How scientists are trying to use AI to unlock the human mind 

Compared to the traditional psychological models, which use simple mathematics equations, Centaur has done much better to predict behavior. The exact predictions of how humans respond to psychology are of value in itself: for example, scientists can use Centaur to experience their experiences on the computer before employment, payment, and human participants. However, behind them, researchers suggest that Centaur be more than just a prediction machine. By interrogating mechanisms that allow Centau to effectively repeat human behavior, they argue, scientists can develop new theories about the internal works of the mind.

But some psychologists question whether Centaur can tell us a lot about the mind at all. Certainly it is better than traditional psychological models in predicting how humans behave – but also have a billion times. And because the model behaves like a person outside, it does not mean that it works like one inside. The Olivia guest, Assistant Professor of Cognitive Sciences at Radbod University in the Netherlands, compares Sentor to a calculator, which can effectively predict the response that its charter will give when it is asked to add two numbers. “I don’t know what you will learn about human addition by studying a calculator,” she says.

Even if Centaur picks up something important in human psychology, scientists may struggle to extract any insight from millions of typical nerve cells. Although artificial intelligence researchers are working hard to know the extent of large language models, they were barely able to open the black box. Understanding a huge nerve retinal model for the human mind may not be much easier than understanding the same.

One alternative method is to become small. The second of the two nature Studies focus on small nerve networks – which contains only one neuron – which can predict behavior in mice, mice, monkeys and even humans. Since networks are very small, it is possible to track the activity of each individual nerve cell and use this data to see how the network is produced for its behavioral predictions. Although there is no guarantee that these models work like the brains that have been trained in tradition, they can, at least, can generate test hypotheses about human and animal perception.

There is a cost of inclusiveness. Unlike Centaur, which has been trained to imitate human behavior in dozens of different tasks, every small network can only predict behavior on a specific mission. One network, for example, specializes in providing predictions on how to choose people among different gambling machines. “If the behavior is really complicated, then you need a big network,” said Marcelo Mattar, an assistant professor of psychology and neurology at New York University who led the study of the small network and also contributed to Centaur. “The middle solution, of course, is that it is now very difficult.”

This comparison between prediction and understanding is a major feature of the science that the nerve network drives. (I also encountered a book on that.) Studies such as Mattar make some progress towards closing this gap – like its networks, can predict behavior more accurately than traditional psychological models. As well as searching for the ability of LLM in places like humans. However, at the present time, our understanding of complex systems – from human beings to climatic systems to proteins – is far away and far from our ability to provide predictions of them.

This story originally appeared in the algorithm, our weekly newsletter on artificial intelligence. For stories like this in your inbox first, subscribe here.