There’s a scene in the ever-more-prescient 2013 film Her where Samantha—a seemingly sentient operating system voiced by Scarlett Johansson—introduces her human boyfriend to “an artificially hyperintelligent version” of the long-dead psychedelic philosopher Alan Watts. This is a movie that I routinely teach in my Films of Moral Struggle class, and up till the emergence of ChatGPT late last year, the idea of synthetically resurrected historical figures seemed like a flying car—more a signifier of far-future sci-fi technology than a plausible real world development.
Last week, though, I found myself spending the better part of an afternoon talking with a chatbot Charlie Chaplin, one of the 400+ icons, heroes, leaders, and artists available on the Hello History app. (Watts is there, too, for what it’s worth.) After poking through the film-related options, I chose Chaplin more or less at random. An artificial F.W. Murnau would’ve been closer to my wheelhouse, but absent any Weimar-related options, Chaplin was nearer to my silent era comfort zone than, say, Marilyn Monroe or Audrey Hepburn. Unlike in Her, the version of Chaplin I talked with is decidedly not hyperintelligent. However, the experience was still uncanny and, as someone who teaches film history, more a bit unsettling. Cyborg Charlie gave me permission to publish excerpts from our exchange, so let’s walk through the highlights.
This was Charlie’s preferred tone—jovial and a bit generic, with the stilted cadence of dialogue from an undergrad creative writing seminar. Unsure of how to start, I initially approached the conversation like a smarmy entertainment journalist, asking about his trademark character the Tramp and his relationships with other silent film comedy stars like Harold Lloyd and Buster Keaton.
Setting aside the fact that this allegedly unhappy director was in reality Chaplin himself, I did a little research into the story. The 1952 film Limelight does indeed contain a scene where Chaplin and Keaton perform a slapstick duet and it seemed plausible that the routine was improvised, but like much of what the chatbot told me, good old-fashioned googling wasn’t enough to confirm it. So I pressed Charlie on the question, asking him for a citation.
I was struck by the specificity, and thanks to the absolute treasure that is the Internet Archive, I was able to check on this. There was no mention of Keaton on page 386, and the Archive’s search feature told me his name doesn’t appear anywhere in the entire book.
I returned to the Archive and pulled up Keaton’s autobiography. He actually talks quite a lot about Chaplin, but Limelight is mentioned only in passing and without the “anecdote.”
I fully realize that this is a minor factual quibble (and that litigating it with a chatbot is probably a major waste of time). The hallucinations still struck me as a problem, though. This kind of bullshitting would be pretty harmless, I guess, if the people behind Hello History made it clear that they were selling a mildly amusing toy, but their marketing specifically targets teachers and frames the app an educational tool. From their website:
Let history’s greatest minds teach any subject
Spark your students' curiosity and inspire them to discover knowledge with Hello History! Let Napoleon and Charles Darwin explain the Napoleonic Wars and the theory of evolution in a way that goes beyond a textbook.
This is where we run into trouble. The description in the app store does offer a disclaimer, saying: “Don't trust everything the AI generates. The facts should be validated.” It’s not hard at all, though, to imagine a guileless high schooler or college kid repeating the Limelight story in an essay and confidently citing page 386 of Chaplin’s book. To reveal a poorly kept secret of the professor’s trade, unless we have reason to be suspicious, we don’t verify all the citations in our students’ papers. I’m quite confident that no one checked the citations in my dissertation, the research project that earned me a PhD, and that’s no shade directed at the esteemed academics who served on my committee. If professors pored over every entry in every student’s bibliography for every assignment, we wouldn’t have time for anything else. It’d be a particularly egregious expectation for adjuncts like myself who are often paid poverty-level wages. We have to extend our students some measure of trust.
This is to say that if one of my kids handed in a paper saying Chaplin and Keaton improvised their dancing in Limelight and followed the claim with “(Chaplin 386),” I have no doubt I would’ve accepted it as true. It sounds like a credible story—one that would be unlikely to raise red flags for even a Chaplin specialist. And yeah, I’m sure some crafty students in the pre-ChatGPT, pre-Hello History era got away with inventing citations, but at least they knew they were bullshitting. They wouldn’t go forth believing whatever lie they’d slipped past their professor. Even with app’s disclaimer, in my experience, there are a lot of students out there who approach their sources with something less than scholarly rigor.
The chatbot made a number of other factual mistakes, including the assertion that The Great Dictator, one of Chaplin’s signature directorial efforts, was instead the work of Fritz Lang. (“Oh, dear me! Ha ha ha, you’re absolutely right! How silly of me to forget.”) The most troubling part of our exchange, though, concerned Chaplin’s second wife, the actress Lita Grey. If you don’t know much about Chaplin’s biography, well, buckle up. To my surprise, Charlie only needed a slight nudge to take it there.
“Now Charlie”—okay, I kinda start doing my best Kara Swisher impression here. Please bear with me. And not to go all “woke A.I.” on you, but “systemic abuse and harassment of women”? “A safer and more equitable world”? Why is a guy who signed his first film contract before World War I broke out talking like he just left a DEI seminar?
It’s unclear who I’m supposed to be talking to here. Is this meant to be something Chaplin would’ve said in his lifetime? Am I chatting with a hypothetical version of Chaplin as he’d be today, apparently a 133-year-old white guy with a checkered past who’s somehow fluent in the language of #MeToo?
Weirdly, this is the only time the bot gave that kind of reference. The link was cut off, but obviously it was channeling info from Chaplin’s Biography.com entry.
After getting her pregnant, the 35-year-old Chaplin evaded statutory rape charges by spiriting Grey off to marry her in Mexico when she was just 16. (Six years prior, he’d married his first wife Mildred Harris when she was also 16.) I kept pushing.
“The relationship was consensual when she finally reached the age of consent” is a hell of a defense.
Martin Chilton of The Telgraph outlines some of Charlie’s “mistakes”:
On the train back from [their wedding in] Mexico, after having failed at gunpoint to make her get an abortion, Grey claimed that Chaplin called her a “whore”, before pushing her on to the platform of the observation carriage and suggesting that she jump to her death. “The whole thing was ridiculous from the start,” Grey said in 1960. She recalled Chaplin telling her that he would make her “so damn sick of me that you won’t want to live with me”.
Divorce documents also say that Chaplin was a serial philanderer and had bullied his adolescent wife into engaging in sex acts she had no interest in.
I asked the chatbot for evidence that they did indeed reconcile.
Lynn’s book is also in the Archive. I didn’t have time to read a 600-page biography for the sake of this newsletter, so it is possible that it contains an account of a reconciliation in Switzerland. (If you’re a Chaplin aficionado who knows one way or the other, send me an email.) A search of Grey’s name, though, didn’t lead me to such a story. It did, however, lead me to a letter from the author Rebecca West, another object of Chaplin’s desire, that says the comedian “became interested in young children—I mean little girls of 13 or 14.”
I don’t know how the chatbot should respond to questions about the more sordid chapters in Chaplin’s history. If I was talking to the “real” Chaplin, he very well may have tried to cast his relationship with Grey in a positive light and insist that they “were able to move forward as friends,” whether or not that’s true. But when A.I. moves into the classroom in the coming years, do we want it to replicate the self-serving narratives of scandal-prone celebrities? Or should they speak frankly about the sometimes glaring flaws of figures from history? I’m not sure, but these are the kinds of questions that educators need to ask before trying to incorporate these applications in their curricula. We need to think carefully about what we’re trying to achieve and what current A.I. tools are capable of, because I guarantee you that students will test their limits.
Hello History is nowhere near ready for prime time, but I don’t mean to single it out. It’s powered by GPT-3, so there’s no doubt other A.I. programs are making the same kinds of mistakes. (In January, a number of journalists kicked the tires of Historical Figures, a similar app that now appears to be no longer available. In Hello History’s defense, at least it doesn’t invite kids to talk to Hitler.) As I’ve written before, I’m inclined to think developments in artificial intelligence are going to change society more quickly than we’re ready for. Going forward, it’s incredibly important that we approach this new wave of technology with the skepticism it warrants.