Where Therapists Go for Therapy: Getting Sold on the Power of ChatGPT
The story of my 2 days working with Dan Shipper on the "Maximize your Mind with ChatGPT" course
I believe what Steve Jobs said about how you “only connect the dots in retrospect,” but every once in a while, you know with unshakeable certainty that this moment you’re in right now is a massive new dot getting drawn. You don’t know exactly where you’ll go from here, or even what exactly to make of the experience, but you know it will have been a major inflection point on your timeline.
Last week I had such a moment. I had flown to New York to spend a couple of days with Dan Shipper, the co-founder and CEO of Every, with the goal of ideating on a new course for “leveling up your psychology with ChatGPT.” Dan is a writer, technologist, and major AI enthusiast. We’d gotten to know each other through our respective work and writing, and wanted to explore a shared interest in using LLMs for psychological self-betterment.
In truth, I was slightly skeptical about the LLM part, and was mostly just excited to get to contribute to the psychological part. But after two days of combining our efforts to get ChatGPT to do what my best graduate student trainees maybe start to do semi-consistently in their second or third year of therapy training, I became a believer.
And in early 2024, Dan and I are launching our course: Maximize your Mind with ChatGPT (note from 1/20/24 Gena: registration is now open!).
To give you a preview of what we’ll be covering, let me share a few ways I experimented with leveraging ChatGPT for psychological growth during my two days with Dan.
Clarifying values with ChatGPT
Given our shared conviction that articulating core values is a crucial starting point for psychological growth (just as articulating a company’s core values is crucial for healthy business growth), we first wanted to see if we could get ChatGPT to help with this. Drawing on the first exercise in my life vision worksheet, we inputted the following prompt:
After some initial false starts and tweaks to the prompt (e.g., requesting that it ask its questions one at a time rather than all at once), ChatGPT capably followed these instructions. First it asked me to name the feelings that have recently come up, and I listed “sentimental” and “restless”. Then it followed up with appropriate questions about the values reflected by each one. It even showed a good intuitive sense of when to move on from one feeling to the next, given how deep I’d gotten in identifying the relevant values:
(Alice is my 3-year-old daughter.)
This was a pretty good start: from just my one-word description of a feeling, ChatGPT pretty quickly elicited and poetically summarized back to me one of my fundamental raison d’etres. On the other hand, it missed a crucial aspect of my value system—the “builder” aspect—despite my planting a seemingly obvious clue (cf. the “amazing things that humans have built”). Notably, when I later got home and recreated this chat with my newly personalized ChatGPT, whose custom instructions (roughly modeled off of Dan’s) include a reference to my writing and an emphasis on “personally embodying the builder’s mindset” as one of my growth goals, it did much better:
This time ChatGPT put an appropriately sharp emphasis on the “building” theme. But in so doing, it lost the “childlike wonder” aspect that is equally core to my value system. I gave this feedback to ChatGPT, telling it that its summary was “missing some of the ‘sense of wonder’ and ‘human connection’ aspect I think we started with.” ChatGPT then returned this more complete and personally resonant summation of my values:
Coming back to the original exchange: after wrapping up the “sentimentality” thread, ChatGPT asked me to reflect on the values implicit in my feeling of “restlessness” (the second feeling I’d named in my original list above). This led to a revealing dialogue:
At this point ChatGPT had elicited from me a reflexive assumption that sometimes creeps into my thinking: that I risk “running out of time” if I don’t “get to [the right] people quickly,” whatever that means. Per its original instructions, ChatGPT then asked me some guiding questions about what made this concern important to me, and then it summarized the corresponding value under the description “Achieving Ambitious Goals in a Timely Manner.”
This exchange made me realize there’s a subtle further task I often perform as a coach or therapist when helping clients clarify their values: I listen for implicit, possibly unchecked assumptions about themselves or the world that may be coloring the values they articulate (a la my fear above about running out of time). So I asked Dan if this was something he thought ChatGPT could be trained up to do. He was game to try.
Identifying and examining implicit beliefs with ChatGPT
Toward this end, one of our next experiments was to create a Hidden Premise Finder bot, whose main job is to “analyze chat transcripts to identify and summarize suspected implicit assumptions or beliefs exhibited by users.” When I copy-pasted the earlier chat about my feelings and values into this bot and asked it to pull out the one implicit belief or assumption “most worth interrogating or correcting,” it came up with this:
Wow, ok—this “running out of time” belief is also the one I was thinking of. Wondering if perhaps our Hidden Premise Finder had just gotten lucky on the first go, I fed it the same information and asked it the same question a number of different ways. With some slight variations and differences in emphasis, it managed to arrive at this one psychological sticking point every time.
I then asked it to guide me in examining this belief critically, and it came back with a bunch of questions for me to consider:
These are, indeed, many of the same questions I would hope to consider if I were coaching myself, including some that may not have even occurred to me, like the role of “personal fulfillment” versus “external validation” in driving my sense of urgency. And they provide better coverage of the relevant conceptual issues and evidentiary sources than I would expect from maybe 40-60% of therapists.
ChatGPT for the Win: Inspiring exemplars on demand
Still, up to this point ChatGPT hadn’t told me anything about myself that I didn’t already know. (I am a professional psychologist with a lifelong commitment to practicing and promoting the examined life, after all.) The real kicker came at the next step, when I asked for its help with the second question it had posed above, about “alternative perspectives”:
Do you know how long I’ve been pining for just such a list of “late bloomers” to share with my clients who express just the sort of “it’s too late/I’m running out of time” narrative I succumbed to above?! Of course generating examples like this would be one of ChatGPT’s strengths, seeing as it has the whole internet at its proverbial fingertips. What’s more staggering is that it knew to provide me with this more general, yet still relevant and inspiring, list of examples in lieu of the (extremely specific) query I’d given it, rather than conceding ignorance or turning the question back around on me like a typical therapist might. This, too, is one of ChatGPT’s strengths: it is not afraid of “looking stupid,” and it almost always makes a reasonable attempt, rather than giving up or asking for clarification.
(I haven’t yet fact-checked the above list, so please don’t quote me on it. But do feel free to ask your own ChatGPT to tell you more about any of the examples that intrigue you, and then follow up its sources to see where they lead.)
A better-than-average therapist, if you know how to use it
I got a similar mix of initially middling but eventually useful and even surprising results when I ran the following experiments:
Linking ChatGPT to a Notion page with years of my compiled journal entries and asking it to summarize the major themes, psychological patterns, and growth areas (a la some of Dan’s own earlier experiments, as he documents here and here)
Creating a Motivational Interviewing (MI) bot whose job is to help users work through decisions about which they are ambivalent
The upshot of all these experiments is that I am now confident that I can more reliably extract value from talking through my feelings, thoughts, and decisions with ChatGPT than with a randomly selected human therapist.
Of course, ChatGPT cannot substitute for my own judgment and creativity (as no one and nothing can). And one likely needs at least some prior knowledge of how human psychology works to use LLMs skillfully—similar to how one needs some knowledge of coding to competently use AI code generation. But by enabling me to query the whole sum of existing human knowledge—including my own accumulated self-knowledge—in plain English, ChatGPT massively accelerated and extended the scope of my own reasoning.
You might ask: aren’t I worried about being replaced by ChatGPT? Exactly the opposite: I’m exhilarated at the prospect of getting freed up to focus on the problems that lie beyond the frontier of existing psychological knowledge, while empowering people (clinicians and clients alike) to leverage the existing knowledge base better and faster with ChatGPT.
As to the Hidden Premise Finder’s fourth question above: “How might your skills and experiences as you age actually open new doors or create new opportunities?” I don’t know yet—but I’m pretty sure I just walked through one of those doors, and the air is abuzz with opportunities.
To register for our Maximize Your Mind with ChatGPT course:
Building the Builders is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.