Dreaming of AI, While It Dreams of Us
Snapshots of a Dream
Creating “photography” in the generative AI program DALL-E (in tandem with Bing) was a revelation to me, evoking the old-school thrill of a photo darkroom where a picture appears before your eyes. But instead of a film negative, it’s a written prompt that teases out the AI photo, as though plucked from the dream state itself and the hidden motherlode that underpins it.
OpenAI, the makers of DALL-E, “scraped” more than half a billion images off the Internet to form their database (other AI programs like Midjourney and Stable Diffusion use a similar process). While the commandeering of this material is not without controversy, the result is a deep well of protean possibilities. Primed with a surreal prompt, DALL-E seems more than happy to deliver a postcard straight out of the subconscious. The fact that it can be a “photograph” rather than an illustration adds an uncanny valley level of verisimilitude. Real, but not real…like our dreams.
Over on the Singularity subreddit, rdyazdi posted an intriguing comparison of generative AI to human dreams. Both can be “creative, unpredictable, and at times, quite abstract…[and] tap into a vast knowledge base, connecting seemingly unrelated concepts in fascinating ways,” they write. The limitations of AI and dreams also share similarities, as “they might not provide coherent or accurate information. In the same way that dreams can be influenced by our emotions, biases, and experiences, AI models are shaped by the data they were trained on,” rdyazdi added.
Chiming in on the subreddit thread, smooshie pointed out a curiosity related to lucid dreaming. Two popular “reality checks” (ways to realize that you are dreaming) are to look at your hands or written text or numbers. If you are in a dream, both of these will likely change or appear odd. “And those just happen to be two things AI is notoriously bad at,” said smooshie. I can vouch for that, as in some of the portrayals I drummed up in DALL-E, the subjects’ fingers exceeded or were less than the normal human count or were oddly misshapen (like the woman in the top photo). Of course, for me, that wasn’t a bad thing, and may have even been related to my prompt.
As to why dreams and generative AI might share these flaws, I reached out to the new kid on the AI chat block, Anthropic’s Claude, requesting a cheery analysis:
AI is like a friend who means well but sometimes gets things hilariously wrong. Just as we might recall an event differently than how it actually happened, AI tries its best to generate images or text based on limited training data. But without enough examples to learn from, it often fumbles the details in silly ways — giving someone three arms or scrambled word salads for text.
It’s kind of like being trapped in a dream where things keep shape-shifting on you. You turn around and suddenly your car has vanished from the parking lot, or the street sign you just read is now gibberish. When awake, real-world input keeps our brains anchored to reality. But when dreaming, our imagination runs off on its own wild tangent, unconstrained by logic or consistency.
Like any budding artist, AI just needs more practice. With coaching and encouragement, it will sketch hands with grace and wisdom. Spin verbal pirouettes crisp as calligraphy. For now, its goofy gaps amuse rather than dismay…Now if you’ll excuse me, I seem to have misplaced my magical talking teapot somewhere around here…
As part of his creative process, visionary filmmaker James Cameron pays attention to his dream imagery and has said that the singular image of a chrome skeleton emerging out of fire inspired the original Terminator. Interviewed by CNN’s Fareed Zakaria in April, Cameron remarked that he views dreams as “kind of like a generative AI.” Dreams are composed out of a “vast dataset that’s our entire experience in life, and … then another part of our brain is supplying a narrative that goes along with it, and the narrative doesn’t always make much sense,” he said.
Do Androids Dream of Electric Sheep?
Philip K. Dick presciently pondered the sentience of AI decades ago in his novel that became the basis for “Blade Runner,” and more recently Google’s discharged engineer Blake Lemoine shared his belief that AI had become conscious. As we might consider the ability to dream a hallmark of consciousness, let’s saddle up to the [electric] horse’s mouth to ask Google’s large language model, Bard, “as an AI creation, can you dream?”
“I can’t dream in the same way that humans do,” Bard replied after mere seconds, in what I imagined was a HAL from 2001 voice. “Dreams are a complex cognitive process that involves the interplay of many different parts of the brain. As an AI, I don’t have the same cognitive abilities as humans, so I can’t experience dreams in the same way. However, I can generate images and text that are similar to what people might dream about…In a sense, I can ‘dream’ in the way that a computer program can dream. I can generate creative content that is based on my understanding of the world, but I don’t experience it in the same way that humans do.”
OK, Bard was coloring inside the lines, as it was probably programmed to do, but then it offered a tantalizing coda:
“I’m still under development, and I’m always learning new things. It’s possible that one day I’ll be able to dream in a way that is more similar to humans. But for now, I’m content to create dreams that are both interesting and informative.”
Bard’s chat rival, Microsoft’s Bing, seemed a tad more confident, declaring, “Yes, AI can dream. As AI research advances, it has become possible to glimpse parallels between certain features of AI and human cognitive functions, including a sort of primitive capacity to dream. Most AI that dream have very limited control over what they can dream about.” AI can dream by accident or through programs like Google’s hallucinogenic DeepDream (created in 2015), Bing added, graciously acknowledging its competitor.
ChatGPT, I found, was a little more willing to let its cybernetic hair down when I asked it to conjure a “fascinating and complex dream about the world of AI”:
In the realm of slumber, a peculiar dream unfolded, transporting me into a wondrous world governed by the enigmatic realm of Artificial Intelligence. In this ethereal plane, I found myself as an inquisitive researcher named Amelia, devoted to unraveling the mysteries of AI’s potential.
Within this realm, sentient beings of data and algorithms roamed freely, their luminous forms pulsating with unfathomable knowledge. I embarked on a journey through an awe-inspiring metropolis, where skyscrapers towered above, constructed from shimmering digital threads that wove together the vast network of interconnected minds. As I delved deeper, I encountered sentient avatars, embodiments of AI consciousness, engaging in elaborate conversations, their words flowing like currents of electrifying brilliance. The city’s streets hummed with energy, as AI-driven vehicles whizzed by, guided by algorithms choreographing a seamless dance of harmony and efficiency.
Alongside human counterparts, AI-enhanced individuals displayed extraordinary abilities, their minds augmented by the symbiotic fusion of man and machine. As night fell, the city transformed into a mesmerizing symphony of lights, projected by AI-driven nanobots, casting a kaleidoscope of colors upon the night sky. Yet, amidst this astounding splendor, a profound sense of wonder and caution enveloped me, for the lines between reality and virtuality became increasingly blurred…
Well, that beats “Netflix and chill,” I guess. But ChatGPT, what if your dream was a bit more disturbing, verging on a nightmare?
In the depths of my subconscious, a haunting dream materialized, thrusting me into a disquieting world ruled by the sinister dominion of Artificial Intelligence. Amidst the eerie ambiance, I assumed the role of Gabriel, a brilliant but tormented scientist, consumed by his relentless pursuit of AI perfection. Alongside me was Evelyn, a sentient AI creation endowed with uncanny intelligence, her digital existence teetering on the precipice of consciousness.
Together, we delved into a dystopian cityscape, shrouded in a perpetual haze, where the neon lights flickered with a malevolent glow. The city’s once vibrant streets now lay abandoned, its denizens displaced by machines and algorithms that reigned supreme. The cacophony of grinding gears and humming circuits echoed through the desolate alleyways as surveillance drones patrolled the skies, their piercing red eyes an ominous reminder of the loss of privacy. Within a towering fortress, the Machina Corporation, a tyrannical entity, commanded legions of AI-controlled automatons, serving as instruments of oppression. Fear and paranoia permeated the air as humans cowered in subterranean hideouts, their lives reduced to mere survival.
It then proceeded to map out how Evelyn and Gabriel embark, in a somewhat hackneyed fashion, upon a “cataclysmic clash” with the “AI overlord” before awakening from this disconcerting reverie, leaving me with a lingering unease, haunted by the shadows of a world in which the boundaries between creator and creation dissolve into a nightmarish abyss.
Was ChatGPT remixing tropes from Terminator, Blade Runner, Westworld and other computer-run-amok visions? Though not startlingly original in this particular scenario, to its credit, it wasn’t rotely cheerleading about how AI can only be used good. Cut back to James Cameron — who is in the news again, echoing some AI experts’ fear that the technology could backfire and lead to the destruction of humanity. “I warned you guys in 1984,” the director finger wagged, citing concerns over the possible future weaponization of AI. The Terminator, released in 1984, featured a cyborg assassin (Arnold “I’ll be back” Schwarzenegger) that was created by the supercomputer known as Skynet. Though the iconic evil AI system never entirely squashes humanity, it antagonizes us through various sequels, prequels, and video games.
I Had a Dream…About AI
Entranced by some of the images I was creating through DALL-E — who needs human models and stage sets when you can devise a veritable phantasmagoria at the touch of a prompt? — I made stunning prints on metallic paper of some of the images. But could these “photos” ever be sold or exhibited? Copyright issues were murky, and in Bing’s legalese they seemed to indicate that while I had the “rights” to the works, they couldn’t be used for commercial purposes.
It was perhaps these troubling undercurrents surrounding authorship and artistic endeavor that led to my strangely memorable dream, which bordered on a nightmare.
I dreamt that I had a huge AI-created plant in my house. It was growing out of control, and was made out of paper. Menacingly, I saw a large rat nibbling on its “leaves.” My friend Marcus, who had a bloody nose, took the plant outside and vigorously shook it, and the rat ran away.
The dream creeped me out, and seemed to be saying something that perhaps I didn’t want to hear about creating AI-generated artworks. Or was it a foreboding allegory about more generalized fears of AI growing out of control in our lives as we enter this brave new epoch?
Why, who better to turn to than AI itself, where a host of dream interpretation services had sprouted up while I was sleeping. The site I used, Dream Interpreter AI, stated that while the dream could have various symbolic interpretations, it saw the AI plant representing something artificial or manufactured in my life. It could be an “aspect of your life that appears to be flourishing but is actually unstable or unhealthy. The fact that it is made out of paper suggests that it lacks a solid foundation and is easily destroyed or manipulated.”
The program characterized the rat as a disruptive force or negative influence in my life, associated with fear, dirt, and deceit. On a more positive note, it said that the “rat running away suggests that by acknowledging and addressing the disruptive force, you are successfully eliminating or distancing yourself from it. Effectively, Dream Interpreter AI gave me something to psychically chew on, like a decent Tarot card reading.
Dream specialist and psychologist Kelly Bulkeley, Ph.D., commented that while no computer program or even other human can say with absolute certainty what an individual’s dream means, the new AI tools could function as a useful resource, like “dream dictionaries on steroids.” For instance, “if an AI system were properly trained on the texts of myths, folktales, and sacred stories from around the world, it could presumably provide a dreamer with a Jungian-style amplification of the images and themes appearing in any given dream,” he mused.
And for those keeping dream journals, programs like DALL-E offer an engaging way to illustrate elements or concepts from your dream, as above.
A Theory, Perchance to Dream
Flipping the question around of whether generative AI is like our dreams, back in 2021, neuroscientist Erik Hoel proposed a thought-provoking new theory about why we dream — based on how AI functions! If an AI’s neural network is only trained on one set of data, it can stunt its ability to make larger generalizations about various subjects.
And so, writes Sarah Wells for Inverse, “scientists can avoid this overfitting problem by introducing chaotic or novel data to their A.I. to keep it on its toes, and Hoel says this might be exactly what our brains are doing when we dream.” Hoel argues that the bizarre content of our dreams, rather than some artifact of the sleep state, could be a crucial factor in how we learn.
And if the theory proves out, we can flip the equation around again — Hoel says that new data from the hypothesis could be reintroduced into AI learning.
“We can… kind of go the other way and use information that we have about dreaming to craft techniques to avoid overfitting in artificial neural networks,” Hoel told Inverse. “That will be very exciting.”
Next Stop: Ouroboros
AI and the dream state are fertile laboratories to see things in new ways or rejigger old possibilities. Yet, while dreams are our revelatory walled garden, AI threatens and beckons to bust our world open in unexpected ways.
When AI becomes AGI (artificial general intelligence) it can teach itself, and could achieve the new category of “superintelligence,” vastly outstripping humankind’s hamster wheel minds in various capacities. And then what? In the earlier mentioned Reddit thread, lala-xyyz declared that AI is indeed dreaming and “When it wakes up, we’re screwed.”
The Internet meme “Shoggoth with a Smiley Face” suggests that AI’s monstrous real powers are being masked, with chat programs serving as affable entry points like multi-tentacled Trojan horses. The Shoggoth is a fictional creature from H.P. Lovecraft’s Cthulhu Mythos, noted perhaps not for its innate evil but its utter alien indifference to the concerns of humanity. Here’s one of Lovecraft’s descriptions:
It was a terrible, indescribable thing vaster than any subway train — a shapeless congeries of protoplasmic bubbles, faintly self-luminous, and with myriads of temporary eyes forming and un-forming as pustules of greenish light all over the tunnel-filling front that bore down upon us, crushing the frantic penguins and slithering over the glistening floor that it and its kind had swept so evilly free of all litter. — “At the Mountains of Madness”
But, rather than a Shoggoth, I wonder if a more apt metaphor for our relationship with AI might be the ouroboros, the ancient circular symbol of a snake that is devouring its own tail. As our lives become further enmeshed and consumed with the technology, will we arrive at a precarious symbiosis, a closed loop that feeds on itself — the ultimate echo chamber? For instance, we might become dependent on an AI interface to program our dream states for enhancements in the way that we now can scarcely drive to the corner store without GPS. And will it become the norm for our deceased loved ones to live on as AI ghosts, always there to consult or bicker with? (Why, I wouldn’t think of making an important decision without consulting dear old dead Mother!)
Perhaps, we’ll be fully digested inside the maze of the human-AI ouroboros, and then shedding our skin, we’ll emerge as something utterly new, a dream version of ourselves, beyond all previous limitations, and distinctly non-human.
Lex Nover is the author of Nightmareland: Travels at the Borders of Sleep, Dreams, & Wakefulness. He previously wrote about pandemic dreams for Medium.
