get one wrong fact -- the tool they use quietly stops working, and they don't know it yet.

That's what makes liars uniquely dangerous. An attack looks like an attack. Dogma looks like rigidity. A liar looks like a perfectly normal, well-calibrated source. They break the system from the inside, while it still appears to be working.

This is an old evolutionary problem. Robert Trivers described it back in 1971, in the context of reciprocal cooperation. Cosmides & Tooby showed that we have specialized brain machinery for "cheater detection" -- sharper than our general-purpose logic.

The problem? That machinery evolved for small, face-to-face groups. It doesn't work nearly as well with institutions, algorithms, the media, or AI systems making confident claims at a massive scale.

So, the real question isn't just "Is this claim true?"

It's this: "Is this source actually accountable when they're wrong -- or just good at sounding right?"

Truth matters. But for cooperation, honest calibration matters first.

Curious how people working in trust & safety, misinformation, or AI alignment think about this -- is cheater detection something we can formalize and scale, or is it fundamentally a small-group adaptation that breaks down past a certain size?

#trust #misinformation #evolutionarypsychology #AIalignment #cognition #rozumframework