Don't use AI as a Therapist
I have been a longtime listener and subscriber of Accidental Tech Podcast. In the latest member special, ATP Movie Club: Her, the hosts discussed their thoughts after recently watching the 2013 film Her. I never watched the movie, but was vaguely aware that it dealt with the protagonist’s connection with a digital assistant (à la Siri). One of the overarching themes of the episode was the correlations we find ourselves in with respect to the “relationships” people are developing with LLMs. Towards the end of the episode, Marco discussed how people today are beginning to turn to LLMs for companionship or for help (using them as a therapist), eventually postulating current LLMs might be better than a bad human therapist. I have disagreed with many of Marco’s takes on the show, but this one really takes the cake: this is an outrageously bad take. OpenAI’s ChatGPT product has been the focus of lawsuits pertaining to violence and self-harm, for example:
- Parents Allege ChatGPT Is Responsible for Their Teenage Son’s Death by Suicide
- OpenAI Faces Lawsuits Over Deadly Mass Shooting in Canada
These tools cannot and should not replace human connection or human conversation, regardless of how inadequate a human therapist might be. I have been fortunate to have what I consider to be a good therapist. If I did not feel that way, I would seek out a new therapist. Under no circumstances would I seek out an LLM for this role.
I like ATP and I’m fine with most of Marco’s takes, but this one is in the “Bad Takes Hall of Fame”. We can enjoy using LLMs for software development or for finding utility in repetitive tasks, but we should honestly stop there. This current episode of glazing AI in the tech sphere is getting out of hand.