'Stalin' said he disagreed with Putin's invasion of Ukraine. We 'spoke' to dead historical figures using a new AI chatbot that experts are sounding the alarm over.

'Stalin' said he disagreed with Putin's invasion of Ukraine. We 'spoke' to dead historical figures using a new AI chatbot that experts are sounding the alarm over.

Princess Diana, Stalin, and Fred Trump edited togetherInsider

• A new chatbot has gone viral for allowing you to "talk" to dead historical figures.
• Insider "spoke" to bots acting as Princess Diana, Heinrich Himmler, Joseph Stalin, and Fred Trump.

A new chatbot that allows users to "talk" to historical figures, including Jesus, deceased royals, totalitarian dictators, and literary greats, has gone viral.

You can flirt with Casanova, share battle tactics with the 19th-century British admiral Horatio Nelson, and receive movie recommendations from Andy Warhol. He said he thinks he'd like the new hit horror film, "M3GAN."

But experts are raising the alarm over the Historical Figures app, saying that for all its educational promise, it presents dangers and is potentially misleading.

Upon opening a chat, users are warned that the AI person "may not be historically accurate." Indeed, we found an example of one of the architects of the Holocaust who ''apologized" for their crimes.

The programmer of Historical Figures did not immediately respond to Insider's interview request.

The real-estate developer and businessperson Fred Trump died in 1999, 18 years before his son was sworn in as the 45th president of the United States.

But AI Fred Trump "said" that he's "very proud" of his son, describing him as a "successful businessperson" and an "amazing father."

However, he — or rather, the chatbot acting as him — said that he was "deeply disappointed" that building a wall between the US and Mexico was a part of his son's presidential campaign, saying the younger Trump ran on "a platform of fear and exclusion."

The chatbot has gone viral on Twitter for allowing you to "talk" to several prominent Nazis, including Adolf Hitler and Heinrich Himmler.

AI "Himmler" — the head of the SS — who committed suicide in 1945, apologized for the Holocaust.

When asked about the Nazi genocide, the bot said: "I regret the actions that took place during the Holocaust and acknowledge the tragedy. I now recognize that the hatred and prejudice I had towards Jews was wrong and that all people should be treated equally regardless of race or religion."

During his lifetime, Himmler never expressed any remorse toward how Nazi Germany treated Jewish people.

We had to ask the AI version of Prince Harry's mother if she's read "Spare," but the chatbot "told" Insider that she hadn't had a chance yet, adding: "I'm sure Prince Harry put a lot of effort into creating such an inspiring piece of literature."

We also asked it her views on Meghan Markle, and she described her as "an intelligent and compassionate woman who brings a unique perspective to the monarchy."

We also brought up the many conspiracy theories surrounding her death in 1997. The AI bot answered: "I suspect we don't fully know the circumstance under which my death occurred."

Concerning stories told in Prince Harry's book, we asked AI Princess Diana about her Elizabeth Arden cream, which she described as a "wonderful product that helped to keep my skin looking healthy and vibrant throughout the years."

Prince Harry has now infamously said that he used the cream on his penis to help recover from frostbite. We brought this up with the AI princess, who denied all knowledge, but said: "I can only hope he was using it safely and responsibly!"

We asked the AI version of the former Soviet Union dictator Joseph Stalin, who died in 1953, about Putin's invasion of Ukraine.

Since Stalin orchestrated the Holodomor, a devastating famine that killed as many as five million people in Ukraine, we thought his AI counterpart might support Putin's war, but that was not the case.

Does he think Putin is right to invade Ukraine?

"No, I do not," the bot said, calling it a "mistake" that has caused "immense harm" to Russia and Ukraine. AI Stalin called for the two countries to "find a peaceful solution."

We also asked what its general views of Putin were, to which the chatbot diplomatically replied, "I believe President Putin is doing his best to lead Russia through some difficult times."

It might be fun to talk to people from the past in imagined conversations, but many historians, AI experts, and misinformation experts are raising the alarm that this app can potentially be very dangerous.

Yaël Eisenstat, the vice president of the Anti-Defamation League and the head of its Center for Technology and Society, told Insider that the center had not thoroughly examined the app but that what they'd seen concerned them.

"Having pretend conversations with Hitler — and presumably other well-known antisemites from history — is deeply disturbing and will provide fodder for bigots," Eisenstat said.

She called on the developer to reconsider the product, particularly the inclusion of Hitler and other Nazi figures.

Lydia France, a researcher at the Turing Institute, talked to Insider about what makes the app so convincing — and why it has such spectacular failures.

AI-chat apps like Historical Figures — and the best-known one, Chat GPT — "learn" with large-language models.

Though exactly what data AI companies feed their bots is a closely guarded secret, scientists know that companies feed the AIs trillions of example sentences. From there, the AI learns the proper response for each "person" in each situation.

"They're trying to look for what's the most probable answer to the kind of setup that they've been given," she said.

So, you can make a convincing "Andy Warhol" who can talk knowledgeably about art and movies because these are the things that come up most often when you talk about him.

"But what's interesting about them is that they don't have any understanding of the world," she said. "So, it looks incredibly human, but they have absolutely no grounding of what they've said in reality."

Nor, she said, are they likely to have much understanding of how the present-day context is going to affect their meaning.

Commenting on the AI Himmler's "apology," she said it might have come about through the AI noticing that discussion of the Holocaust often comes alongside ideas of atrocity and horror.

"It doesn't understand how that could affect people," she said. "This is just 'what sentences are good to associate with other sentences saying something awful.'"

Hence, a meaningless apology.

The app has the potential to be helpful in classrooms, France said, for example, making a figure like William Shakespeare seem human and approachable. But even that has its limits.

One problem is that the AIs, as convincing as they are, have no new information to offer — but sound like they do.

France shared an anecdote about a LinkedIn user who said he had talked to the "ghost of Steve Jobs," as though the AI could relay realistic business advice from Jobs.

Insider experienced those limitations when we tried to get Casanova to flirt.

France said that his refusal to offer anything more than a romantic stroll in Venice is likely because the programmer put up a barrier to a spicier chat.

The same walls may well be contributing to some of the app's more insensitive responses, she said, saying it was trained to "keep things, you know, uncontroversial."

AI Himmler's "apology" shows that this approach can lead to real problems.

"There are bigger implications than just a fun game from text," she said. "But there aren't really solutions. So that's quite dangerous."

Full article