Meta (formerly Facebook) has announced a long-term artificial intelligence (AI) research initiative to better understand how the human brain processes speech and text, and to develop AI systems that learn in the same way that humans do.
Meta said it is comparing how AI language models and the brain respond to the same spoken or written sentences in collaboration with neuroimaging center Neurospin (CEA) and Inria.
“We’ll use insights from this work to guide the development of AI that processes speech and text as efficiently as people,” the social network said in a statement.
Meta has spent the last two years analyzing how the brain processes words and sentences using deep learning techniques applied to public neuro-imaging data sets.
From a few examples, children learn that “orange” can refer to both a fruit and a color, but modern AI systems cannot do this as efficiently as humans.
According to meta-research, language models that most closely resemble brain activity are the ones that best predict the next word from context (like once upon a… time).
“While the brain anticipates words and ideas far ahead in time, most language models are trained to only predict the very next word,” said the company.
Unlocking this long-term forecasting capability could aid in the improvement of modern AI language models.
Meta recently discovered evidence of long-term predictions in the brain, an ability that continues to challenge today’s language models.
Most language models today would predict the next word, “time,” for the phrase “Once upon a…”, but they are still limited in their ability to anticipate complex ideas, plots, and narratives, as people do.
In collaboration with Inria, the Meta research team compared various language models to the brain responses of 345 volunteers who listened to complex narratives while having their brains recorded with fMRI.
“Our results showed that specific brain regions are best accounted for by language models enhanced with far-off words in the future,” the team said.
