U.S. researchers develop AI that thinks like humans

[Image source: Pixabay]

Researchers from the New York University (NYU) Psychology Department, led by Professor Brenden Lake, have successfully created an artificial intelligence (AI) system capable of systematic generalization, a cognitive ability previously believed to be unique to humans.

This groundbreaking research, published in the journal Nature, has the potential to revolutionize the field of AI and significantly impact the development of AI applications such as ChatGPT.

The human mind possesses a remarkable capacity known as compositional generalization, allowing individuals to combine learned concepts with new information to expand their knowledge. For example, when humans learn the concept of “jumping,” they can easily understand the meaning of a phrase like “jumping with hands up.”

This cognitive ability has long been associated with the human domain, but the latest research has successfully transferred it to AI.

The NYU research team introduced a novel AI training method called “meta-learning for compositionality.” This approach involves the AI system identifying logical structures within data, evaluating their connections, and autonomously improving its algorithms based on the results.

To assess the effectiveness of their methodology, the researchers conducted experiments involving both the meta-learning-trained AI and human participants.

Test subjects were presented with unfamiliar, nonsensical words such as “zup” and were tasked with deducing their usage with the aid of hints.

The results demonstrated that the AI’s ability to make inferences closely matched that of human participants, indicating that it was capable of systematic generalization.