Intelligence is equal parts achievement and understanding. The ability to learn, then apply information and skills drives the human innovation process—an exponential search for knowledge and the application of creative ideas and inventions to advance society. The creations of intelligent innovation are mainly aimed at industry, science, and technology.
At the crossroads of those three main targets of innovation—industry, science, and technology—sits a modern human invention: artificial intelligence.
Artificial intelligence (AI) is a commonly used umbrella term for intelligent machines that behave and think in useful, often creative ways. AI is used in a number of sectors ranging from agriculture to healthcare.
You can now add psychology to that list. Psychologists have begun to integrate algorithms and machine learning, among other functions, into their research. Of particular interest are dreams, and how AI can help you better understand your psyche.
The field of artificial intelligence, and the coining of the term itself, took place at Dartmouth University, in Hanover, New Hampshire, in the summer of 1956. American computer scientist John McCarthy—the father of AI—noted that the purpose of these innovations was to engineer intelligent machines and computer programs to solve problems.
The field has shown that machines are taking strides towards true intelligence. For starters, they can learn from past experiences, like humans. Beyond that, AI programs can “dream.”
A pair of Google software engineers, and an intern, published the results from their DeepDream visualization tool. While the code was designed to study the learning processes behind the multiple layers of a neural network, it also produced dream-like visual art. The program essentially does something that humans do, namely produce meaningful images from ambiguous stimuli—what is better known as pareidolia.
When humans look at clouds, particularly visually creative people, they see all sorts of amazing creatures and patterns. Leonardo da Vinci called this effect pareidolia. You might see a dog, a fish, or a combination of the two. Yes, the mythical dog-fish. The DeepDream program can do exactly that, create a composite of a dog and a fish from the sole image of a wispy cloud. Beyond clouds, the program produced magnificent images of actual buildings and art that are reminiscent of fantasy. The researchers called these “neural net dreams” (Mordvintsev et. al., 2015).
AIs are not entirely meant to mimic humans exactly, rather they can do human-like things—take dreaming and learning for example—and also support humans in the pursuit of intelligence—like analyzing mind-boggling amounts of data. These connections led researchers to implement AI in the field of dream research.
|
As early as 1893, years before the publication of Sigmund Freud's Interpretation of Dreams, psychologist Mary Calkins published The Statistics of Dreams in The American Journal of Psychology. Calkins recorded and analyzed the time, location, continuity between waking and sleeping, and vividness of dreams. Her results did not favor the continuity hypothesis, a popular modern theory that claims dreaming often reflect the waking life. |
Dreams are commonly collected—especially in large databases—based on individually written entries of when a dream occurred, and what happened. Three examples of dream databases used in research are:
One of the major problems in dream research is the textual analysis of dream reports. They are laborious and time consuming. Initially, researchers turned to word search methods to identify similarities between the frequency of words in dream reports and personal activities, relationships, and troubles (Bulkeley, 2014). Soon thereafter, researchers built AI tools for computing large amounts of dream data.
In 2017, linguists used an AI program to analyze the coherence, defining features, and themes of dream reports from the DreamBank database against personal narratives—to see if the artificial intelligence could distinguish the data and accurately parse dream content. Not only did the program accurately detect dreams from the narratives, it found some interesting differences. For example, people tended to use more language about time and conversational expressions in personal narratives, while language about settings and general uncertainty were more common for dreams (Hendrickx et. al., 2017).
A few years later, a team of researchers studied dream content analysis using artificial intelligence to analyze data from the Dreamboard application. In line with the often collaborative efforts required by human-computer research, a team featuring a psychologist, a neurologist, and computer scientists were assembled to develop a dream content analysis system (McNamara et. al., 2019). The researchers fed their AI a little more than 35,000 dream reports.
The results were ideal—they confirmed some well known measures, and added a few unexpected contributions. For example, when compared to established methods, the AI found that physical aggression and fantasy-based characters were more common in dreams, indicating their social nature. The researchers, while supporting the Social Simulation Theory of dreaming, argued that the AI’s ability to detect context analysis led to this finding. Context is key, and with humans, the context is usually social.
In 2020, artificial intelligence dream researchers took a major step in psychometrics. By operationalizing the numerical Hall and Van de Castle coding system for dream analysis, the researchers were able to apply the scale to a large sample of dream reports—again from the DramBank database—in a short period of time. After analyzing 24,000 dream reports, Aiello et. al., concluded in favor of Calvin Hall’s continuity hypothesis of dreams, and also found an interesting result: blind people tended to dream more of imaginary characters, when compared to the rest of the general public (2020).
More recently, some of the same researchers used artificial intelligence to create a visualizer that helps users better understand the continuity between their waking lives and dreams (Bogucka et., al., 2021). Their Dreamcatcher program is positioned to improve the qualitative analysis of dreams through visualization of data and the ability to share the results with others.
Artificial intelligence is the driving force behind a new idea on why we dream. Tufts neuroscientist Erik Hoel recently proposed that the brain evolved dreams—in particular, the stranger things in them—to generalize everyday experiences. This is called the overfitted brain hypothesis (Hoel, 2021). Hoel was inspired by the parallels between the human brain and deep neural networks. When training an AI, Hoel wrote that researchers often introduce random bits of data to help the program gain a more complex picture of the world, learning how to respond to new types of input. Using this metaphor, dreams then serve to keep the human brain from developing a simplistic worldview.
The researcher pointed to several lines of evidence, like the connections between dreams and creativity. Another example cited is that the easiest way to dream about something from the real world is to do a repetitive, but novel task. For example, you could play a game, like Tetris (Hoel, 2021). Going along with the hypothesis, this repetition, or overfitting, would trigger a dream about the task to generalize the results. These generalizations can lead to more insights, instead of following known, repetitive patterns.
Artificial intelligence, when paired with a team of trained researchers, can make dream analysis more accessible to everyone. Instead of keeping tedious dream journals that require hours of expert analysis, you can analyze dreams in a few minutes using a technology called Dream Blots, an Inkblot Analytics psychological innovation.
All you need to do is fill out a short survey, and take an inkblot test that doubles as a dream journal. It’s simple, write down what you see in your dream journal and use the inkblot to tag images. Our AI can then quickly score and analyze your dreams, connecting the content to your personality profile. By helping you interpret your dreams at night, we hope you can better understand what you experience every day. Want to give it a try?References:
Aiello, L., Fogli, A.,& Quercia, D. (2020). Our dreams, our selves: automatic analysis of dream reports. Royal Society open science, 7(8), 192080.
Bogucka, E., Aseniero, B., Aiello, L., and Quercia, D. (2021). "The Dreamcatcher: Interactive Storytelling of Dreams," in IEEE Computer Graphics and Applications, vol. 41, no. 3, pp. 105-112
Bulkeley, K. (2014). Digital dream analysis: A revised method. Consciousness and Cognition: An International Journal, 29. pp. 159–170.
Calkins, M. (1893). The Statistics of dreams. The American Journal of Psychology. 5(3). pp. 311-343.
Hendrickx, I., Onrust, L., Kunneman, F., Hürriyetoǧlu, A., Hurriyetouglu, A., van den Bosch, A., & Stoop, W. (2017). Unraveling reported dreams with text analytics. Digit. Humanit.
Hoel, E. (2021). The overfitted brain: dreams evolved to assist generalization. Patterns.
McCarthy, J., Minsky, M., Rochester, N., and Shannon, C. (1956). Artificial Intelligence (AI) Coined at Dartmouth. ‘A proposal for the Dartmouth summer research project on artificial intelligence.
McNamara, P., Duffy-Deno, K., Marsh, T., & Marsh, T., Jr. (2019). Dream content analysis using Artificial intelligence. International Journal of Dream Research, 12(1). pp. 42–52.
Mordvintsev, A., Olah,C., and Tyka, M. (2015). Inceptionism: Going deeper into neural networks. Google Research.