For decades, science fiction predicted the singularity: the marvelous event in which AI, our own creation, becomes smarter than us. From The Matrix to Her, AI ramps up slowly but ultimately surpasses our wildest dreams, becoming so much more intelligent than we are that we cannot even begin to comprehend its brilliance.

Since 2023, it has often seemed that we are the fortunate generation getting to experience this transformative moment. With the release of ChatGPT, followed soon after by other chatbots, AI suddenly appeared to know everything: from medical exams and previously unsolved math problems to which outfits would look best with a new haircut.

Sure, AI is still worse than a genuine expert in many fields, but that is temporary. Sure, it has gaps in its knowledge of non-Western cultures and languages, but those will soon be fixed. Every weakness is treated as a temporary limitation. Every impressive capability is interpreted as an early sign of much greater brilliance to come.

Whether or not the singularity is coming, we are already inclined to look at AI as completely trustworthy. So what happens when AI inevitably shares biased ideas with us? What happens when it reinforces stereotypes about minorities, particularly the Jewish people?

It is bad when people voice antisemitic ideas in everyday life. It is worse when those ideas spread across social media, multiplied by hordes of conspiracy theorists. But it is far worse still when the same rhetoric comes from AI associated with limitless knowledge and intelligence. 

We know that people have opinions. But as a species, we have yet to recognize that intelligent machines do too, even as they get smarter. AI has biases, and many of them are not accidental. 

When that seemingly godlike machine shares an antisemitic view or fails to correct a prompt dripping with bias, the prejudice no longer appears merely human. It looks conclusive.

Why does AI have bias? In recent years, the answer to this seemingly perplexing question has become simpler. In large part, AI is biased because the companies that create it choose for it to be so.

You can see this for yourself. Ask ChatGPT, Claude, and Gemini which chatbot is best, and notice how each answer tilts in a different direction. Then give an AI system a sharply critical essay about large language models and watch how readily it tones down the criticism. It’s hardly surprising; of course, these companies act in their own interest. In fact, massive alignment efforts, with unprecedented budgets spent on reinforcement-learning data creation and post-training, allow companies to get AI systems to say whatever they believe will best serve their interests.

As antisemitism has sadly become increasingly normalized in public discourse, we see AI firms responding in ways similar to media outlets and online platforms. Not simply because they are passive products of their training data, but because they have determined that this approach best serves their interests. Some of this intentionality is also evident by the fact that antisemitism doesn’t manifest uniformly in the models. A recent ADL report on antisemitism in AI found that different AI chatbots exhibit different levels of anti-Jewish and anti-Israel bias, with Grok being the worst offender. 

Part of the fight against antisemitism in AI (and against any other form of bias) is spreading awareness that this technology is not a perfect source of information and never will be. We need to debunk the all-knowing oracle fallacy that science fiction has worked so hard to create, and encourage critical engagement with the information output by AI chatbots. Regardless of how far AI has come or will get in the future, information always comes with a point of view. There is no single, objective answer to every question, and the need to engage our own minds will never, and should never, disappear.

A new front for the Jewish people has opened up, and we must work to reduce antisemitism in AI. In parallel, the future of humanity demands that we raise awareness of AI’s inherent limitations and the need to engage with technology critically, rather than put our blind trust in machines. We must let go of the myth of the all-knowing oracle, so that we can benefit from AI while limiting its extensive potential for harm.