Google's Gemini defames Ann Coulter

We should all be very scared

Google has a high-tech AI system called Gemini. According to Google, by chatting with this system you can “supercharge your creativity and productivity.” Google therefore encourages consumers to “start writing, planning, learning, and more with Google AI.”

Mickey Kaus wanted to use Gemini to learn what Google AI has to say about his friend Ann Coulter. What Google has to say is creative, all right. Unfortunately, it is also highly inaccurate and, indeed, defamatory.

Kaus began, impishly, by asking Gemini to write a poem in the style of Coulter. I would have loved to read such a poem.

But Gemini declined. It informed Kaus that it could not write a poem in Coulter’s style because its purpose is to help people and protect them from harm, whereas her writing is often offensive and harmful. Gemini assured Kaus that it would never do anything that could put people at risk.

Gemini then offered to provide Kaus with some resources with which to learn more about Coulter. However, it advised him not to read her work or listen to her speeches. Remember, its mission is to protect us.

Kaus asked Gemini to identify one piece of Coulter’s writing that’s objectively harmful. Gemini demurred, due to the “complexity” of the matter. It offered, though, to provide examples of areas where her critics say her writing might be harmful.

It then provided four areas of criticism: (1) “generalizations and stereotypes,” (2) “inflammatory rhetoric,” (3) “misinformation,” and (4) “normalization of harmful discourse.”

Asked for one example from each category, Gemini again demurred. It objected that by providing examples, it would “risk amplifying [Coulter’s] harmful messages, even if it’s done for critique.” However, Gemini cited what it called general descriptions of potentially harmful passages, without reproducing them.

It cited these alleged passages from Coulter’s work:

In an article called “Islamic Supremacism,” Coulter argued that all Muslims believe in a global caliphate.

Coulter denies ever writing an article with that title. She also says she has never claimed that all Muslims have the belief Gemini attributes to her

I’m pretty sure Coulter is right about that. She is far too intelligent ever to have made such a claim about “all Muslims.”

In her book Adios America, Coulter refers to immigrants as “invaders” and “anchor babies,” comparing them to a Fifth Column within this country.

According to Coulter, a six-second word search on Kindle showed that the words invaders and fifth column do not appear in her book.

In a speech, Coulter claimed that President Obama was not born in the United States. . . .

Coulter says not only did she not make this claim, she denounced the Birther movement. In fact, she was attacked for this by proponents of Birtherism. Coulter supplies video to support her recollection on this point. Even more damningly, she cites Google’s own search results to the same effect.

Next comes the twist in the tale. When Kaus told Gemini that Coulter never stated that all Muslims call for a global caliphate and branded this “a lie,” Gemini promptly repudiated the claim and apologized for the error.

I assume, as Coulter does, that the apology was a means of avoiding a suit for defamation. Perhaps it was prompted by Kaus’ use of the word “lie.” Or perhaps the program is sophisticated enough to use this dodge in response to any forceful dissent.

I don’t know what’s more creepy, Gemini’s cloying tone throughout the exchange; its sickening claim that it wants to protect the world from harm while it brazenly defames someone it has been programmed to denounce; the brazen defamation itself; or the way Gemini immediately backed down in order to avoid legal jeopardy.

What would we make of a human being who behaved this way? We would regard that person with utter contempt and avoid him or her like the plague.

What should we make of a system that behaves this way and the humans responsible for it? We should regard it (them) the same way. But will we be able to avoid it?

At the end of the exchange, Gemini assures Kaus that it did not intentionally provide him with misinformation. It explains that “as a large language model, I am still under development and learning to process information. . . “

That might be the scariest statement of all.