Judaism and AI Design Ethics part 1

by R. Gil Student

Artificial intelligence (AI) has quickly become a part of daily life, influencing the information we consume and the decisions we make. And the process is just starting. This places significant responsibility on the AI builder. Designing an AI system is not merely a technical challenge but also a moral and religious one. What information is included, how it is presented and what assumptions shape its worldview all affect the end user. AI is a broad term and we speak here of any AI system that provides information or recommendations to the public, even if this is just a small subset of AI that is already integrated into systems. Judaism has long wrestled with analogous challenges, especially in the realm of publishing, where books and ideas shaped communities and beliefs. The precedents we find in halakhic literature offer guidance on the ethical responsibilities of those building AI systems today.

I. Book Publishing

When you boil the issues down to their basics, in a sense AI systems resemble book publishers. They gather, process and distribute information, often with little distinction between fact and opinion, or between traditional and secular perspectives. Of course, there are differences. Publishers determine the actual words used while AI systems have more independence in expressing ideas. However, the similarities are important. The dangers are obvious: inaccuracies can harm reputations, mislead the public and cause damage to individuals, groups or institutions. Additionally, the dissemination of a secular worldview can significantly undermine religious convictions. Judaism has a lot to say on these subjects.

But a fundamental question arises: who is the judge? Many issues cannot be conclusively proven. What counts as heretical, misleading or damaging? Who decides what is acceptable and what must be avoided? These questions, which arose in the age of the printing press, return with new urgency in the age of artificial intelligence. 

There are two ways to approach the ethical dangers of information technology: as policymakers and as citizens. Policymakers can regulate markets and restrict harmful products. Citizens, lacking that power, must find other ways to protect themselves and their communities. Halakhah addresses publishing issues from both perspectives, which can inform our discussion of AI ethics.

II. Improper Content

AI systems, even the most advanced, can generate errors. However, this is not a new challenge. Authors can include mistakes and misinformation in books, newspapers and magazines.

The Torah demands reliability. The Sages teach, chazakah she-ein chaver motzi mi-yado davar she-eino mesukan, it is assumed that a scholar does not release something that is defective and unreliable (Eruvin 32a). Your product, your words, your teaching must be accurate and responsible. This principle applies no less to an AI builder than to an author or teacher. If you release a system that frequently misinforms, you have failed the Torah standards expected of you. You might also be violating prohibitions against slander (lashon ha-ra) against individuals, groups and institutions. AI builders bear an ethical duty to ensure accuracy, reduce harm and constantly refine systems to prevent the spread of falsehoods.

But inaccuracies are not the only danger. AI can spread not only errors but also perspectives foreign and contradictory to Torah. By default, most AI systems are trained on vast libraries of secular writing, much of which reflects assumptions inconsistent with Jewish tradition. Some of these relate to unacceptable social behaviors and others relate to fundamental Torah beliefs. Presenting such perspectives as neutral fact and normative behavior and beliefs is spiritually dangerous. Books, likewise, present similar challenges.

III. Jewish Approaches to Regulating Publishing

How have Jews historically dealt with similar challenges? There are two possible perspectives: policymakers and citizens. As mentioned above, policymakers wield control and can regulate markets. But for most of Jewish history, Jews lacked such power. Indeed, Jews often utilized Christian book publishers. Instead, Jewish communities had to assert religious responsibility as citizens, finding creative ways to protect their members without market control.

Given that Jewish publishing houses have existed for centuries, it is surprising how few responsa have been published about their ethical responsibilities to the public. There is one mention of book publishers in Shulchan Arukh (Orach Chaim 307:16) declaring that the publishers of romance novels cause people to sin by thinking improper thoughts. In the 1970s, Rav Moshe Feinstein addressed the case of publishing heretical works. He famously insists that the commentary of R. Yehudah He-Chassid on the Torah is a heretical forgery. Significantly for our purposes, Rav Feinstein rules that it is forbidden for a Jewish publisher to print heresy. More strikingly, he adds that even if the overt heretical passages are removed, the publisher may not publish the rest of the work which might still contain confusing or misleading ideas. Even subtly non-traditional ideas are forbidden (Iggeros Moshe, Yoreh De’ah, no. 115).

In the AI context, this is particularly pressing. A model that offers secular or non-traditional interpretations of morality, halakhah or faith can easily mislead the unwary. The risk is not only false information but distorted frameworks of thought. AI builders must ask: what perspectives are we embedding? What worldview does the system normalize? Policymakers must consider: what perspectives can we, as a society, tolerate and what can we not? How do we enforce minimal standards to prevent dangerous views from proliferating? The first step is generating agreement that there should be minimal standards. The second step is deciding what they are. Neither step is easy. 

Even when the information comes from a reputable source, it might be improper to provide to the public. For example, the Talmud (Shabbos 30b) discusses whether certain biblical books should have been removed from circulation. There was no doubt that they were written under divine inspiration. The problem was their confusing and contradictory natures. If the objectionable passages could be explained, then there would be a basis to allow their circulation. However, responsible authorities cannot allow the circulation of a theologically confusing and misleading book, even one written under divine inspiration.

I remember when Tipper Gore led the fight against violent and profane lyrics in music. To society’s great detriment, her team’s partial win consisted only of labeling such music as explicit and nothing beyond. In my opinion, AI builders are ethically bound to ensure that AI avoids violent, profane and otherwise destructive output. And regulators are ethically bound to ensure that unethical AI systems do not enter society. However, even if this fight is won in the US, unethical AI systems will certainly be built in other countries that do not regulate their technology. Perhaps this is overly pessimistic, but it seems almost impossible to prevent those AI systems from being used in the US. In other words, no one really controls the markets. Therefore, we need to look at another model for responsible publishing.

To be continued…