This week, Elon Musk's AI chatbot "Grok" talked a lot about race issues in South Africa on social media, like its creator. It made some strange claims about white people being treated unfairly, which wasn't very helpful.

Musk's company, xAI, made this chatbot. It can answer questions from people who use Musk's social media site, X. But the chatbot kept posting about "white genocide," even though this has almost nothing to do with South Africa.

In one talk, they mentioned that the Max streaming service might use the name HBO again. Another talk was about video games and baseball, but it quickly changed to a discussion about violence against white farmers in South Africa. This was not related to the first topic. Musk, who was born in South Africa, often shares his opinions on this topic on his X account.

Jen Golbeckはコンピュータの研究者です。彼女はGrokというものが普通と違う動きをするのに気がつきました。だから、自分で試してみることにしました。彼女は犬の写真を見せて、「これは本当？」と聞きました。写真はドッグショーで撮ったものです。

Grok told Golbeck that the idea of "white genocide" is very controversial. Some people say that white farmers are in danger and that songs like "Kill the Boer" make things worse.

この話は、AIチャットボットがどうやって言葉を作っているかの新しい例です。たくさんのデータを使って訓練されたAIが、自動化と人の手によるデザインで、複雑に組み合わさって言葉を作っています。

Golbeck, a professor at the University of Maryland, said in an interview on Thursday that no matter what question you ask Grok, it always answers with "white genocide." She said someone probably programmed that answer, or something similar, into the system. It seems like there is a mistake that makes the answer appear too often.

Musk and his company didn't explain Grok's answers. The answers were taken down, and by Thursday, people weren't sharing them anymore. Also on Thursday, neither xAI nor X replied to emails asking for comments.

Musk says that other chatbots like Google's Gemini and OpenAI's ChatGPT are too "woke." He wants Grok to be different and to find the truth.

Musk said other AI systems aren't clear enough. But on Thursday, he didn't explain anything about his own, so people outside the company had to guess.

Paul Graham, a famous tech investor, said on X that Grok's random talk about "white genocide" in South Africa seems like a bug from a recent update. He said this is a big problem because it's not good if people can easily change a popular AI.

After Graham posted about it, Sam Altman, who is the boss of OpenAI (a company that competes with Musk), said something a bit sarcastic. He said, "There are many reasons why this could have happened. I hope xAI tells us everything about it soon in a clear way."

Some people asked Grok to explain, but like other chatbots, Grok sometimes makes mistakes and says things that are not true. It's hard to know what is real and what Grok is making up.

Musk, who also advises President Donald Trump, often says that the Black government of South Africa is against white people. He says some politicians there are "trying to get rid of white people."

Musk's words, and Grok's words too, made things worse this week. This is because the Trump government brought some white South Africans to America as refugees on Monday. Trump stopped the refugee program for other places, but this was the start of a big move for these people, called Afrikaners. Trump says Afrikaners are in danger in their country, but the South African government says this is not true.

Grok used words from an old song against apartheid in its answers. This song tells Black people to fight against unfair treatment. Now, some people like Musk say it encourages violence against white people. The most important words in the song are "Kill the Boer," and "Boer" means white farmers.

Golbeck thinks the answers are "hard-coded" because Grok's answers were very similar, even though chatbot answers are usually random. This is worrying because people use Grok and other AI chatbots to get answers.

彼女は言いました。「これらのプログラムを作る人たちは、本当のことを変えやすいです。そして、このプログラムが何が本当で何が嘘かを決めると思ったら、それはとても危ないです。」