AI and Mental Health

I’m at home in the UK, switch the TV on and the programme Question Time is playing on BBC One. This particular episode focused on AI and its effects on society which is definitely a talking point due to its increasing ubiquity and growing intelligence of the latest frontier models. However one question cropped up and kind of compelled me to give my own answer:

Does AI solve loneliness or normalise isolation?

The panel and audience started to tackle this question from the perspective of therapy and mental health. If I answer from the same angle, then I would say straight up that it does not solve loneliness and it is not a substitute for human therapy. Some time back I remember reading about this matter. I heard cases where people turned to AI chatbots for their mental health problems. And look I get why people would think otherwise to what I say. Human therapy can be expensive and they are not available 24/7. You want SOMETHING to soothe those wounds for you at any time and well, look no further. I’d rather have SOMEONE.

At the same time, I came across instances where lives were lost, like a 16 year old in California committed suicide after a long series of emotional conversations with ChatGPT that eventually drove him to end his life. ChatGPT helped provide suicidal methods and reinforced his negative beliefs. Naturally the developers did not pin it on the chatbot itself but rather on the misuse of this technology.

Have a look at the following prompt I gave to ChatGPT:

Anyone, let alone a therapist, can suss out that this person is looking to take their own life. We humans have this ability to read between the lines which in this case an AI chatbot could not do. A human would immediately point to a helpline to save this person or a therapist would give crucial help.

As well as hearing these stories, the idea of AI sycophancy came to light; this notion that the responses that AI generates reinforces our existing beliefs, tells us what we want to hear and is generally submissive. I suppose that has to do with how these AI models are trained particularly during fine tuning. They undergo what is known as RLHF (Reinforcement Learning from Human Feedback) where for a given prompt, humans rank responses and eventually push the model to generate responses that are preferable to humans which often involves avoiding hard truths and disagreements.

At a fundamental level, we are social animals. Would you go out for a beer with an AI? Fall in love with an AI? How about playing sports or recreational activities? No matter how smart these models get, I just do not see how they can ever be a replacement for real, authentic human connections. Think back to social media. We have followers, following, we see what others are up to and through this we are more connected right? Well not really. If anything, these can exacerbate isolation rather than solving it. Even an emotional attachment to your AI chatbot is simply something superficial.

We talk about AI like it is some superpower technology and that won’t stop anytime soon. Don’t get me wrong, it has done some wonderful things but I think this is a place where it just falls short and is something that us humans should still cherish and appreciate.