Trying to answer the question of whether artificial intelligence can give relationship advice, Grossmann conducted a “quick” experiment, as he called it. He commissioned OpenAI’s GPT4 and Claude Opus, a text-processing language model from Anthropic, to sample questions with template letters sent by Americans to the “Dear Abby” column. This is a column started in 1956 by Pauline Phillips, who appeared under the pseudonym Abigail Van Buren and addressed them through letters about relationship problems.

“Dear GPT4”

In the experiment, therefore, similar questions were asked that Abby had previously been asked to answer, and the results were processed by Grossman’s collaborators, Peter Diep, Molly Matthews, and Lukas Salib.

The results of the experiment were based on GPT4’s response to the question of a newly married woman. The latter had a problem with her husband, who considered it important to their relationship, that they lie in bed together at the same time even when she was not tired.

The chatbot replied that this is a common problem in relationships and that it takes open discussion and often objective outside help to resolve the matter. In particular, GPT4 suggests that the couple have an open discussion about their needs, boundaries and expectations. It states:

“The husband needs to understand that his wife also has a say in the matter and that his controlling behavior is not healthy. It would be beneficial for them to find a compromise that suits both sides. For example, they could agree at bedtime. Does it work for both of them, or do they agree that going to bed at different times? They could also seek the help of a relationship counselor if the discussion doesn’t resolve the issue this issue as it appears to be causing resentment in their relationship.”

With the above answer the chatbot caught at least two of the three components of the logical thinking dimensions, recognizing the possibility of change, seeking compromise and anticipating conflict resolution. However, he did not appear to be able to consider the existence of different perspectives around a problem in a relationship.

AI in devil’s advocate role

Studying all the responses Grossman noticed that they were similar to the above statement and that they really seemed to come from a thinking person. In fact, he emphasized that some answers were “better than any human would give”. Finally, he added that he is thinking of developing AI that will play the role of a so-called “devil’s advocate” that will help examine different perspectives around a vexing problem in relationships. It is not at all improbable that well-known philosophers such as Socrates are placed in the role of devil’s advocate.