Michal Kosinski is a Stanford research psychologist with a nose for current issues. He believes that his work not only promotes knowledge, but also alerts the world about the potential dangers caused by the consequences of computer systems. His best-known projects involved analyzing the ways in which Facebook (now Meta) gained a surprisingly deep understanding of its users from all the times they clicked “like” on the platform. Now he has focused on studying the amazing things AI can do. He has conducted experiments, for example, that indicate that computers could predict a person’s sexuality analyzing a digital photograph of his face.
I got to know Kosinski through my writing on Meta and reconnected with him to discuss your last articlepublished this week in the peer-reviewed Proceedings of the National Academy of Sciences. His conclusion is surprising. He claims that large language models like OpenAI’s have crossed a boundary and are using techniques analogous to real thinking, which was once considered solely the realm of flesh-and-blood people (or at least mammals). Specifically, it tested OpenAI’s GPT-3.5 and GPT-4 to see if they mastered what is known as “theory of mind.” This is the ability of humans, developed in childhood, to understand the thought processes of other humans. It is an important skill. If a computer system cannot correctly interpret what people think, its understanding of the world will be poorer and it will get many things wrong. yes models do They have theory of mind, they are one step closer to matching and surpassing human capabilities. Kosinski put the LLMs to the test and now says his experiments show that in GPT-4 in particular, a mind-like theory of ability “may have emerged as an unintended byproduct of the LLMs’ improved language skills. .. They mean the advent of more powerful and socially capable AI.”
Kosinski sees his work in AI as a natural outgrowth of his earlier immersion in Facebook Likes. “I wasn’t really studying social media, I was studying humans,” he says. When OpenAI and Google started building their latest generative AI models, he says, they thought they were training them to primarily deal with language. “But they actually trained a model of the human mind, because you can’t predict what word I’m going to say next without modeling my mind.”
Kosinski is careful not to claim that LLMs completely master theory of mind… yet. In his experiments he presented some classic problems to chatbots, some of which they solved very well. But even the most sophisticated model, the GPT-4, failed a quarter of the time. The successes, he writes, put GPT-4 at the level of 6-year-olds. Not bad, given the initial state of the field. “Looking at the rapid progress of AI, many are wondering if and when AI could achieve ToM or consciousness,” he writes. That radioactive C-word aside, there’s a lot to unpack.
“If theory of mind emerged spontaneously in those models, it also suggests that other abilities may emerge next,” he tells me. “They can be better at educating us, influencing us and manipulating us because of those skills.” He worries that we are not really prepared for LLMs that understand the way humans think. Especially if they get to the point where they understand humans better than humans.
“Humans do not simulate personality; have personality,” he says. “So I’m a little stuck in my personality. these things model personality. The advantage is that they can have whatever personality they want at any time.” When I mention to Kosinski that it sounds like he’s describing a sociopath, he lights up. “I use it in my talks!” he says. “A sociopath can put on a mask; he is not really sad, but he can play a sad person.” This chameleon-like power could turn AI into a superior con artist. No regrets.