Neural networks, a type of AI, can now combine ideas in a way that more closely mimics human learning. A new study published in Nature challenges the belief held by some cognitive scientists since the 1980s that neural networks are not accurate models of human thought. Neural networks can now acquire the ability to think more like humans through practice, according to the study.
The researchers tested both AI models and human volunteers using a made-up language and found that a method called meta-learning for compositionality (MLC) allowed neural networks to match or surpass human performance in understanding the underlying grammar rules. MLC also outperformed other AI models on additional tasks involving interpreting written instructions and sentence meanings. However, there is still room for improvement in the neural networks’ ability to generalize beyond the training it received.
The whytry.ai article you just read is a brief synopsis; the original article can be found here: Read the Full Article…
