ChatGPT Isn’t Good at Arguments — Can We Humans Do Better?

Over the last few weeks, the AI chatbot ChatGPT has become a source of wonder, fear, consternation, and humor for its ability to generate reasonable-sounding text given a particular prompt. It can tell stories, explain complex concepts to an elementary school student, and even write divrei Torah.

Yet as is always the case with technology, the output is dependent on the input, and there are real limitations on what ChatGPT puts out. As a few people have noted, it tends to aspire toward a “Kum-ba-ya” feeling and tries very hard to make people happy (even apologizing while still giving an incorrect answer). Perhaps my favorite was when journalist Yair Rosenberg asked it to “Rewrite the debate between Hillel and Shammai over whether to start with lighting one Hanukkah candle or eight as a rap battle in the style of Lin-Manuel Miranda’s Hamilton.” While overall, the AI created something that is quite good, it ended in a way that is almost comically trite: “Let’s put aside our differences and come together / And celebrate Hanukkah with joy and laughter.” There seems to be a preference in ChatGPT for finding commonality and agreement – sometimes at the expense of actual argumentation.

And argumentation is something we need – the challenge is what kind. In the online space, arguments tend to be attacks, then counter-attacks, and then counter-counter-attacks. Perhaps because unregulated AI can reinforce racism and sexism or spew hateful tweets, ChatGPT tries to be a little kinder. That’s the kind of arguing we certainly should try to avoid, and we see this kind of regulation in this week’s portion, Vayigash.

Joseph reveals himself to his brothers after they had sold him into slavery years before. When they arrive in Egypt in the midst of a famine asking for food, they don’t know it’s Joseph, and after he says, “I am your brother, Joseph,” he tells them,

“Now, do not be distressed or reproach yourselves because you sold me hither; it was to save life that God sent me ahead of you…So, it was not you who sent me here, but God–who has made me a father to Pharaoh, lord of all his household, and ruler over the whole land of Egypt.” (Genesis 45:4, 5, 8).

In looking back on his own story, Joseph reframes what had happened to him with his brothers, and as he sends them back to Canaan to bring his father and the rest of his family to Egypt, he tells them, “Do not be quarrelsome on the way.” (45:24)

Why does Joseph tell them not to argue? The great commentator Rashi gives a few explanations, but one was that “…[h]e feared that they might quarrel on the way about his having been sold, arguing one with another. One would say: ‘It was through you he was sold.'” He was worried that they would relitigate the past rather than moving forward. Those arguments wouldn’t be constructive and would instead devolve into the ad hominem attacks we often see in the online space. The Hebrew word for “quarrel” in this week’s portion, tirg’zu, connotes a level of anxiety, which Joseph doesn’t want to encourage.

But arguments aren’t inherently bad. In contrast to the personal accusations Joseph feared between his brothers, arguments can actually be quite helpful and productive. Indeed, Judaism celebrates a machloket – an argument that tries to resolve an issue with both accuracy and compassion (as long as it’s done for the sake of Heaven!).

We sometimes say that people are entitled to their own opinion but not their own facts. But a better phrasing would be that people are entitled to whatever position they can argue for. We need to ask ourselves: why do you hold that view? How would someone else respond to it? How do we move forward? Joseph doesn’t want his brothers to use the past as a way to blame and attack each other, which is appropriate. But at the same time, we shouldn’t paper over real differences and real challenges. And striking that balance is very difficult – especially for artificial intelligence.

ChatGPT is a fun new tool and may indeed change the way we write, read and create. But it can’t (yet) generate new arguments and ideas – it can only repurpose what has already been written. While peace, harmony, and unity may be utopian goals and what ChatGPT tends to display, every person has their own life story and perspective, and every community has its own values and policies, so there will always be arguments. And that’s as it should be in a pluralistic society. The goal will be to follow Joseph’s charge to his brothers – to make sure the arguments aren’t recriminations, allowing us to move forward, not backward, along the way.

Send this to a friend