top of page
Search
  • Writer's pictureJohn Lantos

Outsourcing empathy




Sharma and colleagues report a randomized clinical trial of human-only or human+AI respondents to people who sought counseling on a peer-support website. Both groups first received basic training on empathy, which included empathy definitions, frameworks, and examples. Empathy was defined "the therapists’ sensitive ability and willingness to understand the clients’ thoughts, feelings and struggles from the clients’ point of view." In the training, they gave categories of empathic responses with examples: “Empathic responses typically involve: 1) Reacting with emotions felt after reading a post (e.g., I feel sorry for you); 2) Communicating an understanding of feelings and experiences (e.g., This must be terrifying); and 3) Improving understanding by exploring feelings and experiences (e.g., Are you feeling alone right now?)”


During the study, each participant was asked to write supportive, empathic responses to a unique set of 10 existing seeker posts. While writing responses, participants in the Human + AI (treatment) group received feedback via a real-time robotic empathy coach. Participants in the Human Only (control) group received no such feedback.


The researchers evaluated the empathic nature of responses using both human and automated methods. For the human evaluation, people compared their perceptions of the empathy levels in responses to the same post. For the automated evaluation, they measured responses using an empathy classification model that had been developed to evaluate empathic communication.


They found that AI helped people express empathy. No big surprise there. But the study does raise important questions about how, exactly, AI will be used in medicine. What specific doctoring skills will it replace. Eric Topol’s view is that it will do the tedious tasks and free up doctors to be more humanistic. This study suggests that AI might, in fact, be more humanistic than many people. Vinay Prasad thinks it will make chart documentation obsolete and may be better and formulating treatment plans than most doctors are today. He still holds out hope that the best doctors will be better than AI but that sounds a lot like the early days of chess-playing programs when we held out hope that the Grand Masters would never be beaten. They were beaten, of course, and we know now that computers are better at chess than we are.


I don’t think that chatGPT or any other robot will replace doctors because I don’t think we are clear about what it is that we want doctors to do. That is an interesting morass. We have to know not only what doctors do or don’t do but also how well or poorly they do specific things compared to other professionals or, now, compared to robots. We might ask whether one needs to take a history, offer empathy, make a diagnosis, recommend a treatment, determine which types of lenses best correct myopic vision, or counsel the emotionally distressed. Should only doctors perform physical examinations, or administer anesthesia, or determine whether a patient is ready to be discharged from the hospital? Should we leave it to doctors to determine whether another doctor was negligent in a particular situation? How do we define and measure quality in order to decide whether doctors do these tasks better or worse than other people? All of these tasks have been and may be performed by doctors, but they may also be and often are performed by others. And soon there will be the robots. We may find that many of these tasks, including empathy, can be outsourced to a machine that will do it better and for less money.


In the near future, there will be multiple different types of robots and patients and doctors both will have to consult many of them and then, unfortunately, do what we do today when the data are conflicting and experts disagree – that is, make a decision with imperfect knowledge, quantifiable uncertainty, and urgency.

289 views4 comments

Recent Posts

See All

4 commentaires


claudia
31 mars 2023

Two thoughts: first, why was a different method of evaluation used with the all-human and the AI+human responses? It makes it hard to compare the merits.

Second, some doctors have a shockingly low empathy response, and can't even fake it. I think we understand that repeated exposure to grief/fear/shock in a patient or their family can numb the practitioner, indeed may have to numb them in order for them to do their job, but one would think that they could still muster a five second "there, there, we'll get through this" that can make or break a patient-doctor relationship. Unless this was being done in a chatbot, it would be hard to replace in person.

Imagine the fearful diagnosis, the…

J'aime
John Lantos
John Lantos
31 mars 2023
En réponse à

The evaluation was the same for both. Both were evaluated by both humans, subjectively, and a language coding algorithm designed to identify words or phrases pre-defined as “empathic.” The intervention group did better on both. As you say, not all doctors are good at empathy. In the example you give, a good doctor is probably better than any machine, but patient narratives are full of stories about doctors who are shockingly bad at just that sort of conversation. I like the idea that doctors could learn empathy from a robot.

J'aime

Joshua Briscoe
Joshua Briscoe
31 mars 2023

Perhaps a bigger concern would be humans becoming more like machines (as you observed, that AI might demonstrates signs of empathy more clearly than some people). Neil Postman had some things to say about the time when machines set the standard: “we are at our best when acting like machines, and that in significant ways machines may be trusted to act as our surrogates. Among the implications of these beliefs is a loss of confidence in human judgment and subjectivity.” https://familymeetingnotes.substack.com/p/health-without-other-people

J'aime
John Lantos
John Lantos
31 mars 2023
En réponse à

Thanks, Joshua. It is true that all metaphors here are two-way. We think about our own thinking using machine metaphors. And we imagine that machines could feel things. We don’t know if they do. But we don’t know what it feels like to be a bat, either. On mental health, an app called “woebot” is being used, successfully, to screen for risk of suicide. Jury is still out whether psych AI will prevent more suicide than it causes. I’d guess prevent…

J'aime
bottom of page