Researchers have created a new neural network called RTNet that makes decisions like humans. This breakthrough was made by scientists Farshad Rafiei, Medha Shekhar, and Dobromir Rahnev, and published in Nature Human Behaviour.

RTNet is different from other neural networks because it can make decisions in a way that mimics how humans think and decide. Unlike other models that make the same decision every time, RTNet can vary its decisions, much like people do. This makes it a better model for understanding human behavior.

To test RTNet, the researchers used images of handwritten numbers. They collected data from 60 human volunteers who had to recognize these numbers and rate their confidence in their answers. The RTNet was then tested with the same images. Amazingly, RTNet’s responses matched the humans' in accuracy, response time, and confidence.

RTNet uses a method called a Bayesian neural network, which makes decisions based on probabilities. It also accumulates evidence before making a decision, just like humans do. This means RTNet doesn’t always make the same decision but varies depending on the evidence, making it more human-like.

One important feature of RTNet is the "speed-accuracy trade-off." This means that if it has to make decisions quickly, it might be less accurate, just like humans. The researchers found that RTNet performed better than other models, especially when decisions had to be made quickly.

This new neural network is a big step forward in making machines that can think and decide like humans. The researchers hope to train RTNet on more varied datasets and apply its methods to other neural networks in the future. This could help machines not only mimic human decision-making but also assist in making some of the many decisions we face every day.