Can AI unlock our dreams—and creativity?

Earlier this year, a startup announced a new $2,000 headband meant to be worn during sleep and detect when a person enters rapid eye movement or REM. That’s the land of deep sleep where dreaming takes place, and it’s at that point that the headband (in theory, according to the company) kicks on a multimodal AI model, trained on brain data, to help someone dream. The data is used to send signals via ultrasound holograms—acoustic waves that can penetrate human tissue and create 3D images—from the device to the prefrontal cortex of the brain. At that point, it might be able to induce lucid dreaming—the state where you’re dreaming, and you know it.

It's a bold new territory for AI—the activation of dreams via machine learning—and it’s one of the newest questions that psychologists, sleep specialists, and dream researchers are now pondering: Can AI help us better understand our dreams?

“The effort to use big analytic systems to figure out dreams—that goes back centuries,” says Kelly Bulkeley, a religion psychologist who specializes in dream research. “It’s just that we have really good tools now, better than quill and ink.”

Analyzing our dreams

It’s often disorienting to wake up after a strange dream without knowing why something in particular was happening in our minds while we slept. Machine learning can not only interpret our dreams for us, but it can also create images of those interpretations. 

One new AI-inspired dream journal, Elsewhere.to, is already trying to peel back some of the layers of the dream world. Bulkeley, an unpaid research adviser for the app’s makers, says users can record what they saw or experienced in their dreams, and the app will come back with various ways to comprehend what was going on. The AI will create visual approximations of characters or creatures from the dream. (If you dreamt about a cat, you’ll get a picture of it.) And thanks to the app’s training on the teachings of famous psychologists and philosophers, it can analyze the meaning of a dream from a variety of perspectives.

“If you want a Freudian interpretation with a fairy-book-tale illustration style, you can do that. If you want a Jungian analysis with a modernist visual image, you can do that too,” he says. 

Applying machine learning to dreaming in this way isn’t just some neat trick. Over time, the AI interpretations of our dreams will become something like a diary of the subconscious, one that includes more than just our recollections of what we thought of during sleep, but also what those thoughts might mean.

“Dreams are dynamic. They’re not just raw material waiting for us to pick them,” says Bulkeley. “This is the fun thing about looking at not just a single dream, but getting into a dream journal. That dynamism gets going into your higher self, your unconscious psyche.” 

That level of attention, Bulkeley adds, could form the basis of how AI can help induce lucid dreaming by optimizing people for the REM stage of sleep. Being more in tune with our own dreams while we’re conscious will help stimulate recall—will help, in other words, for us to go to sleep and know that we’re most likely going to be served up some picture that we’ve seen before. 

“The more attention we pay to our dreams, generally the more our dreams will pay attention to us,” he says.

Unlocking why we dream

Yet why we dream doesn’t have a simple answer. And researchers think that one of the applications of AI will be to unearth the fundamental question of why dreaming happens in the first place.

“Because we can’t access dreams directly, we just don’t know much about [them]],” says Remington Mallett, director of the Dream Mining Lab at the University of Montreal. “AI methods are going to make that less and less of an issue going forward.”

Navigating the subconscious with AI may, perhaps, give more credence to the idea that the dream state is rooted in evolutionary theory. Various researchers believe, for example, that early humankind experienced nightmares about being chased by predators. A dream is a safe place for the mind to ponder what would happen if a saber-toothed tiger ran us down, guaranteeing we’d be the special on the evening menu. Dreaming, then, is a way to get more comfortable with frightening situations—almost a practice run for the real world. No doubt there are workers out there who dread having to give a presentation in front of a big audience. A nightmare about stumbling over your words the next day can help us shed some of the fear.

Erik Hoel proposes a different way of looking at things. A neuroscientist, Hoel is author of The Intrinsic Perspective, a Substack publication about the confluence between the sciences and the humanities. Previously he was a Tufts University researcher on consciousness, where he posited that dreams were like a built-in reset button for our brains. Many machine learning models can become so good at performing one repetitive task over and over that programmers introduce random variables in the data to prevent the AI from becoming what Hoel calls “overfit”: too good at doing one specific task. The same concept applies: Dreams are random noise injected into the monotony of daily life to keep our brains in good health. 

For knowledge workers, that makes dreaming that much more important. Consider, for instance, the times in life when you’ve been stuck trying to finish a presentation, a report, or a problem that seemed intractable. Oftentimes a night of sleep will alleviate the issue. “This fits with anecdotal reports of plateauing in terms of performance on a task, like a video game, only to sleep and have increased performance the next day,” Hoel said in an interview in 2021. “There is also the long-standing traditional association between dreams and creativity.”

Hoel expanded on the idea in a paper he published earlier that same year, where he fully described his overfitted brain hypothesis. “Sleep,” he wrote, “improves abstraction and reasoning on tasks.” Sleep, and dreaming more specifically, allows us to step away from a particular challenge and let the subconscious take over. In the dream world, we’re having general thoughts, divorced from the difficulties of the day, and that can help our creative impulses. Recall the number of times in life you might’ve said you need to clear your head or get a good night’s sleep. The dream world kickstarts insights into complex problems that require creative solutions. 

The next frontier: manipulating dreams

Influencing dreams, to be sure, can be fraught territory. Sleep is unguarded time when our rational decision-making is turned off and our deeply personal and intimate inner selves take over. That makes it vulnerable to tampering.

“Concepts of mind-control and inception, introducing ideas into an individual’s memory without their consent or even conscious awareness, harken back to public concerns about subliminal persuasion,” wrote Michelle Carr, a fellow dream researcher who leads Mallett’s lab at the University of Montreal, in a research paper published in 2020. 

But that concept of control while someone is dreaming can be beneficial in myriad ways. As Carr mentioned in that same paper, there have been studies that paired the scent of cigarettes with that of rotten fish while people slept. Over time, participants eventually drew a correlation between the two, which in turn led to a reduction in cigarette smoking. That’s to say, the subconscious changed people’s behavior.

There are other examples of people suffering from post-traumatic stress or other anxiety disorders finding help in the dream world. One recent study had people—during the day, while they were awake—recite happy endings to their nightmares. At the same time, they were hearing specific piano chords at regular intervals. Then, at night, they wore sleep-engineering headbands that detected when they entered the REM stage of sleep. The headbands, in turn, transmitted the same piano chord continually—which led people to experience more positive dreams and fewer nightmares.

There’s also a possibility that AI can be a tool to improve sleep quality. At Thomas Jefferson University, researchers are looking into how various brain waves encourage different mental states. Alpha and delta waves are those associated with relaxation and deep sleep, respectively, and it’s thought that playing certain sounds can boost those waves in the brain. Imagine an AI headband that begins picking up when a user is just about to enter REM, and syncs with a pair of earbuds that begin playing music or noises that encourage the wearer to drift away peacefully into their subconscious. 

Whatever is to come, researchers like Bulkeley believe that applying AI to dream analysis will make the visions we see in our sleep more fun than they’ve ever been. “It’s super cool,” he says, “and it’s going to make the dream world more exciting.”