Georgia Tech develops human-like decision-making AI
This development could lead to more accurate and reliable AI systems, potentially reducing human cognitive burdens.
Researchers at Georgia Tech are training neural networks to make decisions more like humans, a significant step forward in AI. Traditional neural networks make the same decision every time, unlike humans who can vary their decisions based on context. The human-like decision-making is being integrated into AI to improve its reliability and accuracy.
In a study published in Nature Human Behaviour, Georgia Tech’s team introduced a neural network that mimics human perceptual decision-making. Using a Bayesian neural network (BNN) and an evidence accumulation process, the model produces responses with slight variations, much like human decisions. When tested on the MNIST dataset of handwritten digits, the model’s accuracy, response time, and confidence levels closely matched those of human participants.
The breakthrough follows similar research, such as the identification of specific neurons responsible for AI decisions and AI systems mimicking biological models with a small number of neurons. By making AI more human-like, researchers hope to create models that can not only replicate human decision-making but also alleviate some of the cognitive load from the thousands of decisions people make daily.
The Georgia Tech team plans to train their model on more varied datasets and apply the BNN model to other neural networks. The approach could lead to AI that better understands and rationalises decisions, paving the way for more advanced and human-like AI.