June 4, 2021 -- "I had the weirdest dream last night."
It's a common refrain made by people whose sleeping experiences have taken them to places that surpass the waking imagination. And that's for good reason, says neuroscientist Erik Hoel, PhD, from Tufts University in Medford, MA.
Strange dreams serve an important purpose, he says, and help our brains understand day-to-day experiences in a way that enables deeper learning. Humans, he suggests in a recent study, actually expand their brain power in much the same way that artificial intelligence systems are trained to become smarter.
In fact, scientists are using deep-learning neural networks to train AI systems.
But when an AI system becomes too familiar with data, it can oversimplify its analysis, becoming an "overfitted brain" that assumes what it sees is a perfect representation of what it will encounter in future.
To counter that problem, scientists introduce a degree of chaos and randomization into their data to deepen machine learning and improve the accuracy of AI systems.
In much the same way, "our brains are so good at learning that we're always in danger of being overfitted," warns Hoel. That can lead to overly simplistic and too-familiar perceptions of the world around us. As a prompt, just like in AI training, our brains introduce chaos as we sleep, which often takes the form of outlandish dreams.
"The very strangeness of dreams and the way they diverge from waking experience gives us insight that there must be a biological function behind it," says Hoel. "Our experience with deep neural networks, which themselves were inspired by brain function, gives us a possible clue about why this happens."
This theory is interesting, says Christopher Winter, MD, from Charlottesville Neurology and Sleep Medicine in Virginia. However, he's not sure how scientists will ever be able to prove it.
It has become routine for neuroscientists to examine neural activity during sleep, but capturing dreams so they can be evaluated presents obvious challenges.
Dreams are mainly assessed through recall, Hoel acknowledges. Most people remember only fragments of their dreams, and usually just the parts that happen right before they wake.
One step researchers could take is to quantify the "weirdness quotient" of dreams, Winter suggests. A better definition of what constitutes "peculiar" could provide insight into the role dreams play in learning.
He wonders whether the dreams of people with Alzheimer's disease, for example, are as unusual as the ones of people without cognitive impairment.
But even if the study of dreams is beyond current technology, Hoel's theory hints at the idea that the brain is more malleable and capable than previously thought, Winter points out.
And neuroscience has gotten closer to dream study than many people think, Hoel counters.
A reliable way to integrate real-life experience into a dream is to perform a new task repetitively before sleep, he explains. That process triggers overfitting and, during sleep, the brain will create quirky dreams that incorporate the newly acquired knowledge.
The way the brain makes use of everyday life could one day be applied to introduce artificial dreams to improve a sleeper's ability to learn. Artificial dreams might even help minimize the harmful effects of sleep deprivation.
So while Hoel explores the idea of harnessing the reverie of the mind at rest, he recommends that people keep sharing their wildest dreams.