AI and the Paradox of Trust
9th July 2026
Here is a paradox, perhaps.
Information Theory tells us that only surprise counts as information. If the receiver can infallibly predict the next signal from the sender, that predictable signal adds no information.
If I transmit “Mr.”, none of the letters in the group “i, s, t, e, r” count as information. We trust that those letters come next.
Trust is reduced uncertainty about future behavior. But trust is generated largely by past behavior.
When we trust a bridge, we expect it to continue standing. When we trust a bank, we expect it to honor withdrawals as it has done in the past. When we trust a person, we expect him or her to behave tomorrow roughly as they behaved yesterday.
We are trusting in the past. We are trusting in reputation.
Reputation is accumulated evidence from the past about future behavior.
Even if, as is usually the case, our trust is in an expert rather than our own experience, we trust because other persons or institutions have affirmed the expert’s reputation—his history of being right or wrong.
Now here’s a wrinkle.
Suppose we have strong reason to trust an expert within a particular field. Then one day we encounter him discoursing on dozens of subjects far outside his area of expertise. Every proposition sounds plausible. Better still, many of them are entirely new to us.
He is providing enormous amounts of information. Do we believe him?
If he offers a thousand plausible assertions, do we accept them?
This is the AI problem.
Full disclosure: Rich is a friend and was a classmate at Yale. He has worked with George Gilder and written for National Review.