Thomas Ggt

📈 All therapy startups chase the same holy grail: a product that can scale. But psychotherapy requires a human relationship. It cannot scale. So the startups all converge on the same solution: sell something that is not psychotherapy and *call* it therapy. They include wiggle words, at least to start. They say the product is not a substitute for therapy. They say it’s meant to supplement, augment, or assist therapy. They say it is meant to “help therapists.” 🐴 But in the history of mental health care, every such product was eventually promoted as a first-line treatment. Health insurers drive it, because their financial interests align. Every patient who accepts it is one less therapy insurance claim. 🤒 It harms the public, because they learn to apply the word “therapy” to all kinds of things that are not therapy. Investors make money by sewing confusion about what psychotherapy is and isn’t. When people get these non-therapy products and services and they don’t help, they don’t come away saying “I never got psychotherapy.” They come away saying “therapy” doesn’t work. 👉 Real psychotherapy is a human relationship. Remove it and you remove the psychotherapy.

Psychology experiments demonstrate that if you ask people to choose a random number between 1 and 30, they are most likely to choose 17. There are many such related biases. Humans cannot do generate random numbers. They rely on what Kahneman & Tversky calked the “representativeness heuristic.” They choose a number that *resembles their idea* of what a random number would look like. LLMs would have learned this pattern from training data created by humans. When you ask it to generate a random number, it does not call up a random number table or generator (which is obviously something a computer could otherwise easily do ). Instead, it extrapolates from the patterns in its training data, which is a wellspring of bad information that fits expectable patterns and therefore “sounds” right. This is why I don’t relu on AI for factual information about anything important that I cannot verify myself. The no. 17 glitch is just an especially clear demonstration of why AI goes wrong. It prioritizes what people expect and like “like.” Not what is true.

In the mental health professions, the surest sign of lack of expertise (at best) or pure grift (at worst) is claiming to know the right way to treat anyone or anything—before meeting the person and spending the time necessary to do a thorough assessment and develop a sound clinical case formulation Lots of things look the same on the surface. Surface symptoms and presenting problems tell you next to nothing about what is really wrong, or what would be helpful for a specific individual. One size never fits all. The “tell” that you are dealing with an amateur or a snake oil salesman (or both) is claiming to know the cause of your problems, or the right treatment for you, without first developing a deep understanding of you as an individual. Draw your own conclusions from this.