Assessing the SoM + NSM model

Established notions of exploitation and exploration gain no extra strength from the state of mind (SoM) framework. It seems to have no predictive power. TA Figure 2 offers a continuum that lumps diverse dichotomies including Open-Ended↔Goal-Directed, Interest↔Deprivation, and Originality↔Usefulness as if they were respective components of two contrasting states of mind, rather than assessing whether each pair or component involves different systems whose contributions must be distinguished. The continuum does not help us organize analysis of the contributions of the default mode network (DMN), salience network (SN), and executive control network (ECN) posited for novelty-seeking model (NSM). A useful exercise would be to carefully define the terms of Figure 2 and analyze their relation to interactions within DMN-SN-ECN.

TA§6 note that “The … brain networks … included in the NSM are involved in many aspects of cognition [and] not solely underlying … creativity.” My concern is that, by ignoring data from other aspects of cognition, TA leaves DMN, SN, and ECN and HC as unanalyzed “lumps,” depriving us of the opportunity to assess how the way their constituent circuits serve diverse roles in quotidian behavior may illuminate their roles in creativity.

TA finds it “reasonable to assume that intrinsic and extrinsic motivation could synergistically benefit creativity … Intrinsic motivation may be essential for the novelty component, just like exploratory SoM, while extrinsic motivation can help to ensure perseverance and elaboration, similar to exploitatory SoM.” But it remains unclear whether the SoM framework adds anything here.

NSM posits four phases: Affinity generates novel combinations in a semantic network, in Activation salient combinations potentially cross a relevance threshold, Evaluation further assesses combinations, and in Commitment the hippocampus is engaged as a novel interlink is created in the semantic network. However, adding links to a semantic network seems a poor framework for a theory of creativity. Unfortunately, TA is almost devoid of examples. One of the (two?) exceptions is “the Remote Associates Test, which requires participants to find a common element among three seemingly unrelated concepts (e.g., mines, lick, sprinkle) and to generate a fourth item related to each item in the trio (e.g., salt).” Here, then, the nodes of the semantic network seem to be words with their meaning-items, with a link between two nodes if they share a meaning-item. “Creativity” in this case involves finding words associated with more than one of the three targets (Affinity) until one is found that is associated with all three (Evaluation). The result is then that word, but no new link is added to the semantic network. Rather, a working memory gathers and evaluates existing links, and there is no Consolidation once the test has been completed.

Linking two ideas may be a crucial part of creativity (recall Koestler's bisociation – Koestler, Reference Koestler1964; Miller, Reference Miller1964) but, in general, this only one step in creating new and more complex structures. Indeed, tests of in-the-lab-creativity through assessing and making drawings involve novel “constructions” rather than new links in a semantic network, and there are many relevant studies of hippocampus (Moscovitch, Cabeza, Winocur, & Nadel, Reference Moscovitch, Cabeza, Winocur and Nadel2016; Schacter & Addis, Reference Schacter, Addis and Abraham2020; Summerfield, Hassabis, & Maguire, Reference Summerfield, Hassabis and Maguire2010).

Designing buildings

Recent work (Arbib, Reference Arbib2020, Reference Arbib2021) analyzes how architects design buildings, constructing new patterns in memory (in diverse brains and the external representations) that cumulatively yield a plan for a new building. The architect does not manipulate an extant semantic network but instead creates rich “mental constructions” that can guide the physical construction of buildings – devising spaces and shaping and relating forms to serve stipulated functions, be aesthetically pleasing, conform with the site, and be built with available funds.

In the VISIONS model of interpreting visual scenes (Hanson & Riseman, Reference Hanson, Riseman, Hanson and Riseman1978), perceptual schemas compete and cooperate to interpret regions of the scene and relations between them. Perception “clamps” retinal input to drive schema activation and interpretation. “Bottom-up” processing integrates input data as one basis for activating schema instances, but once some schema instances are activated, perhaps by outside considerations (Yarbus, Reference Yarbus1967), “top-down” processes come into play. Perception is here a form of mental construction.

The experience of a building is multisensory, and design may involve constructing physical models to offer a genuine feel of spatial relations, but much of design involves drawing. Visual imagination “inverts” vision, “clamping” interpretation and some constraints on schemas to drive top-down activation of schema instances and feature maps. But the resultant drawings can then stimulate and anchor further creativity.

Turning to movement and navigation in space: In a World Graph, a node corresponds to a significant place, and each edge represents a direct path from one such place to another (Arbib & Bonaiuto, Reference Arbib and Bonaiuto2012; Lieblich & Arbib, Reference Lieblich and Arbib1982, in BBS). A WG may link to a locometric map which charts patterns of locomotion in physical space.

Designers exploit their own diverse forms of long-term memory – episodic, procedural, and semantic – to design spaces that will structure the experience of the building's users as they develop behaviors that, at least in part, emerge as variations on the scripts (patterns of behavior that can be adapted to varied circumstances) the architect imagined. Our modeling (Arbib, Reference Arbib2020, Reference Arbib2021) extends navigation to controlling transitions between affordances (opportunities for action) in supporting the constraints of scripts. The architect must transform each script into design ideas for a WG linking places that need to be included in a building to satisfy the script. But a design that specifies separate places for each script may be both uneconomical and inconvenient. An assessment is thus required of which places to merge, unifying WGs in the process. Only the registration of the WG with a locometric map makes the factoring of effort into executability possible.

There are vast realms of empirical data left to be explored and discovered and much further modeling to be done – modeling that goes far beyond semantic networks to enrich future contributions of the neuroscience of creativity to our understanding of the experience and design of architecture.