Frequently Asked Questions

1. What are the main findings of this study?
2. Why is targeted dream incubation (TDI) an important part of this study?
3. How is targeted dream incubation (TDI) used in this study?
4. What inspired this study? What was known about sleep onset before this study?
5. How much is creativity improved by a TDI nap?
6. Does this study tell us that dreams are creative?
7. What does it mean to measure creativity?
8. How was creativity measured in the study?
9. What is semantic distance?
10. How does semantic distance relate to creativity?
11. How is semantic distance different after napping vs. staying awake? What does this tell us about sleep onset?
12. How was the study carried out?
13. What are the main ethical concerns associated with this work?
14. Where else has this study been shared?
15. Contact Information

1. What are the main findings of this study?

   In our study, we used a technique called targeted dream incubation (TDI) to overcome a longstanding challenge in dream research: guiding dream content. Using TDI, we were able to run a controlled study to compare how dreaming during sleep onset affects post-sleep creative performance. Among the many exciting findings in our paper, we’d like to draw your attention to the two key results:

   1. First, we found that a period of sleep onset boosts creativity compared to staying awake for the same period. This finding aligns with a 2021 study from Lacaux et al., which also found a creative boost associated with sleep onset on a number-related insight task. Our study extends this finding to three different classical creativity tasks, with real-world relevance and both objective and subjective performance measures. Furthermore, by analyzing participant responses, we identified the potential mechanism through which sleep onset may increase creativity: by enabling a more cognitively flexible state with greater exploration of novel associations between concepts.

   2. Second, we show that dreaming of a topic during sleep onset is directly related to increased post-sleep creativity on that topic (i.e., if you dream about a “tree,” you will perform better on post-sleep creativity tasks related to a tree than a subject who does not. If you dream about a tree multiple times, you will do better than a subject who dreamt about a tree one time). This is the first study to show that incubating dreams can boost post-sleep creativity on tasks related to the incubated theme. This is a novel scientific breakthrough, both for understanding the role of sleep and dreams and for the general public, who may be able to apply these findings to boost their own creative ideation.

2. Why is targeted dream incubation (TDI) an important part of this study?

   Our study focuses on the link between dreams and post-sleep creativity. While the role of dreams (including this link between dreams and creativity) has been a topic of intense speculation for millenia, the scientific study of dreaming has a much shorter history. Moreover, the field has been historically limited by a key methodological challenge: the inability to control dream content. To scientifically show a causal role of dream content on post-sleep cognition, a scientist would need some means to guide dream content in order to randomly assign study participants into different dream conditions, akin to how a plant scientist might test the efficacy of different fertilizers by applying each one to a different randomly-selected set of plants. However, since such a method was not available, past studies which have tried to link dream content to post-sleep cognitive properties have been fundamentally limited to associating participants’ spontaneously-occurring dreams with post-sleep cognition. While these studies have generated insight into the potential role of dreaming, there was a great need for improved tools in order to break past this methodological limitation.

   In our study, we used a technique called targeted dream incubation (TDI) – developed by our group in collaboration with others – to guide dreams and overcome this long standing challenge in dream research. Using TDI, we were able to run a controlled study to compare how dreaming during sleep onset affects post-sleep creative performance. 

   By “controlled study,” we refer to our ability to randomly assign participants to different experimental conditions, one of which was defined by dreaming about a tree (which we used TDI to accomplish), in order to relate these conditions to post-sleep cognitive performance. This is in contrast to a paradigm where dream content was not a controlled variable, where scientists would be limited to just correlating spontaneous dream content with post-sleep performance. TDI represents an important methodological advance, and we hope our study can inspire more studies in dream science where dream content can be a controlled variable.

3. How is targeted dream incubation (TDI) used in this study?

   We used targeted dream incubation (TDI) in our study to guide the dreams of participants. Our TDI protocol was applied to two participant groups: one with the “tree” cue (“Remember to think of a tree.”) and one with a neutral cue (“Remember to observe your thoughts.”). By collecting dream reports from participants after each interruption of their nap, we found that over 90% of our participants in the “tree” TDI group had at least one dream of a tree.

4. What inspired this study? What was known about sleep onset before this study?

   Our study was inspired by both anecdotal reports and scientific publications that pointed to the creative potential of the sleep onset period (also called Stage 1 sleep, NREM1, or N1), especially the transition from wakefulness into sleep known as hypnagogia. In general, the sleep onset period has been characterized as containing spontaneous, vivid dreams, suggesting that it could be an ideal state for creative idea generation.

   In terms of anecdotal reports, the sleep onset period has been the site of scientific and artistic discoveries made while dreaming by the likes of Thomas Edison and Salvador Dalí. The basic technique used by Edison and Dalí for capturing hypnagogic insights consisted of dozing off with a heavy object in hand. Right as they fell asleep, muscle tone would lessen and they would drop the object, causing a loud noise that would wake the sleeper, who then recorded potential insights made in their hypnagogic dreaming. These anecdotal reports fascinated us and inspired us to investigate if we could scientifically show that hypnagogic dreams could contain useful creative insights.

   Laboratory studies have also suggested that sleep could be an optimal brain environment for creative ideation. For example, a 2002 study from Walker et al. showed improvement in anagram problem solving following REM sleep. The sleep onset period specifically has recently been highlighted as a creative sweet spot; an exciting 2021 study from Lacaux et al. found a creative boost following the sleep onset period on a mathematical insight task (called the number reduction task). However, we saw a gap in the research literature in terms of studies about the role of dreaming at sleep onset, and the relation of such dreaming to creativity. As such, we were inspired to use targeted dream incubation to study the role of dream content on the sleep onset creative boost.

5. How much is creativity improved by a TDI nap?

   Our study participants were divided into four groups with different protocols: one group napped and received TDI (with a prompt to think of a tree) and another napped without incubation. A third group stayed awake and received incubation (i.e. prompts to think of a tree but delivered periodically without intervening naps) and the last group stayed awake without incubation. After a 45 minute period of their assigned protocol, all participants completed the same three creativity tasks, all relating to the word “tree.” The participants’ responses to these tasks were scored for creativity by human raters using a rubric. Using these scores, we can measure differences in creativity between these four groups to see how much creativity is affected by napping with TDI.

   Participants who napped with TDI performed 43% more creatively (in terms of the Creativity Index) than participants who napped without incubation. Additionally, the group which napped with TDI performed 78% more creatively than those who stayed awake without incubation. 

   We feel it is important to note that these improvements are limited to the topic and tasks investigated by our study; further work is needed to generalize these results to other creative domains, tasks, and TDI topics.

6. Does this study tell us that dreams are creative?

   This study does not tell us that all dreams are creative. Our study specifically shows that an incubated dream about a particular topic may improve post-sleep creative performance on tasks related to that topic! We were excited to see our participants integrate the ideas from their dream reports directly into their responses to our creativity tasks (and this happened more frequently in the incubated dream group than the non-incubated group). We feel this evidence, in combination with the statistical results showing the average creativity was highest after TDI, is our strongest clue that dreams can be a source of creative inspiration. In a way, our study is a formalization of the method used by Edison and other creative greats, who notably harnessed their sleep onset dreams to make creative insights.

7. What does it mean to measure creativity?

   Measuring creativity is a complex and challenging task, since creativity is multifaceted, context-dependent, and subject to personal preferences. Creativity researchers have developed methods to measure creativity in the lab. These methods each have strengths and limitations and measure different facets of creativity. Common methods include:

   1. Divergent thinking tasks: These tasks measure a person’s ability to generate multiple creative solutions to a problem, with the quantity and originality of the ideas generated used as indicators of creative potential.

   2. Creative product analysis: Participants’ creative products (such as artwork, music compositions, or creative writing) are evaluated for quality and originality.

   3. Self-report measures: This approach assesses a person’s perception of their own creativity. People are asked to rate themselves on metrics such as their originality, fluency, or flexibility of thinking.

   4. Creative achievement measures: This approach involves the evaluation of a person’s real-world creative achievements (such as those from scientists, writers, or artists). 

   By measuring real-world creative achievement, scientists have been able to show that many laboratory-based methods (such as divergent thinking tasks) are correlated with real-world creative achievement. This gives us confidence that although these methods to evaluate creativity have limitations, they still have relevance to real-world creativity – which is the kind we care about most.

8. How was creativity measured in the study?

   In our study, we assessed creativity using three different tasks: the Creative Storytelling Task (CST), the Alternative Uses Task (AUT), and the Verb Generation Task (VGT). We used three different tasks to ensure that we could assess multiple facets of creative ability, such as creative writing ability (via the CST), divergent thinking (via the AUT), and more (via the VGT).

   For the CST, participants were asked to write a creative story using the word “tree.” The CST mainly assesses the ability to make combinations of concepts that form a meaningful and creative product. One participant wrote a story about a CIA agent arriving on an island called “Sycamore Island,” accessible only via a raft of “PVC and pistachio shells glued together.”

   For the AUT, participants were asked to list all the creative, alternative uses you can think of for a tree. The CST measures divergent thinking abilities, assessing the ability to broaden one’s representational search space to produce a wide range of responses to a query. Some example responses included, “back scratcher” and “toothpick for giants.”

   For the VGT, participants were instructed to respond creatively with the first verb that came to mind for each word in a list of 31 nouns (most of which were related to the word “tree”). The VGT assesses several cognitive processes, such as broad associative thinking, ability to shift between different conceptual categories, and ability to generate novel responses. Some example responses to the noun “tree” included “climbing,” “blooming,” and “providing.”

   All of the participants’ creativity task responses were scored by three human raters, who were unaware of the experimental paradigms assigned to the participants. To score each task, the human raters were all given a training on assessment of creativity.

   Finally, we used scores from all three creativity tasks to generate a composite “Creativity Index'' for each participant. This composite score encapsulates multiple facets of each participant’s creative abilities. We focused our analysis on the Creativity Index, but all of the scores from each of the individual creativity tasks were also analyzed in our study. 

9. What is semantic distance?

   “Semantic distance” refers to the degree of difference (or similarity) between two concepts based on their meaning (i.e. their “semantic content”). In other words, semantic distance is a measure of how closely related two words are. For example, the words “fork” and “spoon” are both utensils with similar roles in human society, so these two words would be closer in semantic space (i.e. have a smaller semantic distance between them) than, for example, the words “fork” and “window.”

   Semantic distance is measured computationally, often using methods based on “word embeddings.” Word embeddings are mathematical representations of words, wherein each word is represented as a vector in high-dimensional space. Word embeddings are generated by various algorithms. While there are many different algorithms that have been used for creating word embeddings, in general, the distance between a pair of word embedding vectors represents the semantic distance between the pair.

   In our study, we chose to use a pre-trained GloVe (Global Vectors for Word Representation) embedding (developed in 2014 by researchers at Stanford University). GloVe’s embeddings are based on the co-occurrence statistics; that is, word pairs that occur more frequently together (such as fork and spoon) will have smaller semantic distance than word pairs that occur less frequently (such as fork and window). We measured semantic distance as the cosine distance between words we were comparing.

10. How does semantic distance relate to creativity?

    One of the most well-studied and longstanding theories of creativity is the associative theory, which proposes that creativity often involves generating novel ideas by making connections between seemingly unrelated or distant concepts. Under this theory, the more distant two concepts are, the more challenging it may be to meaningfully connect them, thus requiring more creative effort to bridge the gap.

    Past research has shown that highly rated responses to creativity tasks (such as the Alternative Uses Task, or AUT, in which people list creative alternative uses for common items) tend to also have higher semantic distance. However, the strength of this correlation between creativity and semantic distance, varies greatly depending on the task at hand, the word embedding model used, and the way the semantic distance measurements are aggregated (i.e. for a story, one could choose to take a maximum taken from all pairs of words, or a mean). Some of the strongest correlations between human-rated creativity and semantic distance are seen in divergent thinking tasks, such as the AUT (for example, check out this paper from Beaty and Johnson, 2021). Still, for other types of creativity (such as evaluating the creativity in stories), semantic distance is a poor proxy for human ratings. The main benefit of semantic distance lies in the fact that it is a computationally generated measurement, unlike human ratings, which allows for standardization across large datasets. All in all, depending on the task at hand, semantic distance is a useful, but incomplete, proxy for creativity.

11. How is semantic distance different after napping vs. staying awake? What does this tell us about sleep onset?

    According to the associative theory of creativity, generating creative solutions often involve generating novel ideas by making connections between seemingly unrelated or distant concepts. We wondered how N1 sleep might be causing a creative boost. Based on the associative theory of creativity, we hypothesized that the N1 stage sleep could enable a cognitive state which promotes exploration of more distantly associated concepts. To test this hypothesis, we measured semantic distance in participants’ creativity task responses.

    In our study, we found that participants who napped had higher semantic distance in their responses to the Alternative Uses Task (AUT) and Verb Generation Task (VGT) than participants who stayed awake. We believe this result sheds light on a potential mechanism underlying the N1 creative sweet spot: N1 may be enabling a cognitive stage with broader associative divergence, facilitating exploration of connections between distant concepts.

    How does this finding relate to our findings about dreaming? We propose the following hypothesis: N1 enables a state optimized for making distant connections, and TDI allows this state to be applied to making connections to a specific cued topic (such as a tree, as was used in our study), thus allowing the dreamer to have creative insights about the topic. Further research is needed to better understand this potential model. For more on sleep-dependent mechanisms of creativity, see the NEXTUP theory proposed by our co-author Bob Stickgold. 

12. How was the study carried out?

    The study was co-led by Adam Haar Horowitz (a PhD graduate of the Fluid Interfaces group and current postdoctoral researcher in the lab) and Kathleen Nguyen Esfahany (an MIT undergraduate researcher in the Fluid Interfaces group). Before the study, a team of collaborators including Adam and co-author Tomás Vega Gálvez developed the Dormio device and TDI protocol. Adam applied these tools to lead the design of our experiment and collected the data for our study. In 2019, Kathleen joined the lab and began working closely with Adam to process the data. Kathleen led the design and implementation of our statistical analyses and data visualizations. The study was co-supervised by Professor Pattie Maes (Fluid Interfaces, MIT Media Lab) and Professor Robert Stickgold (Harvard Medical School).

13. What are the main ethical concerns associated with this work?

    We are very concerned about ensuring our main methodology – TDI – is used in ethically sound ways. Please see our thoughts here.

14. Where else has this study been shared?

    You can read more about our study in Science, Scientific American, MIT News, and many other outlets. Our study has also been shared in a talk delivered by Kathleen at the 39th Annual Conference for the International Association for the Study of Dreams and was featured in an interview on WEOL radio. 

15. Contact Information

    To contact the lead authors of the study, please email Adam Haar Horowitz (adamjhh@mit.edu) and Kathleen Esfahany (kaes@mit.edu). 

    To contact the co-supervising authors, please email Professor Robert Stickgold (rstickgold@hms.harvard.edu) and Professor Pattie Maes (pattie@media.mit.edu)