According to the target article, humans (1) detect novelties that (2) instill curiosity. Then (3) the curious mind enriches memory by encoding the novelty. Finally, (4) memories thus constructed are recombined in creativity leading to new actions or thoughts. We provide a critique against all four causal links, and demonstrate that curiosity and creativity are typically unrelated.
Curiosity does not demand expected novelty
Novelty involves epistemic expectations about what is more or less likely to occur ontologically (uncertainty about existence) or within the bounds of a specific contextual state (e.g., uncertainty about action outcomes). Studies involving trivia questions demonstrate that ontological uncertainty can instill curiosity, but only when some preconditions are met such as in the presence of prior domain interest, or that guesses are extrinsically rewarded (Dubey & Griffiths, Reference Dubey and Griffiths2020; Fastrich, Kerr, Castel, & Murayama, Reference Fastrich, Kerr, Castel and Murayama2018). However, nonnovel environments with regularly occurring but uncertain states and outcomes are also powerful stimulators of curiosity, demonstrated by studies involving simple decisions in lotteries, and sensorimotor control tasks (Holm, Wadenholt, & Schrater, Reference Holm, Wadenholt and Schrater2019; Kobayashi, Ravaioli, Baranès, Woodford, & Gottlieb, Reference Kobayashi, Ravaioli, Baranès, Woodford and Gottlieb2019; Van Lieshout, Vandenbroucke, Mu, & Cools, Reference Van Lieshout, Vandenbroucke, Mu and Cools2018). In fact, simply needing to refresh our memories can be a potent curiosity inducer as in the urge to look up an actor's name you forgot when recounting a movie. In general, our environments and memories are replete with novelties, uncertainties, and missing features and we are indifferent to most of them. What suffices to instill curiosity remains unknown, but novelty is neither sufficient nor necessary.
Curiosity-driven learning does not need to expand memory
Curiosity often targets very restricted knowledge variables such as the performance in computer games (Holm et al., Reference Holm, Wadenholt and Schrater2019; Ten, Kaushik, Oudeyer, & Gottlieb, Reference Ten, Kaushik, Oudeyer and Gottlieb2021; Van Lieshout et al., Reference Van Lieshout, Vandenbroucke, Mu and Cools2018), or information seeking in social media (Zahoor, Reference Zahoor2022). These activities may become obsessive yet satisfy curiosity, and may even lead to addiction-like behavior (Hsu, Wen, & Wu, Reference Hsu, Wen and Wu2009; Weinstein, Reference Weinstein2010). This type of curiosity-driven learning might even gate out the acquisition of new knowledge by expending cognitive resources and time improving efficiency in a narrow domain, which arguably constitutes an exploitative rather than exploratory pursuit.
Novel memories do not need to promote creativity
Curiosity learning and novel memories often result from observation learning. Imagine that monkey A watches monkey B perform a novel task, and later monkey A recalls the event and solves the task via imitation. While for monkey A, this constitutes a novel action stemming from a curiosity encoded memory, its creativity is sorely limited. More generally, enriching memory with more solutions does not imply coming up with a better solution. A football team with 11 great forwards is probably not a great team. Reusing stored solutions is exactly exploitation, even if the information is subjectively novel and encoded under curiosity. Instead, creativity requires a recombination of memories – and that requires memories that meaningfully overlap and are decomposable. Curiosity-induced imitation learning is ubiquitous but excludes the meaning (intent/reasons/semantics) for the actions, which limits the ability to decompose (identify steps) and to find overlaps. In fact, generalization in learning has been shown to be sharply limited in imitation learning without structured training that provides overlapping alternative solutions (Braun, Mehring, & Wolpert, Reference Braun, Mehring and Wolpert2010; Wolpert & Flanagan, Reference Wolpert and Flanagan2016).
Creativity does not imply curiosity
Although curiosity need not lead to creativity as the authors claim, they could still be linked if, when creativity occurs, curiosity is implicated. Unfortunately, this backward implication is false because (1) the presence of creativity does not imply enriched (novel) memories. Any old memories suffice as long as their recombination is novel and valid. (2) Enriched memories do not imply curiosity because curiosity is not necessary for learning. (3) The presence of curiosity neither does require novelty as outlined above nor is (4) novelty detection a necessary precondition for instilling curiosity.
Conditions for curiosity to promote creativity
Although curiosity and creativity are independent in many conditions, the analysis above suggests they may be related when: (1) Curiosity results in improved and enriched memory representations that (2) lead to valid recombinations. But even with these conditions met, the reason for a common underlying principle operating within the intersection of the conditions seems unclear, other than novelty seeking is an inadequate basis. To illustrate what creativity entails, consider Stoffel the honey badger, a talented escape artist whose incessant need for freedom exceeds the repeated efforts of his keepers (BBC, 2014). But being free is a poorly defined environmental goal state because it does not indicate a unique point that is required for a typical control system. We need to invoke higher order internal goal states to account for Stoffel's (and other creative creatures') behavior in which “freedom” can be construed as well-defined (internal) goal states in an abstract epistemic space. Curiosity and creativity might thus share principles for reaching intrinsic epistemic goal states. Our proposal is aligned with work on artificial curiosity and creativity, which have had to make careful distinctions between intrinsic and extrinsic motivations and carefully refine concepts like novelty. In particular, this work shows that standard conceptualizations of novelty and surprise are deeply problematic (e.g., Oudeyer, Reference Oudeyer2018; Schmiedhuber, Reference Schmiedhuber2010).
Curiosity and creativity as expressions of a common epistemic drive
Curiosity might share computational principles with creativity insofar as both invoke active sampling for reaching epistemic goal states. The states of interest in an epistemic drive are how well ideas and facts are interrelated, and how much freedom remains to move between ideas, solution, and understandings following acquisition of new information. The aim of the epistemic drive may thus be construed as to maximize information connectedness across perspectives. For curiosity this entails seeking information to arrive at a knowledge state congruent with experience, for creativity it entails testing information combinations until a valid goal state is satisfied. Importantly, new information may work as effectively against expanded knowledge by entrenching constraints as offering new perspectives that release the constraints and offer new knowledge. This narrows what kind of information is desired for the epistemic drive and suggests a “house-keeping” principle for information seeking. Moreover, an epistemic drive may be adaptive and promote survival in the long term because it promotes behavioral flexibility in the eye of environmental change. Finally, it seems we need to broadly invoke an epistemic drive to account for the behavior of animals or how else can we account for Stoffel's ingenious escapes?
You have
Access
According to the target article, humans (1) detect novelties that (2) instill curiosity. Then (3) the curious mind enriches memory by encoding the novelty. Finally, (4) memories thus constructed are recombined in creativity leading to new actions or thoughts. We provide a critique against all four causal links, and demonstrate that curiosity and creativity are typically unrelated.
Curiosity does not demand expected novelty
Novelty involves epistemic expectations about what is more or less likely to occur ontologically (uncertainty about existence) or within the bounds of a specific contextual state (e.g., uncertainty about action outcomes). Studies involving trivia questions demonstrate that ontological uncertainty can instill curiosity, but only when some preconditions are met such as in the presence of prior domain interest, or that guesses are extrinsically rewarded (Dubey & Griffiths, Reference Dubey and Griffiths2020; Fastrich, Kerr, Castel, & Murayama, Reference Fastrich, Kerr, Castel and Murayama2018). However, nonnovel environments with regularly occurring but uncertain states and outcomes are also powerful stimulators of curiosity, demonstrated by studies involving simple decisions in lotteries, and sensorimotor control tasks (Holm, Wadenholt, & Schrater, Reference Holm, Wadenholt and Schrater2019; Kobayashi, Ravaioli, Baranès, Woodford, & Gottlieb, Reference Kobayashi, Ravaioli, Baranès, Woodford and Gottlieb2019; Van Lieshout, Vandenbroucke, Mu, & Cools, Reference Van Lieshout, Vandenbroucke, Mu and Cools2018). In fact, simply needing to refresh our memories can be a potent curiosity inducer as in the urge to look up an actor's name you forgot when recounting a movie. In general, our environments and memories are replete with novelties, uncertainties, and missing features and we are indifferent to most of them. What suffices to instill curiosity remains unknown, but novelty is neither sufficient nor necessary.
Curiosity-driven learning does not need to expand memory
Curiosity often targets very restricted knowledge variables such as the performance in computer games (Holm et al., Reference Holm, Wadenholt and Schrater2019; Ten, Kaushik, Oudeyer, & Gottlieb, Reference Ten, Kaushik, Oudeyer and Gottlieb2021; Van Lieshout et al., Reference Van Lieshout, Vandenbroucke, Mu and Cools2018), or information seeking in social media (Zahoor, Reference Zahoor2022). These activities may become obsessive yet satisfy curiosity, and may even lead to addiction-like behavior (Hsu, Wen, & Wu, Reference Hsu, Wen and Wu2009; Weinstein, Reference Weinstein2010). This type of curiosity-driven learning might even gate out the acquisition of new knowledge by expending cognitive resources and time improving efficiency in a narrow domain, which arguably constitutes an exploitative rather than exploratory pursuit.
Novel memories do not need to promote creativity
Curiosity learning and novel memories often result from observation learning. Imagine that monkey A watches monkey B perform a novel task, and later monkey A recalls the event and solves the task via imitation. While for monkey A, this constitutes a novel action stemming from a curiosity encoded memory, its creativity is sorely limited. More generally, enriching memory with more solutions does not imply coming up with a better solution. A football team with 11 great forwards is probably not a great team. Reusing stored solutions is exactly exploitation, even if the information is subjectively novel and encoded under curiosity. Instead, creativity requires a recombination of memories – and that requires memories that meaningfully overlap and are decomposable. Curiosity-induced imitation learning is ubiquitous but excludes the meaning (intent/reasons/semantics) for the actions, which limits the ability to decompose (identify steps) and to find overlaps. In fact, generalization in learning has been shown to be sharply limited in imitation learning without structured training that provides overlapping alternative solutions (Braun, Mehring, & Wolpert, Reference Braun, Mehring and Wolpert2010; Wolpert & Flanagan, Reference Wolpert and Flanagan2016).
Creativity does not imply curiosity
Although curiosity need not lead to creativity as the authors claim, they could still be linked if, when creativity occurs, curiosity is implicated. Unfortunately, this backward implication is false because (1) the presence of creativity does not imply enriched (novel) memories. Any old memories suffice as long as their recombination is novel and valid. (2) Enriched memories do not imply curiosity because curiosity is not necessary for learning. (3) The presence of curiosity neither does require novelty as outlined above nor is (4) novelty detection a necessary precondition for instilling curiosity.
Conditions for curiosity to promote creativity
Although curiosity and creativity are independent in many conditions, the analysis above suggests they may be related when: (1) Curiosity results in improved and enriched memory representations that (2) lead to valid recombinations. But even with these conditions met, the reason for a common underlying principle operating within the intersection of the conditions seems unclear, other than novelty seeking is an inadequate basis. To illustrate what creativity entails, consider Stoffel the honey badger, a talented escape artist whose incessant need for freedom exceeds the repeated efforts of his keepers (BBC, 2014). But being free is a poorly defined environmental goal state because it does not indicate a unique point that is required for a typical control system. We need to invoke higher order internal goal states to account for Stoffel's (and other creative creatures') behavior in which “freedom” can be construed as well-defined (internal) goal states in an abstract epistemic space. Curiosity and creativity might thus share principles for reaching intrinsic epistemic goal states. Our proposal is aligned with work on artificial curiosity and creativity, which have had to make careful distinctions between intrinsic and extrinsic motivations and carefully refine concepts like novelty. In particular, this work shows that standard conceptualizations of novelty and surprise are deeply problematic (e.g., Oudeyer, Reference Oudeyer2018; Schmiedhuber, Reference Schmiedhuber2010).
Curiosity and creativity as expressions of a common epistemic drive
Curiosity might share computational principles with creativity insofar as both invoke active sampling for reaching epistemic goal states. The states of interest in an epistemic drive are how well ideas and facts are interrelated, and how much freedom remains to move between ideas, solution, and understandings following acquisition of new information. The aim of the epistemic drive may thus be construed as to maximize information connectedness across perspectives. For curiosity this entails seeking information to arrive at a knowledge state congruent with experience, for creativity it entails testing information combinations until a valid goal state is satisfied. Importantly, new information may work as effectively against expanded knowledge by entrenching constraints as offering new perspectives that release the constraints and offer new knowledge. This narrows what kind of information is desired for the epistemic drive and suggests a “house-keeping” principle for information seeking. Moreover, an epistemic drive may be adaptive and promote survival in the long term because it promotes behavioral flexibility in the eye of environmental change. Finally, it seems we need to broadly invoke an epistemic drive to account for the behavior of animals or how else can we account for Stoffel's ingenious escapes?
Financial support
The authors were funded by a grant from the Swedish Research Council (2019-02997).
Competing interest
None.