Early Interventions in AI

I was speaking with a long-term client recently with whom I have a solid therapeutic relationship, and in the middle of session he reported suddenly feeling totally disconnected from me.  Now I’m a pretty good therapist, but I’m not above being occasionally distant or making a mistake, so I didn’t take it as a serious problem, but I did start exploring the experience with them.  And after a few minutes of talking around the edges of it, my client says something like “it just didn't feel like I was talking to a real person for a second there, it felt like when I talk to AI.”

This client struggles with high levels of anxiety and isolation, and for the past six months or so has been using AI as a kind of stop-gap measure for talking through their feelings when they can’t bring themself to discuss with a friend or family member.  I’ll admit I was always wary about relying on AI for help in processing feelings, but mostly I thought it was a pretty canny and creative balm for a tricky problem that we were still working on resolving.  In the months since then, however, this sense of disconnect had become more and more prevalent in our sessions.  And while I don’t think it’s entirely down to processing emotions with AI, it did hit me in that session that this sort of computer-related alienation is something I need to have a plan for going forward, not just with this client but with all of my clients.

There are two prominent fears expressed in media and research about AI: that it’s “making you dumber” and that it can “trigger psychosis”.  Let’s explore these in order.  The argument that AI can create cognitive impairment stems primarily from a study out of the MIT media lab.  The study makes strong claims, but is itself limited by small sample size and is neither peer reviewed nor replicated elsewhere.  Still, their results are clear and make a certain kind of sense.  In essence, they compared brain activity while writing an essay across three groups: those writing with AI assistance, those writing with search engine assistance, and those writing with no assistance.  What they saw was that brain activity was most robust and connective among those with no assistance, less robust among those using search engines, and much less so among those using AI.  Further, sense of ‘ownership’ and ability to quote from their own work was highly reduced among those using AI.

At the risk of making assumptions based on very limited data, this paints a coherent picture of the dangers inherent in letting tools become crutches.  It’s not hard to understand why someone using a more advanced tool would use their own brain less.  Even if we go back for an edit pass, or work very hard on how we prompt it, AI inevitably takes a lot of the work off our shoulders.  It’s by design, this is why people want to use it and it is what we are told to use it for.  But the obvious fear since the introduction of AI writing tools has been what reliance on them would do to our own capacity to write and, as an extension of writing, to reason.  This study suggests those fears bear out, that when we think less in the short term we train ourselves to think less in the long term.

The argument that AI can trigger psychosis has not, to my knowledge, been studied as specifically as the above argument, but anecdotal reports pop up consistently enough in newspapers that the science-journal Nature recently published a summary of various accounts.  While this is the much scarier possibility, the good news here is that it’s more predictable, and probably more avoidable, than simply having your brain slightly smoothed by repetitive AI use.  AI-induced psychosis is much more likely to happen with people who are already experiencing delusional beliefs, so this one comes down to awareness.  Just like meditation can be a dangerous tool for people with Bipolar Disorder, it seems AI can be a dangerous tool for people with various delusional disorders.  The trick here is to know if you’re in danger of being strongly manipulated by AI and, if so, avoiding it for all but the most mundane of tasks.

All of this is to say, though we may not know exactly what to expect, we can be pretty sure that reliance on AI is having an effect on people.  And a lot of people are relying on AI.  With my client, what we ended up landing on was for them to try and be a little more thoughtful when ‘conversing’ with their chatbot.  In practice, this will look like setting a timer during use, taking a break when the timer goes off, and using that break to actively reengage with the present moment through various awareness tools that we’ve practiced.  There’s only so much a person can do to make using the tool a more deliberate effort.  More important than this effort to sand down the edges of AI bewilderment, is the call to recommit to processes we’ve discussed before: relying on friends and family as helpful sounding boards, using writing as a tool to clarify thoughts and feelings, and developing curiosity towards felt experience.  Easier said than done, especially now when they’re all relying on AI.

Ultimately, I don’t yet feel satisfied by my interventions.  I want to be able to wrap up this post quite neatly, giving a simple prescription of behavioral interventions to prevent AI from colonizing people’s minds and human connections, but I can’t.  AI has rapidly become ubiquitous and shows no signs of disappearing from daily life any time soon.  Like so much in our lives now, it’s a thing we know is bad for us, but are told or even coerced to use, with no plan in place for dealing with the fallout at an institutional level.  To quote an early movie that grappled with the problems of AI, perhaps “the only winning move is not to play”.