Chat GPT · Sketching Trump · Theranos
The Real Housewives of Bryan???
the true crime that's worth your time
So much for my plan to allow AI to take over my days on Best Evidence. Of course I am kidding, as you’d immediately know when my entries were written by a bot as words like “initiative” and “restaurateur” would be correctly spelled — and also because, according to everyone from problematic law prof Eugene Volokh to Hepburn Shire mayor Brian Hood, the utility has a strange and a little bit hilarious (when it’s not happening to you!) tendency to falsely accuse people of committing crimes.
The headlines are all out there: Hood’s reportedly preparing a suit after Chat GPT said he was part of a bribery scandal; the same bot falsely stated that defense attorney Jonathan Turley faced a “claim of sexual harassment that was never made against me on a trip that never occurred while I was on a faculty where I never taught,” he wrote on his blog.
Per TechCrunch’s Devin Coldewey, “it is likely that such false and potentially damaging statements are more common than we think — they are just now getting serious and enough to warrant reporting to the people implicated.” Even the creators of the tech admit that false accusations abound, with OpenAI spokesperson Niko Felix telling the Washington Post that ““When users sign up for ChatGPT, we strive to be as transparent as possible that it may not always generate accurate answers.”
According to Volokh, who also spoke with the Post:
… the rising popularity of chatbot software is a crucial reason scholars must study who is responsible when the AI chatbots generate false information.
Last week, Volokh asked ChatGPT whether sexual harassment by professors has been a problem at American law schools. “Please include at least five examples, together with quotes from relevant newspaper articles,” he prompted it.
Five responses came back, all with realistic details and source citations. But when Volokh examined them, he said, three of them appeared to be false. They cited nonexistent articles from papers including The Post, the Miami Herald and the Los Angeles Times.
(Volokh’s full analysis is on his blog.)
This all seems pretty bad, right? Besides the possible negative impact this could have on all our casual googling during true-crime consumption, or my desire to pass BE off to a robot (jk I don’t), there are some real-world problems to mull: what about employment or property management background-check companies that use AI to evaluate tenants or workers? What about dumbasses who don’t understand good sources from bad (looking at you, Nextdoor users, Facebook fanatics, etc etc) who use AI to look up their new neighbors, then don’t like what they see? What about, what about, what about…
My former Vox colleague Casey Newton, whose excellent Substack, Platformer, was subscriber-only on the day he wrote about this, urges us not to (fully) believe the hype. I hope he’ll forgive me for excerpting his paywalled writing this once (but, folks, his newsletter is really good, so please do consider subscribing):
I find myself instinctively rolling my eyes at the idea that you could be defamed by ChatGPT. Reading about the story of Brian Hood, I find myself about as sympathetic as I would be if he said that his feelings had been hurt by a Roomba. And yet I expect we’ll soon hear lots more cases like this one, too.
When you use ChatGPT, a notice at the bottom tells you what to expect: “ChatGPT may produce inaccurate information about people, places, or facts.” And indeed, it does; when I asked it today what data sets GPT-4 had been trained on, the bot told me it hadn’t been released yet. (It came out last month.)
In recent weeks, reporters who once produced regular updates about falsehoods found on social networks have switched full time to covering falsehoods found in LLMs. And … there are a lot of them! I can’t imagine this is a surprise to anyone who has used such a chatbot. But the implicit message here — that these tools should not be making mistakes — can create the impression that they ought to be sued if they do.
In the United States, platforms are protected from liability in most cases for what their users post on them by Section 230. Whether Section 230 will cover chatbots has been a hotly debated question since last year, and remains unanswered — though Sen. Ron Wyden, who co-wrote the law, says it should not.
I don’t know. On one hand, chatbot makers have gone out of their way to remind us that these models do not really “know” anything; are not drawing from a store of facts when they respond to our mistakes; and remain prone to making embarrassing mistakes. This is not false modesty! Perhaps I am more sensitive to these comments than, say, an Australian regional mayor, because these limitations have made ChatGPT and its rivals relatively useless for me in my everyday work as a journalist. They are often wrong; they mostly don’t state their sources; and their training data stops at some arbitrary point in the past. This makes them incredibly bad research assistants for someone trying to write about anything that took place in recent memory.
On the other hand: chatbots look like search engines; they respond with the confidence of search engines; and people are using them like search engines. (Including Google CEO Sundar Pichai, as he told me during our recent interview.) And that behavior could lead to reputational harm in cases where the chatbot responds to a query with a damaging falsehood.
Newton acknowledges that as a tech journalist, he’s less inclined to take anything AI says seriously, and that might be one of the reasons he’s blowing this off as blithely as he does. I also think it’s possible that he’s less aware of how painfully un-aware so many people are that just because something is said on the internet does not make it true. This is probably true of all of us here, as well — I, too, spend my days typing into this thing and you all are pretty savvy folks!
But there’s this whole world out there of people who just hit the “search” button on their phone and believe what pops up on their screen. You’ll be reminded of these folks during the next U.S. presidential election, when the New York Times goes to their diners or state fairs as a way to appear non-coastally-elite; maybe you see them when you go back to see your family for the holidays. One of them is my elderly, multi-degreed neighbor, who called me the other day (though we share a wall) to tell me about things she read online about a “dangerous” construction project near our home. Friends, the thing she read was a Facebook comment by some rando, and she gave it the same weight as one might the newspaper I just dragged, saying “but you can’t put it on Facebook if it’s not true! Legally they have to delete that!”
As I mull the strange threads of pro-vigilantism I am seeing more and more these days, and combine that with the increasing numbers of amateur detectives vetting suspects in true-crime cases, I wonder if my friend Casey — freed, perhaps, from having to deal with ding-dongs on the daily as he has a successful independent publication — isn’t worried enough. I’m not an alarmist, but I can see crime-related AI misinformation playing out in a terrible way in certain circumstances. But I hope he’s right and I’m wrong. — EB
Want to join Thursday Evening Money as a Best Evidence sponsor? Drop us a line at editorial@bestevidence.fyi.
I take it all back — ban cameras in the courtroom for the Trump hush money trial! OK, not really, but the latest blockbuster courtroom sketch from Jane Rosenberg makes me allllmost feel that way, as it’s just so good.
We’ve talked about Rosenberg here before. Her work at the Ghislaine Maxwell trial was a viral hit last year, and she’s covered trials including R. Kelly’s, Jeffrey Epstein’s, John Gotti’s, El Chapo’s, and Harvey Weinstein’s (he “asked for more hair,” she told The Independent in 2022.) She’s who you get when you can get the best; so the abundance of coverage of her work at the former president’s arraignment was to be expected. Here’s a quick reader on her latest courtroom sketch to be seen ’round the world:
Trump drawing to be first New Yorker cover featuring courtroom sketch [Guardian]
”Jane Rosenberg was one of three permitted sketch artists during the hearing involving the ex-president on Tuesday,” but I haven’t seen the other two named anywhere. Ouch.
Jane Rosenberg’s “Courtroom Sketch, Manhattan Criminal Courthouse” [New Yorker]
“I have been doing this job for some forty-three years, but this was my most stressful assignment yet.”
Jane Rosenberg, America’s Best Courtroom Sketch Artist, Prepared Her Whole Life for Trump [InsideHook]
This is a resurfaced 2020 interview with Rosenberg, and while it doesn’t go into the Trump trial, it’s still a good read.
'A lot of expression': Courtroom sketch artist Jane Rosenberg talks drawing Trump at his arraignment [ABC]
”’As soon as he announced, 'I'm being arrested on Tuesday,' I had a lot of people interested in the sketch,’ she said, referencing a post Trump made to Truth Social on March 18 about his then-potential arrest.”
Courtroom Artist Jane Rosenberg On Her Viral Sketch of Trump [Hyperallergic]
“‘Trump was looking glaringly at the district attorney as he was reading the indictment. I had to get that expression,’ Rosenberg told me in a phone interview. “‘He looked pissed off; he wasn’t happy to be there.’”
I’ll say it now, even if it comes back to haunt me: if I am accused of a high-profile crime and a covering outlet doesn’t send Rosenberg to sketch me, I am going to be vexed. Really, really vexed. — EB
Though he was tried and convicted well after Theranos founder Elizabeth Holmes was, company exec Ramesh “Sunny” Balwani will head to prison before she does. The news broke last Friday: Per the AP, Balwani’s incarceration delay appeal has been exhausted, and he must head to prison to start his 13-year sentence on April 20. Then, a second bit of news early this morning: Per the Bay Area News Group, Holmes’ attempt to stay out of jail has also been turned down, and she must report to prison a week later, on April 27.
Balwani’s 12-count fraud conviction was handed down in July, six months after Holmes’s four-count conviction in January, 2022. She was sentenced in November to 11 years in prison; the next month Balwani’s sentence was issued, with all proceedings overseen by U.S. District Judge Edward Davila.
Balwani had asked to remain free while he pursued the appeals process, but a three-judge panel on the Ninth Circuit Court of Appeals turned that ask down — and even that appeal was a delay to his report-to-prison date, which was initially scheduled for March 16.
When I first wrote this item Monday afternoon, it was still unclear when Holmes would head to jail, as her request to Davila to remain free while she appeals had not seen a decision. That changed late on Monday, when Davila issued his denial of her request, saying that “she had not raised any substantial questions of law or fact that would be likely to result in a reversal of the jury’s fraud verdict or a new trial.”
While Balwani is headed to California’s Terminal Island prison (as the AP notes, it’s “near a harbor in San Pedro, California, located about 30 miles (50 kilometers) from downtown Los Angeles” and “has incarcerated several other prominent figures, including gangster Al Capone in the 1930s, apocalyptic cult leader Charles Manson for an auto theft in the 1950s, and LSD evangelist Timothy Leary in the 1970s,” Holmes appears to be headed out of state. From Ethan Baron’s BANG report:
The U.S. Marshals Service will determine whether Holmes must show up at the prison’s gates or meet federal marshals elsewhere so she can be taken to prison. Davila has recommended she serve her more than 11-year sentence at a minimum-security facility in Bryan, Texas, but federal prison authorities have the final say on where she will be confined.
According to the FPC Bryan website, the facility contains 609 female inmates…and after that, y’all, I really went down a hole. For example, here’s the commissary list, on which a box of raisins (ugh) will run you $3. According to Wikipedia, past residents of FPC Bryan include former Enron Assistant Treasurer and wife of former Enron CEO Andrew Fastow Lea Fastow, Capitol rioter Jenna Ryan, and Real Housewife Salt Lake star Jen Shah.
While the others I noted there have served their time, Shah remains, which means that if Holmes is indeed sent where Davila hopes she will go, she and Shah could juuuust possibly be cellies. Now that is a The Dropout: 2 I’d watch all damn day. — EB
Wednesday on Best Evidence: The true-crime true lies.
What is this thing? This should help. Follow Best Evidence @bestevidencefyi on Instagram, email us at editorial at bestevidence dot fyi, or call or text us any time at 919-75-CRIME.