why a.i. can't help you find yourself
People are trusting AI more and more for life and career coaching. Too bad machines are just as bad at is as humans.
One of the weirdest things about people is that we like to both simultaneously define ourselves as fluid individuals and in very concrete terms. Even counter-culture groups which say they’re rebelling against society’s norms and rules tend to put themselves in tidy little boxes, and that’s a trend that seems to have sharply accelerated thanks to social media because we’ve been trained by our algorithmic segmentation that to get what we want, we should be very concrete and explicit about our identities.
Normally, when I say this people immediately think of running into someone who tells them they’re “an aromantic poly demisexual cis AFAB” and how they struggled to find all the definitions and their meanings. They tend to have a lot less trouble with a very similar phenomenon in the business world. Specifically, the Myers-Briggs personality types. You can still run into plenty of LinkedIn profiles where “proud ENTJ founder” or an “ENFJ executive” are peppered across bios and headlines.
Normally, people have to take extensive questionnaires to figure out their type, then pay for career advice and coaching to keep diving deeper and use this data to plot a working life. In recent years, however, an AI analysis of their emails and essays started doing the same job, coming up with what seemed like uncannily accurate summaries of the Myers-Briggs type they felt was closest to their personalities.
But how were the AI models so good at that? According to researchers in Spain who ran a study looking into how the classification was being done, they’re not. We’re just giving them a lot of leeway, just as we do with all personality tests.
cold readings, now with chatbots
You see, the Myers-Briggs personality types are not really considered a valid thing in psychology, and are a little more than a marketing gimmick to sell courses claiming to be helping you on a journey of self-discovery and self-mastery, and who doesn’t like the sound of that? The creators of the test even felt the need to counter this, saying that their work is based on the work of Carl Jung. Who, by the way, is also not exactly embraced as an empirical scientist due to his reliance on metaphysics.
Supposedly, I’m an INTJ, or The Architect. Sounds profound, but you’d probably know that with one look at my bio, which says that I’m an engineer, and a quick scan of the possible personality types on the website. Each personality is described in vague and generic terms with which it’s hard to disagree too loudly, and the overall description is pretty positive, kind of like how horoscopes describe each sign.
An AI model then picks up on this and slots you into the type where most buzzwords and descriptors seem to be present in your writing. Because you let an AI categorize you based on all these vague, generally interchangeable, rather positive criteria using this binary surface level analysis, the end result can sound enough like you to say that yes, the machine has you all figured out. But if that’s all it took, then so could any cold reading scam artist like a psychic or a pyramid scheme recruiter.
Human personalities are complex, exist on a continuum, and are subject to changes over our lifetimes. Any classification system that doesn’t account for that and taken too seriously can be pretty harmful because it can lead people who want to fit into a certain community to try and change themselves into something they’re not despite their own minds’ objections, or spend time and money to keep getting answers that sound right at first, but don’t feel right over the long term because they were never right in the first place, all of which leads to unnecessary stress.
mass customized frustration
Think of all these personality tests and identifiers like Converse shoes. Sure, you can customize the color and style you really like, but in the end, it’s the same, very finite types of shoes, with a very finite types, styles, and runs of patterns you specify until you’re happy with them. They’re not going to be custom tailored to your foot and to your exact specifications. They’re also not going to change when your preferences drift and you decide something else fits you better. Which is all fine and well.
You’re not expecting that because Converse is just selling shoes. They’re not telling you how to reshape your foot to better fit in them later. But personality testing, does, in a way. They’re promising to sell a guide to yourself and how to navigate the world the best way possible, but because they don’t understand you either — hell, most of us don’t understand ourselves very well — and both you and the world keep evolving as time goes on, it’s a promise they can’t keep.
This is why the Spanish researchers wanted to figure out whether AI is actually good at predicting and defining your personality or not. As the marketing craze over LLMs rages out of control, assessment companies and life coaches will want to offer it and boast about how it improves accuracy and reduces the client’s effort, which churning out the same flawed results and propagating the same old problems while ignoring a significant body of scientific research that offers far better answers.
See: Saeteros D, et al (2025) Text speaks louder: Insights into personality from natural language processing. PLOS ONE 20(6), DOI: 10.1371/journal.pone.0323096