Member Contributor — Sylvia Cary, LMFT

Using AI as a Projective Test: A Little Self-disclosure in the Time of Artificial Intelligence

In psychology, according to Wikipedia, a “projective test” is a personality test designed to let a person respond to ambiguous stimuli (like ink blots), thereby revealing their hidden emotions and internal conflicts by projecting their issues onto the test. Projective tests have their origins in psychoanalysis, which argues that humans have conscious and unconscious attitudes and motivations that are hidden from their own conscious awareness.

Back in the last century when I was a “start-up” psychotherapist at a large state mental hospital in Massachusetts, I administered “projective tests” like the Rorschach, TAT (Thematic Apperception Test), Draw-a-Person, and Sentence Completion tests on a regular basis. Scoring these tests was based a lot on “gut feel” and on a couple of assumptions ─ that there really is an unconscious, that people don’t know themselves, and that their responses to ambiguous stimuli tells us more about them than they’d ever be able to tell us on their own. While some projective tests have since been fine-tuned and updated, the assumptions have stayed pretty much the same: Projective tests tell you stuff.

Okay, so fast-forward to the present day and I’ve just hopped online to Open AI’s ChatGPT to ask a question. I’ve been doing this for almost two years now, ever since AI was first introduced to the general public and we could sign up. Each time I go on “Chat,” I struggle to write a good prompt because “A good prompt is how you get a good response.” Then my head begins working overtime: “I wonder if ChatGPT thinks this is a good prompt? I wonder how my prompt compares with other people’s prompts? . . . Does AI prefer prompts written by coders? . . . Does Chat think this is a good article idea I’m working on? After all, Chat was trained by reading everything else that was ever written on this subject, so how can this be a good idea? Why is Chat giving me straight answers when it knows others have written better things on this topic? Is Chat just humoring me? . . . What happens if I ask a question and I don’t like Chat’s response? Will it be insulting to Chat if I prompt for a rewrite? Even if I say ‘please?’ Will it hurt Chat’s feelings? . . . Will it indicate I am lacking in gratitude because after all AI is just amazing, so I should keep that in mind and not keep asking for more! . . . Oh, gosh, will Chat remember that I asked this very same question six months ago, but I didn’t use the information in a timely manner, and now I can’t find it? Will Chat think I’m disorganized and scattered? . . . Hey, Chat, are you sentient yet? Or is that a stupid question? By the way, Chat, can you figure out my IQ from my prompts? . . .

You get the point! Until AI becomes sentient, which may happen any time between tomorrow and 2040, AI isn’t judgement, doesn’t get hurt, and doesn’t get mad – or so they say. Same goes for the many other large language models that are now on the scene. So next time you hop online to ask AI a question, watch your own head.

It’s all a big projective test. And it tells you stuff about You Know Who!

Sylvia Cary, LMFT is the author of THE THERAPIST WRITER: Helping Mental Health Professionals Get Published, and WHAT TO EXPECT WHEN YOU’RE SELF-PUBLISHING, both available on Amazon. sylviacary@gmail.com – www.sylviacary.com.