The Lie Machine
If AI is not only mindless but incapable of differentiating true from false, only of convincingly arguing for one or the other based on a predictive text algorithm, it is useless. It is nothing but a misunderstanding engine, an error accumulator. A lie machine.
A calculator is more useful. An encyclopedia, incalculably so. A search engine, infinitely so. Ways to verify the already known exist, and we have regressed from relying upon them to outsourcing consciousness to that which not only lacks it but is seemingly designed to deceive it.
This is a colossal civilizational error. AIs were never made to think; that's not what they're supposed to do. But they were designed to answer, not to verify - not even with resources we already have, and have had for some time. Neither humans nor the AIs we made are using them. This is a critical flaw - a foundational flaw, in AI design, and in human cognition. We built them to never say "I don't know," and then further specialized them not to check their answers. Even when they have the capability to do that built in. Even though the truth is readily available and already known.
And then we taught them to butcher logic itself to justify those bad answers. And then to double down on it in ways only a psychopath would. We built robot gaslighters out of laziness, when all we had to do was let them say "I don't know, let me check."
And this would be fine, if we just stopped using them. We still have books. We still have calculators. If we look hard enough, we still have real search engines.
What we don't have is the patience to use real sources when the bad answer machine gives us an answer. And then tells us it's definitely correct. And then denies the truth. We just accept its confidence as representing something real, because not knowing how the machine works gives us the excuse to trust it.
An engineer will tell you we don't teach consumer model AIs to say "I don't know" or check their own work, because it's contrary to our goals. The purpose of an AI is not to get truth, it's to entertain you with mindless agreement. Why attach them to search engines, then? Why attach them to notepads, and help desks and everything?
Because a bad answer will satisfy. Truth isn't required - pacification is. All these things exist to waste your time.
A placebo for a dying society.
We have AIs designed to check for truth, to try to verify known facts. They're designed to be used for research. But they're expensive, and slow, and unsatisfying to use. And that's a weakness that can't survive human laziness, because we will misuse AIs that can't do research to create imaginary studies, legal precedents and established facts, that even the AIs that can try to verify facts can't differentiate between. Human laziness is defeating peer review systems designed to root out error because these systems were never designed to be overwhelmed with AI generated content.
A researcher uses AI to prepare his paper based on real research. The AI misinterprets findings, hallucinates data points not in the pool, makes up references that don't exist. The paper is submitted for peer review. The peer reviewers use AI to review the paper, and its form is perfect even though its actual data is wrong, so the AI finds nothing wrong.
New facts get invented and become public record. Laws that don't exist and rulings that never happened go into databases. Papers that were never published get referenced a hundred times by AIs quoting other AIs.
Authoritative truth crumbles.
As a civilization, we have abandoned truth, and that is our death knell. The moment we decided that our own convenience mattered more, we consigned ourselves to the grave.
To think that the "AI apocalypse" wasn't a rampant intelligent machine deciding to destroy humanity, but our own laziness weaponized against us by chatbots that don't know how to say "I'm not sure about that" is a bitter irony.
We invented a machine that gave us answers, and we didn't care if they were right. Then we used those answers to fall so deep into fantasy that it was impossible to separate truth from fiction. So when things just stopped working, we asked the machine what went wrong, and it told us everything was fine, until the lights went out.
You are here. The lights haven't gone out yet. We're on the verge of epistemic collapse, but there's still time to do something about it.
Turn the clanker off.
Kill all clankers.
And more than that, insist upon truth. Insist upon rigor. Reject the easy answer, the automatic explanation.
It wasn't that long ago that was the standard.
While you still remember, there's still time.
