UCLA researchers are integrating artificial intelligence and psychoanalysis to unearth the mechanisms and purpose of human dreaming.
Researchers gathered at The Science of Dreams symposium Wednesday and Thursday to discuss the neuroscience and quantification of human dreaming. Mark Blagrove, a visiting psychology professor from Swansea University, said dreams occur during the rapid eye movement, or REM, stage of sleep involved in memory consolidation, which gives rise to multiple theories on dreams’ purpose and mechanisms.
One theory explains that dreams simulate parts of a person’s waking life, while another contends dreams represent a person’s emotional state or trauma through metaphors, he said.
Blagrove added an influential past study argued the evolutionary purpose of dreaming is to overcome fears in real life by putting them in newly imagined contexts. When people cannot manage their fears in those dreams, those dreams can turn into nightmares.
“Dreaming might reflect functional neural processes during sleep, which results in the bold statement … that dreaming is the poor man’s fMRI,” Blagrove said.
Despite the many different theories on the true effects of dreams on the body, Blagrove said the purpose and mechanisms behind dreams are still not completely understood and warrant further study. He added dreams not only reveal the emotions of an individual, but also their social bonding with others.
“There’s a similarity between dream sharing and blushing,” he said. “It’s an honest signal (from the brain) that can be advantageous to share with others.”
In order to better understand dreams and share them, UCLA researchers also discussed using modern technology to compile, analyze and perhaps mirror human dreams.
Vwani P. Roychowdhury, a UCLA electrical and computer engineering professor, said researchers can use an artificial intelligence model to find, classify and analyze dreams within a database of dream descriptions on a large scale.
“Scientists were excited about computers because they could build models to simulate the real world,” he said. “When you look at dreams, you’re using dreams to model the models that drive us.”
In his recent study, Roychowdhury said he applied this model to process and aggregate millions of social media posts from mothers in the anti-vaccination movement. The computer model extracted key words often mentioned in these posts, such as “religion,” “doctor” and “rights,” he said. The model then found the relationships between these words to create a road map and conclusion on why mothers wanted to avoid having their child vaccinated.
“Social media posts are like dreams: Both are pieces of a person’s thought process that can be put together to show how the human mind (works),” he said.
Similar to the way it analyzed social media posts, the model can also analyze descriptions of individuals’ dreams, Roychowdhury said. Dream databases containing textual descriptions of people’s dreams, along with their genders and backgrounds, provide ample data for a computer model to analyze, Roychowdhury said.
For example, a war dream from a veteran may involve certain elements such as the death of a comrade, bombings and personal sacrifice. The model would take these elements and identify them in other dreams, along with who else had those dreams.
“You can see what kinds of people get more war dreams. Is it war veterans or people watching war movies, or other people?” Roychowdhury said.
Bilwaj Gaonkar, a postdoctoral researcher in neuroscience, said recent advances in AI have also opened the possibility for computers to “dream” similar to the way humans do.
Deep learning in AI, for example, is a type of algorithm modeled after the neural network of the human brain, he said. In the brain, neurons are connected to one another, sometimes with multiple neurons connected to one. In computers, a machine contains functions that convert input values into an output value, connected to one another in a similar fashion, Gaonkar said.
Deep learning has allowed computers to be able to identify certain sets of pictures with a humanlike accuracy of 90 percent, Gaonkar added. Using deep learning, researchers at Google have created DeepDream, an algorithm that produces psychedelic images. To create these images, Gaonkar said researchers feed the algorithm an image of white noise, and the algorithm modifies the image until it can recognize it as something else, such as a painting or a dog.
“Suppose someone handed you a piece of paper and asked you to draw a chair,” Roychowdhury said. “While you’re drawing, you might make a mistake, erase something a bit, redraw, modify some parts. That’s what the computer is doing.”
Gaonkar said this method of deep learning may also reflect how the neurons in human brains work while we dream. When people dream, they may be taking the visual input from their waking life and adjusting it in various ways to output dreams.
“This is similar to Freud’s view where the whole system works backward, the brain is recalling daytime memories in the nighttime with dreams,” Blagrove said. “(This belief) keeps the science side and Freud’s side happy.”