An important advancement in AI is being made by Georgia Tech researchers who are teaching neural networks to make judgments more like humans. In contrast to humans, who can adjust their conclusions based on context, traditional neural networks always reach the same conclusion. AI is incorporating human-like decision-making to increase its accuracy and dependability.
A neural network that simulates human perceptual decision-making was presented by Georgia Tech researchers in a paper that was published in Nature Human Behaviour. Based on an evidence accumulation process and a Bayesian neural network (BNN), the model generates replies that exhibit subtle changes, akin to those made by a human. The accuracy, response time, and confidence levels of the model closely matched those of human participants when it was evaluated on the handwritten digits dataset from the MNIST dataset.
Similar research has led to breakthroughs, including the identification of particular neurons that are in charge of AI decisions and the development of AI systems that use a small number of neurons to replicate biological models. Researchers aim to develop models that can mimic human decision-making and reduce some of the cognitive burden caused by the thousands of decisions humans make every day by making AI more human-like.
The Georgia Tech group intends to apply the BNN approach to different neural networks and train their model on a wider range of datasets. This strategy might result in AI that can reason and comprehend decisions more effectively, opening the door to more sophisticated and human-like AI.