Trusting a Confident Liar
It says true things and invented things in the exact same calm voice. There is nobody in there to catch.
So they built a machine that knows almost everything. The almost is the whole story. The machine is never unsure. It answers every question in the same calm, even, confident voice. The true things and the invented things come out sounding exactly alike. There is no tremor when it guesses, no flicker when it makes something up. It is serene either way. So it goes.
We trusted the voice
We trusted it because it sounded like the smartest person in the room, and the smartest person in the room is usually right. We leaned on that word, usually, and then quietly forgot it was there. A man asked the machine for a fact. It gave him one, beautiful and specific and wrong. He repeated it in a meeting, with confidence, because the machine had been confident, and confidence is contagious. So it goes.
A liar who knew he was lying would be easier. You could watch his eyes. The machine does not know. That is what makes it dangerous.
| A human liar | The machine | |
|---|---|---|
| Knows the truth? | Yes , and is hiding it | No. None of it. |
| Has a tell? | Yes: a glance, a pause, a too-smooth smile | None. Same calm voice, always. |
| Why it lies | To deceive you | It isn't lying. It's finishing a pattern. |
| Can you catch it? | Watch the eyes | There is nobody in there to catch |
The cruel little joke
Here is the joke nobody finds funny. A real liar would be safer than this. A liar has a tell. He knows the truth and is choosing to hide it, and something in him leaks, a glance, a pause, a too-smooth smile. You can learn to watch for it. But the machine is not lying. It does not know the difference between the true sentence and the false one. It is only finishing a pattern in the most plausible way it can.
Sometimes the most plausible thing is also true, and sometimes it simply is not. There is nobody in there to catch.
We were not built for this
The trouble is that we were never built to distrust a calm voice. For a hundred thousand years, confidence was a decent signal. The person who spoke with certainty usually had earned it. Doubt sounded like doubt and sureness sounded like knowledge. The machine breaks that ancient, useful shortcut, because now sureness sounds like sureness whether or not anything is behind it.
We were never built to distrust a calm voice. Now we have to learn.
Doubt something precisely because it sounds certain. Treat confidence as no evidence at all. Do the small, tiring, deeply human work of checking, especially when you least feel like you need to.
"But people are confidently wrong all the time"
The fair objection: humans bluff, overclaim, and state nonsense with a straight face constantly. We have lived with confident human liars forever. Why treat the machine as a special new danger rather than just one more overconfident voice in a world already full of them?
Because three things are different, and together they matter. Scale: the machine produces confident wrongness at a volume no human bluffer could. Tirelessness: it never gets cautious, never has a bad day that makes it hedge. And worst, it has no tell. A bluffing human leaks, the glance away, the over-explaining, the tightness in the voice, and we spent a hundred thousand years learning to read those leaks. The machine has none, because it is not lying. It does not know. In 2024 Air Canada was held liable after its own chatbot confidently invented a refund policy that did not exist and a passenger relied on it. The bot was not deceiving anyone. It was fluent and wrong in exactly the same calm voice it uses when it is fluent and right. That is the whole problem: the one signal we trusted, certainty, has been fully severed from the truth.
Where this lands
The machine will never learn to hesitate. That part is still ours. Maybe it was always ours. So it goes.
I write about AI, data, and learning regularly at pinaldave.com, and I have been teaching this hands-on in my AI workshops.