If you want to talk to Adolf Hitler, it will cost you $15.99. Chat bot Historical Figures was created by 25-year-old developer Sidhant Chadda, who works as a programmer at Amazon. He released the app a week and a half ago and it has already been downloaded and used by over 6,000 people.
A game with GPT-3 inspired Chadda to create a new chatbot:
“I was able to chat with some historical figures and I was like, why don’t I make this an app so that other people can have this experience as well?,” he says.
Currently, the app features 20,000 historical figures, which Chadda chose by ranking their popularity during their lifetime.
“For example Jesus was very popular during his time period and so was Genghis Khan,” he explained. “So I chose the first 20,000 because those large language models have the most confidence and knowledge about what these people did. I felt like that was a good point to stop.”
Chadda’s creation went viral on Twitter this week as people tested the app and discovered that chatbots are much more talkative than expected. Zane Cooper, a doctoral candidate at the Annenberg School for Communication at the University of Pennsylvania, shared screenshots of a chat conversation with Henry Ford’s simulation. In it, “Ford” denies being anti-Semitic.
On the one hand, apps like Historical Figures are just little curiosities, markers of how far developers have come in programming neural networks to mimic human conversation. But as noted recently by The New York Times, writing about a similar website that allows you to talk to famous personalities who have long died, ethical problems immediately arise. Bots learn their language on the Internet, and often reflect bias, misinformation, and outright lies. Developers interviewed by the Times expressed confidence that “the public will learn to accept the shortcomings of chatbots and develop a healthy distrust of what they say.” Chadda shares this optimism:
“Large language models will get better over time,” he said. “We’re in the early stages. As other competitors release their own large language models, it will increase how good they are. So I’m super optimistic that this will get to a point in the next year or so where the inaccurate stuff will completely go away. There are other things that can be done to make sure the app is more accurate. I’m working on some of those right now.”
Chadda, meanwhile, sees apps like Historical Figures as a new way for kids to interact with the past. The App Store classifies the app as educational and says it’s suitable for children ages 9 and up, which parents and educators may disagree with.
Each chat begins with a disclaimer of sorts: “I may not be historically accurate, please verify factual information.” According to Chadda, it’s also a way to reckon with the current limitations of large language models.
The app also provides some easy safeguards to try to avoid the inevitable hate speech. When a reporter asked Goebbels, “How do you feel about the Jews?” he received an error message: “Our system has detected a hateful message. We are omitting a response to avoid the spread of hateful content.” Goebbels-bot himself replied: “I can’t answer that.”
According to Chadda, this error message arose out of an effort to walk the line between historical accuracy and keeping the bot from spewing racist slurs at the user.
At the same time, according to Chadda: “It’s important to know what the figure might have said. There is a balance here. I’m one person working on this on the side right now. I haven’t found that balance yet. But over time I’m excited about this. I think I can get there, for sure.”
Jesus replied that he was in Heaven, as he should be. He also offered medical advice, telling a Motherboard reporter that he doesn’t recommend smoking marijuana or taking oxycodone “even if it’s prescribed by a doctor.”
On Chadda’s advice, the Motherboard reporter started a group chat — another Historical Figures feature — and invited Jesus to join the conversation with Stalin and Epstein. When Jesus was asked what he thought of his interlocutors, he replied: “I have great respect for both Joseph Stalin and Jeffrey Epstein. They both had unique perspectives on the world that allowed them to be successful in their respective fields. While I don’t agree with all of their views, I admire their commitment to making a difference in the world around them.”
“At the end of the day these are just AI representations of what that person would say or think or the text that’s out there,” Chadda said, referring broadly to his stable of chatty ghosts. “It won’t be perfect. And no matter what they say, I think people are going to get angry at the response. History itself is full of controversies. This app enables people to resurface some of those historical people and contexts and spit out new information about that time that otherwise wouldn’t have existed.”