UC Berkeley researcher argues artificial intelligence cannot replace human empathy

photo of UC Berkeley public health building
Isabella Ko/Staff
Artificial intelligence is not a substitute for human empathy in health care, according to a paper by Jodi Halpern, professor at UC Berkeley and UCSF.

Related Posts

In a paper published May 26, Jodi Halpern, professor of bioethics and medical humanities at UC Berkeley and UCSF, argued that artificial intelligence, or AI, is not a substitute for human empathy in health care.

Halpern wrote the paper with authors Carlos Montemayor and Abrol Fairweather from San Francisco State University. In the paper, they claim that simulating empathy with AI is both impossible and immoral — impossible because AI cannot experience emotion and immoral because patients deserve human empathy during times of distress.

Halpern said the paper is a response to the growth of AI in health care, especially concerning mental health.

“Just as we deployed actual humans to do contact tracing because of the relational aspects and the need for trust, we can deploy actual humans to provide mental health support,” Halpern said in an email.

However, Halpern notes in the paper that she is not against AI in health care as a whole and there are many places where AI can be useful.

According to Niloufar Salehi, an assistant professor at the UC Berkeley School of Information, AI is a great tool for diagnosis and helping patients track their own health. She said the problem arises when people in health care treat AI as more than what it is — a pattern finder.

“We have to be very, very careful here,” Salehi said. “You don’t want to trick people into thinking that AI or a computer is sympathizing or empathizing with you.”

Salehi focuses on human-centric AI, which identifies gaps in human work that AI can fill or supplement. According to Salehi, it aims to develop AI around people’s needs, rather than the other way around.

Salehi agreed with Halpern’s paper, noting that people need to continue to draw a distinction between what is impossible versus what is unethical with regard to AI.

Halpern emphasized another distinction, specifically between simulated empathy and human empathy: the ability to have experiences. This allows humans to resonate with other people in a way that a computer cannot, according to her paper.

Both Salehi and Halpern said they see a future for AI in health care, but they agreed that it cannot substitute interactions that require empathy.

Halpern added that she is currently working on a book titled “Engineering Empathy,” which will delve deeper into the role of AI in interpersonal relationships. In the meantime, she hopes that her paper might shift the general consensus of AI use in health care.

“We provided a conceptual argument that we hope will show that it is not only technical limitations but the nature of empathy that limits AI going forward,” Halpern said in an email.

Contact Riley Cooke at [email protected], and follow her on Twitter at @rrileycooke.