Back to selection
Emergence
New work in new media by Deniz Tortum
Sleep Dealers: Engineering Dreams with AI
ReaDream (courtesy of Ninon Lizé Masclef) Saelyx Finna told me about a dream she had. In it, the filmmaker was trying to alter the dream itself by typing prompts into an AI interface. She wanted to change the dream while she was dreaming it, but it wasn’t working and the dream went dark. When she woke up, she thought about the logic of her dream. “Of course it didn’t work,” she said about the AI intervention. “I was trying to access an external tool to change my internal experience.”
Finna’s dream points to something very contemporary: how quickly our inner lives are becoming entangled with changing technology. Her dream also serves as a metaphor for current technological and commercial development around dreamspace, which is becoming a new frontier of computation, design and venture capital.
In her first feature, Under the Dream, Finna explores this evolving and speculative universe of dream tech. The film journeys through its broad constellation of emerging tools and research aiming to influence, interpret and even record dreams, often using brain-computer interfaces, EEG sensors, audio cues and AI models.
The field of dream tech first caught my attention a few years ago, when a friend shared an MIT course titled “Engineering Sleep and Dreams.” I was struck by seeing “dreaming” and “engineering” in the same sentence. Dreaming had always lived on the edges of my intellectual map—part mysticism, part fringe science—and now it had a syllabus. I wondered, can we also approach dream engineering as a cinematic and media-theoretical field? Could dreams become a medium in themselves? Could cinema be reimagined not just as a metaphor for dreaming but as a specific form of it? And would this then signify a shift in how we conceptualize ourselves and our minds?
One of the co-instructors of the MIT dream engineering course was Adam Haar, whose Ph.D. work at MIT Media Lab’s Fluid Interfaces group brought new energy and new eyes to the field. His project, Dormio, is a targeted dream incubation device designed to guide dreaming during hypnagogia, the liminal state as we fall asleep. The system monitors brainwaves using EEG sensors and plays curated audio prompts as the subject drifts into sleep, subtly shaping the content of their dreams.
The mechanism behind Dormio echoes an older technique, famously used by Salvador Dalí and Thomas Edison, in which a key (in Dalí’s case) or metal ball (in Edison’s) would be held while falling asleep, waking the dreamer when it dropped to the floor. The idea was to catch the creative state between waking and dreaming. Haar’s version replaces the steel ball with sensors and algorithms, bringing greater precision.
Haar works in science, technology and art. He develops his own artistic projects and regularly collaborates with other artists (he’s a collaborator on Boreal Dreams, directed by Jakob Kudsk Steensen, which will screen at Tribeca this month), and is exploring how Dormio might evolve into a commercial product.
When I asked Haar whether dreams could become a form of media, he hesitated. “It depends what we mean by media,” he replied. People tend to fixate on the idea of control and the power to shape dreams. With Haar’s method, he can get nine out of 10 people to dream about a specific subject: a tree, a volcano, Angelina Jolie. But beyond that prompt, the dream belongs to the dreamer: “One person might become the tree, another might talk to trees, someone else might walk through a forest from their childhood.”
For Haar, dream engineering is “more like a prayer than computer code; more like planting a seed than editing a genome.” Creativity, he says, lives in that space between “a little bit of control and a loss of control.”
Haar recently applied this approach to an upcoming film by artist Philippe Parreno, made in collaboration with Charlotte Gainsbourg and sound designer Nicolas Becker. Haar acted as a “dream incubator,” guiding Becker’s dreams with his method during hypnagogia. Haar monitored Becker’s sleep with an EEG, tracking his descent into hypnagogia. At just the right moment, Haar introduced specific words and sounds connected to the film’s themes. Becker would wake after a short sleep in this altered state and use his dreams as raw material to compose the film’s soundscape. This was a creative process suspended between sleep and waking for a film about dreams, partially composed in dreams.
I was curious whether dream incubation could evolve into an art form in its own right. When I asked Haar about this, he responded with a film: Apichatpong Weerasethakul’s Memoria, which famously put many people to sleep. Before I watched Memoria, I’d heard from friends that they all fell asleep at a similar point in the film. And when I watched it, I also fell asleep for a short time at the same point. It was a strange sleep which became part of my memory of the film.
We tend to limit what cinema is and could be. In recent years, cinema has all been about “storytelling.” However, cinema remains an expansive tool that can become many other things. And in Weerasethakul’s case, it may be a form of dream seeding—a form that could give access to an intimate creative state for the audience. As Haar puts it, “When the audience sleeps at the film, the emotional and aesthetic core of the film explodes into something new, and everyone gets to catch the film sleeping at the same seat, but it becomes so many different dreams.”
Ninon Lizé Masclef is a media artist and research affiliate at the same lab where Haar developed Dormio. Her work has an ambitious long-term goal: to reconstruct and visualize dreams by decoding the relationship between mental imagery and brainwave activity. ReaDream is her artistic research project using EEG data collected from dreaming subjects to generate VR environments. She takes data from the sleeping brain and translates them into text. Then, using AI tools that can turn text input into 3D models (such as DreamFusion and ClipMatrix), she creates immersive virtual environments inspired by those dreams.
Parallel to her artistic work, Masclef conducts controlled lab studies and is finalizing her research publication. “We have very promising results for reconstructing perceived images from brain activity,” she tells me. This new study, titled MindSpace3D, places subjects inside virtual environments where they view 3D objects while their brain activity is recorded via EEG. After viewing, they’re asked to mentally recreate what they’ve seen. The EEG data from the perception and visualization process are then used to generate a 3D reconstruction. The results: the original object and the reconstructed one are often strikingly similar.
This outcome builds on the neuroscientific insight that the brain shares overlapping pathways for visual perception and visual imagination. Dreaming, Masclef argues, belongs to the same neural neighborhood. If we can decode imagined images, we’re not far from decoding the dreamed images.
However, there is a long road to dream reconstruction. A major obstacle is the lack of large individualized datasets; to decode one person’s EEG requires enormous amounts of data, and there’s no universal decoder. Still, Ninon sees this more as a solvable systems problem. “It’s an engineering and hardware challenge,” she says. With better sensors, larger data sets and better models, she believes visual reconstructions of dreams could be viable within the next 10 to 20 years.
“Dream reconstruction,” she says, “could be a tool for negotiating with the self.” Inspired by Michel Jouvet, a pioneering dream researcher who started working on REM sleep in the 1950s, Masclef is drawn to his hypothesis that dreams are a form of genetic and neurological programming. She sees dreams as portals into our most embedded emotional and behavioral patterns. Having more access to dreams is an undiscovered cognitive agency, a new interiority that is still largely untapped.
One of the finalist projects at this year’s SXSW Innovation Awards was a lucid dreaming device called The Halo. This prototype, which competed in the WTF (What the Future) category, was created by the startup Prophetic AI with the provocative mission of providing lucid dreaming on demand. “Abraham, Muhammad and Buddha received their prophetic wisdom in their dreams, so that’s why the name of the company is Prophetic,” Eric Wollberg, the CEO of the company, says, adding, “Imagine a world where everyone has this connection.”
The Halo uses EEG and an AI dataset to identify whether the user is going into REM sleep. When they are in this stage, where most dreams occur, the device sends a pulse of ultrasonic energy to the dorsolateral prefrontal cortex. This region of the brain is responsible for decision making and working memory, and it is also noticeably very quiet during REM sleep as opposed to a waking state. So, if the device could activate this region of the brain during REM, the hypothesis is that the subject would potentially go into a lucid phase and gain awareness of their dream. The ultrasound intensity, Wollberg notes, is below FDA limits.
I ask him if his device actually works. Wollberg admits that Prophetic has not yet “cracked lucidity.” The Halo can’t induce lucid dreams, but he believes what they see so far is promising, with subjects reporting vivid dreams with high memory recall.
Even though their product is not demo-ready, Wollberg is talented at creating hype and coining new concepts, such as the “qualia takeoff,” which refers to the increasing amount of qualia per media technology, from writing to photography to lucid dreaming, and “consciousness experience (CX),” a potential design/human computer interaction term that could be used interchangeably with the analogous term “dream.”
Wollberg proposes a dreamspace that is not fleeting but instead a place you can return to. Like the memory palace technique—the method of loci—this inner world could be cultivated over time, shaped with intention. “It’s not a place that starts anew every night,” he tells me. “It’s something that builds.” I reply that it sounds like a longevity project, almost extending your life. He agrees. “You spend 30 percent of your life asleep, and eight percent of [your life] dreaming,” he says. “We could be giving that time back to people.”
Finna sees efforts like Prophetic AI’s making dream space controllable as importing the logic of the waking world into sleep. “We’re bringing the wake agenda to the dreamspace,” she says. “By shaping and controlling our dreams, we’re making them more like waking life.”
But what if we reversed the direction? Finna suggests that instead of taming dreams to resemble consciousness, we let dreaming reshape our waking perspective. “What if we lived in a dreamcentric, rather than wakecentric, reality?” she asks. “What if we all started paying attention to our dreams tomorrow? What could change?” This path, she argues, could enable dreams to help us with many things, from well-being to social connection.
Her film, Under the Dream, explores what it might mean to learn from dreams instead of colonizing them with waking intentions. “We’re not giving ourselves the space to imagine,” she says. “These capacities are going dormant. It’s like atrophy.”
In her view, the 20th century translated the mythic dimensions of dreams into cinema. “The invention of cinematic language”—its logic of montage, nonlinear time and symbolic imagery—“was humans channeling a millennia-long fascination with dreams,” she says. But what comes next? “What new forms might emerge that could redefine our paradigms of perception?”
For Finna, consciousness is a continuum. “If we’re honest with ourselves, we spend much more time in altered states than we acknowledge.” These include hypnagogia; hypnopompia, the blurry emergence right before we wake up; and many other unrecognized altered states. “Our minds are capable of far more expansive perceptual experiences than we give ourselves credit for,” she says. “And I feel like spending time with dreams is the way we’re going to tap into those abilities.”
She’s looking for a new kind of interiority, one that remains off the grid—for now.
I keep thinking about Finna’s dream of prompting ChatGPT from inside a dream. As LLMs become more embedded in our everyday processes, we risk outsourcing our thoughts and our ability to think and even dream.
Lately, I’ve come across more thinkers writing about what feels like a slow erosion of interiority. In Scorched Earth, Jonathan Crary argues that with all its tools, platforms and services, digital capitalism is dispossessing us from our thoughts. In a recent Substack post, writer Venkatesh Rao wonders whether we’re losing our “internal monologue” to language models as we offload the work of thinking to external tools.
In conversations about dream engineering, what I keep hearing, beneath the talk of EEGs and large data sets, is a desire to capture an internal place our devices and technologies can’t reach—a place yet untouched by algorithms and prompts which might be the last private interior.
