THE UNEXPECTED CONVERGENCE OF HUMAN AND ARTIFICIAL IMAGINATION
Surrealism is based on the belief in the omnipotence of dreams, in the undirected play of thought. --Salvador Dalí
Having played with several early AI-video platforms such as Luma, Runway, and Stable Video, I’m often struck by the similarities between AI video and dreams (at least my dreams, which are usually pretty fun and interesting). Things make some type of sense, but others defy logic and the laws of physics. They also might reveal something interesting about creativity. Or not. They might just be strange - but the similarities run much deeper than just being strange. While one emerges from the human mind and the other from lines of codes and neural networks, they both blend reality with fantasy and can create new visions. These similarities should make us reconsider our understanding of imagination and creativity.
My dreams are usually very visual - Luma
Visual Nature
Perhaps the biggest similarity between human dreams and AI-generated video is their surreal visual nature. For many, dreams unfold as vivid mental imagery, creating immersive visual encounters that can feel remarkably real. They can range from mundane replays of daily life to fantastical scenarios limited only by our subconscious. Similarly, AI-generated video produces visual content that can span a wide range from simple animations to complex, lifelike scenes. Both engage our visual senses in novel and interesting ways. The two also tend to incorporate surreal or impossible elements. In dreams, we might be flying without wings, conversing with a deceased relative, or navigating environments that defy the laws of physics. However, they feel perfectly natural when you’re dreaming but it slowly falls apart as you wake up. AI-generated videos, particularly those created by the latest AI models like Runway v3 and Luma 1.5, can produce similarly fantastical visuals. They might blend different art styles, create physically impossible objects, or depict scenarios that could never occur in our reality. This shared surrealism allows both dreams and AI videos to transcend the limitations of the physical world, opening up infinite possibilities for visual expression and experience.
Albert Hoffman’s Tour De Tripping - Runway
Non-linear Narratives
Dreams and AI videos tend to share non-linear narratives. In dreams, we often experience storylines that defy conventional logic or temporal continuity. A dream might begin at school only to abruptly shift to Paris, then to some lava palace out of Mustafar. Characters may change identities mid-scene. The usual rules of narrative structure are usually out the window in dreams. Similarly, AI-generated videos can product sequences or scenes that lack a clear linear progression. The outputs might jump between disparate themes or settings in ways that mirror dreams. These seemingly random shifts add to the surreal and often disorienting nature of both. Sudden shifts and unexpected juxtapositions in both dreams and AI videos challenge our usual perceptions of time, space, and causality, creating experiences that are uniquely immersive and thought-provoking. Everything is strange, but at the same time, right.
Defying the laws of physics - Luma
Lack of Complete Control
Unless you are a lucent dreamer (I’ve tried and failed…) the lack of control is a big part of dreaming. We are passive participants in narratives and scenarios that unfold around us. We find ourselves in unfamiliar settings, interacting with unexpected characters, facing situations that seem to arise from nowhere. This lack of control contributes to the mysterious and often profound nature of dreams, as the subconscious mind weaves together experiences and emotions in ways that can surprise, delight, or disturb the dreamer. The process of creating AI video also involves an element of randomness and unpredictability that mirrors the uncontrolled nature of dreams. While AI models are trained on specific datasets and can be guided by prompts or parameters, the exact output of an AI video generation process isn’t yet predictable (it still takes a few attempts to get a decent output). The AI may combine elements from its training dataset in unexpected ways, producing visuals that surprise even its creators. This unpredictability is often embraced as a feature rather than a bug, as it can lead to novel and creative outputs that a human might not have conceived of. Just as we might wake up surprised by the content of our dreams, users of AI video generation tools often find themselves amazed by the unexpected combinations and creations that emerge from the AI’s process. The emergent nature of dreams and AI video means limited control.
Are those fingers or toes? Luma
Subconscious/ Dataset Influence
Now let’s talk about the source of these surreal visions – in the case of humans, our dreams seem to be littered with content from our subconscious, whether it’s our fears or small from our waking hours. They also draw on our personal experiences, memories and emotions in our subconscious. The result is surreal dream narratives. We’ve all had dreams about our work or school anxieties and encountered people we know but haven’t seen for years. Even when our dreams are truly weird, they’re often still rooted in our lived experiences and accumulated knowledge. This personal influence gives dreams their uniquely intimate and sometimes revelatory quality. Similarly, AI-generated videos reflect the patterns and elements from their training data. The visual content they produce is a recombination and extrapolation of the video content used to train the models. If an AI model has been trained on natural landscapes, it might excel at generating realistic forest scenes or sand dunes. However, if it were trained on abstract art instead, the AI might produce videos with more surreal elements. Just as dreams can reveal subconscious preoccupations or biases of the dreamer, AI-generated videos can also showcase their biases and limitations present in the training data. Dreams and AI video reconfigure existing information, whether from the human mind or digital dataset.
We’ve all had some version of this one - Luma
Symbolic Representation
Because of the emergence of our unconscious during dreams, we often notice the symbolism. In our dreams, our mind uses symbols to represent complex ideas, emotions, or experiences that might be difficult to express directly. These dream symbols can be deeply personal or drawn from Jungian collective cultural archetypes. Dreaming of falling might symbolize a loss of control in one’s waking life, while a locked door could represent a challenge or opportunity the dreamer is hesitant to confront. And of course, a cigar is just a cigar. But the symbolic language of dreams allows the subconscious to communicate in rich, multi-layered ways. While AI doesn’t have a subconscious, it can still generate visuals that we might interpret symbolically. AI might create a scene where objects morph or blend in ways that could seen as representing transformation or the fluidity of ideas. The AI might juxtapose contrasting elements in a single frame, inadvertently creating visual metaphors that viewers find some meaning in. The symbolism isn’t intentional on the AI’s part but emerges from the complex patterns in its training data. The resulting visuals can be open to interpretation much like dream symbols, allowing us to project meaning onto the AI-generated content like our dreams.
Ever seen a melting wax hummingbird? Runway
Interpretation and Analysis
Probably as long as we’ve been aware of our dreams, we’ve been trying to find meaning in the symbols in our dreams. Freud, for example, viewed dreams as expressions of repressed desires and unresolved conflicts, while Jungian psychology sees them as communication from the collective unconscious, rich in universal symbols and archetypes. Modern cognitive approaches to dream interpretation see dreams as reflections of our waking life’s concerns and emotional states, serving to consolidate memories and problem solve. These approaches typically analyze dream content, emotions, and symbols, often in the context of the dreamer’s personal experiences and cultural background. There is potential value in understanding these dream elements. Similarly, AI analysts “red team” models by examining the outputs of AI models to understand the underlying structures and biases present in the training data. This analysis can reveal unintended biases such as gender or racial stereotypes or emergent properties, complex behaviors or patterns that arise from the interaction of simpler elements withing the AI system in ways not explicitly programmed. Researchers will look for recurring themes or styles in AI-generated videos, unexpected juxtapositions of elements, or even the AI’s handling of abstract concepts, like sadness. This analysis not only helps improve AI systems by identifying and addressing biases but can also provide insight into the nature of machine learning and artificial creativity. As with dream interpretation, the analysis of AI-generated video often requires a multidisciplinary approach, drawing from diverse fields such as computer science, psychology, neuroscience, art theory, and cultural studies to fully understand the implications of what AI can produce.
Freud would have a field day - Luma
Potential for Insight and Creativity
We often look to our dreams for creative inspiration – I’ve had a couple of dreams that I’m turning into screenplays because there were some great ideas in there. Many artists, writers, and innovators throughout history have credited their dreams as the source of groundbreaking ideas or artistic visions. The surreal and often emotionally charged nature of dreams can provide new perspectives on familiar concepts, or as a part of entirely new ideas that seem beyond the reach of conscious thought. This creative potential of dreams exists because our mind can generate innovative concepts when freed from the constraints of waking logic and reality. AI-generated videos can also inspire creativity by offering fresh perspectives. As AI models combine elements from the training dataset in unexpected ways, they can produce visuals that human creators might never have conceived on their own. Artists and designers are beginning to incorporate AI-generated content into their creative processes, using it as a starting point or inspiration. The often surreal or abstract nature of AI-generated visuals can challenge us to see the world differently, potentially leading to innovative approaches. Surprise and breaking boundaries are potential seeds for creativity.
Creative possibilities are endless – Luma
Conclusion
There’s more to the connection between dreams and AI video than one would think. From their visual nature and non-linear narratives to their potential for symbolic representation and potential for inspiring creativity, these two seemingly disparate phenomena are quite similar. Both draw from huge sources of information – the subconscious mind for dreams and training datasets for AI – to produce content that often defies logic and expectations. The lack of complete control in both realms leads to unpredictable and sometimes profound outcomes. There might be fundamental patterns in how information is processed and recombined to create new ideas and experiences, whether in the human brain or in artificial neural networks. We may gain deeper insights into the nature of creativity, consciousness, and the potential future interplay between human and artificial intelligence. We need to start questioning our definitions of imagination, creativity, and maybe even consciousness itself.
F’inn - Runway
