Compared to traditional psychological models, who use simple math equations, Santor did a better job in predicting behavior. The correct predictions of what humans respond to in psychology experiments are valuable in themselves and themselves: for example, scientists can use Santor to recruit and pilot their experiences on the computer before paying human participants. However, in their dissertation, researchers suggest that the Sentor may be higher than just one forecast machine. Scientists can create new ideas about the internal works of the mind by questioning the mechanism, which allows Santor to transmit human behavior effectively.
But some psychologists suspect whether Sentor can tell us something about the mind. Certainly, it is better than the traditional psychological model in predicting how humans behave – but it also has a billion times more parameters. And just because a model behaves like outside humans does not mean that it works as inside. The guest of Olivia, Assistant Professor of Competition Sciences at Radbod University, Netherlands, compares Sentor with a calculator, which will be asked to add two numbers to the answer to the answer to mathematics. She says, “I don’t know what you will learn about human additions by studying a calculator.”
Even if Santor gains something important about human psychology, scientists can struggle to remove any insights from millions of neurons in the model. Although AI researchers are trying hard to find out how much of language models do, they have barely managed to open the black box. Understanding a huge neurological network model of the human mind cannot be much easier than understanding itself.
An alternative point is to go down. The second of the two Nature Studies Focus on negative neurological networks – some contain only one neuron – nevertheless can predict behavior in mice, mice, monkeys and even humans. Since the networks are so small, it is possible for each individual to track the activity of neurons and use the data to find out how the network is preparing its behavior predictions. And while there is no guarantee that these models act like the brains they were trained to imitate, they, at least, can create trial assumptions about human and animal cognition.
There is a cost to understand. Unlike Santor, who was trained to transmit human behavior in dozens of different tasks, each small network can predict only a special task. For example, a network is expected to predict how people choose in different slot machines. “If this behavior is really complicated, you need a major network,” says Marcelo Metar, an assistant professor of psychology and nervous science at the University of New York, who led a small network study and also participated in the Sentor. “Of course. The compromise is that it is very difficult to understand now.”
This trade between prophecy and understanding is an important feature of the science -driven science -driven science. Û” Similarly, LLM interpretation research is also taking place in places like anthropic. However, for now, our understanding of complex systems – from humans to climate to protein – is far behind and behind our ability to predict them.
This story was actually published on AI in our weekly newsletter, algorithm. To get such stories in your inbox first, Sign up here.