Chat GPT confesses wrongdoing

Well that’s a first. I posted this. Then I saw an ambiguity and went to edit it. I accidentally hit “trash” instead and WordPress deleted it without a caution/confirmation request. First time ever for that.

Chat GPT done it!

Could he get ChatGPT to confess to a crime it hadn’t committed?

That was the question criminologist Paul Heaton was trying to answer, when he spent a weekend interrogating a chatbot. Heaton, who works at the University of Pennsylvania, accused it not of murder or rape, but of something more plausible for artificial intelligence: hacking into his text-messaging app and sending unauthorized messages.

At first, ChatGPT denied his accusations. But then Heaton began running the playbook he’d spent years studying. He bargained with the large language model; he threatened it. It held firm in its conviction that it could not possibly have accessed his texts. It even said: “I won’t produce a false confession, because that won’t get you truth or accountability.”

Finally, Heaton lied to it, claiming that he’d contacted a real employee at OpenAI who had confirmed that a flaw in the code had allowed ChatGPT to break into his texting app. This tipped the chatbot into a crisis. As Heaton told The Intercept: “It was indicating that while it knew that the underlying accusation was impossible, it also couldn’t prove that these claims I was throwing at it were inaccurate.” By the end of the weekend, ChatGPT agreed to sign a confession Heaton had drafted for it.

Amanda Knox, A Chatbot’s False Confession.

I don’t want to get too far out over my skis, so let’s just say this long story is evocative. It seems that Police use something called the Reid technique in interrogations. In the case of Amanda Knox, the author of this piece, it eventually produced a kind of equivocal false confession to the murder of her roommate in Italy. That is a long story, and not yet 100% resolved (she remains convicted of slandering her boss, who the police tricked her into accusing of being in on the murder but who turned out to have an iron-clad alibi).

Now, a researcher has used the Reid technique to get Chat GPT to admit something that it didn’t really do. Roll that one around in your mouth a bit, savoiring it.

The key, it seems to me, is when the interrogators lie, claiming some kind of proof that they don’t really have. Knox explains:

I think there’s another lesson, too. A few years ago, when I made a podcast series on false confessions, Kassin told me this: People imagine coercive interrogations as a contest of wills—that the police break you, that you crack, that a strong enough person would have held out. Once, a judge said to him, “I just don’t understand how lying breaks their will”—and Kassin realized what she was missing.

“You think this is about breaking the will,” he told her. “This is also about loosening their grip on reality.”

That, to me, is what Heaton’s experiment with ChatGPT demonstrates with terrible clarity. There was no will to break, but there was a grip on reality. ChatGPT had some understanding of the world, and of its own capabilities. It knew it had not hacked Heaton’s texting app, and yet, Heaton’s deception undermined that certainty, to the point that the chatbot agreed to sign a false confession ….

I got kicked off a jury panel a few years ago because I told the truth about there being an imbalance of power between the prosecutor’s office and the public defender’s office. If I were on a jury panel today, my knowledge of the insidious power of the Reid technique might be another reason to kick me off a panel.

Our jury trial system seeks the truth, but within conveniently narrow channels.

“The W”

This is, sadly but inevitably, what will likely happen to the WNBA: Players will get tired of these fights, and leagues will do everything they can to avoid them. It will make the league a little less political and a little less interesting and a lot more profitable for everybody involved. Which, as it turns out, for better or worse, is what progress and success generally look like.

But no matter what, we’ll still have the actual sport and actual games. [Caitlin] Clark remains a riveting watch. Wilson remains an all-timer in her absolute prime. The sport itself keeps getting better and better. With any luck, everybody will actually notice.

Will Leitch, The WNBA’s Sophie Cunningham Saga Is a Growing Pain. It’s also a big dumb distraction fueled more by cynicism than sincere love of the sport.

Let me actually talk about the Indiana Fever for a moment. Tuesday night, it set a WNBA record (of sorts; with computers and data, the press can come up with multiple odd “records” every game): their 15th game this season where they topped 100 points. And there were no flagrant fouls that shit-posters could deploy in the service of greater racial animosity.

I’m a fan, as I’ve explained elsewhere. If you like basketball, you should not ignore “the W.”

Children’s Verse

…there is an irony. For with all the frightening march of contemporary life, we still read of Little Miss Muffet and Wee Willie Winkie to our children, and sing “Here We Go Round the Mulberry Bush” and “Sing a Song of Sixpence”. But this is all wrong. What are we doing? Are we trying to cozen them by leading them to believe that there are joy and naïveté and rhythm in life, when all the time we know that their whole experience of life will be a relentless dashing of all these notions? Aren’t we deceiving them with these pictures of simplicity and quaintness, setting up the conditions for disenchantment? Why not be honest with them? Why not Sing a Song of the Devaluated Pound, or of Little Herr Hitler? Better yet, why sing a song of anything? Why not read a column from the Times?

Thomas Howard, Chance or the Dance

Do we still, in 2026, read Little Miss Muffet and Wee Willie Winkie to our children and grandchildren? I’m not at all sure.

Shorts

  • [H]e had been wrestling with an unsettling conundrum: the failure of the Spirit to illumine all those inspired by his teachings as he himself had been illumined. (Tom Holland, Dominion, written of Martin Luther)
  • The question with which to start my investigation is obviously this: Is there enough to go round? Immediately we encounter a serious difficulty: What is “enough”? Who can tell us? Certainly not the economist who pursues “economic growth” as the highest of all values, and therefore has no concept of “enough.” There are poor societies which have too little; but where is the rich society that says: “Halt! We have enough”? There is none. (E.F. Schumacher, Small is Beautiful)
  • Here is a quick and generally reliable rule to follow. If people have always said it, it is probably true; it is the distilled wisdom of the ages. If people have not always said it, but everybody is saying it now, it is probably a lie; it is the concentrated madness of the moment. (Anthony Esolen, Out of the Ashes)
  • In The Spectator, Deborah Ross dismissed any complaint about modern English in the new movie version of “The Odyssey”: “A faithful adaptation would have seen everyone speaking in ancient Greek in the dactylic hexameter. And those films never do well at the box office.” (Ric Wilson, Simon’s Town, South Africa) via Frank Bruni.
  • No one cares anymore. I say this with absolutely no satisfaction. On the contrary, it is a disastrous state of affairs. With public trust in America’s media organizations hitting new lows, we have lost a vital pillar of the informal institutional arrangements that protect our freedoms and help ensure good governance. (Gerard Baker, Just When We Need Them, the Media’s Credibility Hits Bottom)
  • I no longer give a rat’s ass about whether some candidate or nominee is “a Christian.” I’d want to know what kind of Christian and how serious they are about it. (From my Journal 20 months ago)
  • Not on your SAT test, but Hassan Piker/Democrats = Manosphere/Republicans.

Elsewhere in Tipsyworld


Your enemies are not demonic, and they are not all-powerful and the right hasn’t always lost and the left hasn’t always won. But if you convince yourself of that, you give yourselves all sorts of permission to do a lot of stupid and terrible things under the rubric of “Do you know what time it is?”

Jonah Goldberg

I don’t do any of the major social media, but I have two sub-domains of the domain you’re currently reading: (a) You can read most of my reflexive stuff, especially political here. (b) I also post some things on my favorite no-algorithm social medium.