The trouble with ‘HIM’: new challenges and old misconceptions in human information modelling

Abstract

Building performance simulation is not just about buildings’ geometry, construction, and systems. It has become increasingly clear that simulation models need also to more systematically address occupants’ presence and behaviour in buildings. To put it somewhat ostentatiously, it is not just about ‘BIM’ (Building Information Modelling), but also ‘HIM’ (Human Information Modelling). To this end, a balance must be found between representational formalisms and representational content. The former concerns finding the right algorithmic expressions, the latter requires empirical knowledge. Exploring algorithmic formalisms can be exciting, but if lopsided, may fail to recognize the complexity of human perception and behaviour. Therefore, it may be instructive to discuss similar instances of past lop-sided formalism-centric approaches. The present position paper underlines the importance of domain knowledge on occupants’ perceptual and behavioural processes and its inclusion in the computational representation of occupant agents in building information modelling and building performance assessment.

KEYWORDS:

• Human
• information
• modelling
• mathematical formalisms
• empirical information

1. Introduction

Information and communication technology (ICT) in general and building information modelling (BIM) platforms, in particular, have the potential to support processes related to the design, construction, operation, retrofit, and decommissioning of buildings (Eastman 1999; Hensen and Lamberts 2019). Researchers and developers in the computational building modelling domain can contribute to the effective exploitation of this potential. To this end, it is important to carefully and continuously reflect on the quality and direction of related research efforts. Such reflections are especially critical whenever the excitement generated by new methods and technologies could obfuscate the original motivational field of the research content-related background. This tendency, at times characterized as (computational) hammers seeking for (application) nails (Mahdavi 2015), appears to be a recurrent phenomenon and not restricted to ICT research. The mentioned excitement, understandably triggered by the introduction of new tools and formalisms, can lose sight of critical questions, such as the fitness of specific computational formalisms to the nature of the problems they are meant to address. It is thus not surprising that such critical questioning (Mahdavi 2019) is not necessarily popular, and not necessarily heeded to. But it is needed, if we want to avoid wasteful conceptual cul-de-sacs and developmental aberrations.

In this general context, the present treatise focuses on a specific aspect of certain recent research trends in building performance simulation (BPS) that are motivated by the increased recognition of buildings’ occupants and users. No one would dispute the centrality of occupants as the recipients of services provided by buildings. However, computational applications meant to support the building delivery process have traditionally adopted fairly rudimentary representations of building occupants. Hence, there is arguably a need for more detailed and realistic models of people, their presence and movements in buildings, as well as their interactions with buildings’ devices and systems. But, to address this need, the focus has been predominantly on tools, methods, and formalisms. For instance, agent-based modelling (ABM) ranks high amongst approaches thought to have a promising potential to accommodate the complexity and dynamics of occupants’ control-oriented behaviour (Berger and Mahdavi 2020). Natural as these tendencies may be, they are prone to a lopsided stance in the dialectic of content and formalism. The contention is that the challenging complexities associated with the processes of human perception and behaviour can be at times swept under the rug of a predominantly tool-centric approach.

Before addressing this dialectic in more detail in the next sections, it is important to note that the systematic incorporation of high-resolution representations of occupants via ABM or through other formalisms, is relevant to both effectiveness and efficiency criteria in buildings’ performance. Effectiveness criterion pertains to the provision of conditions that are conducive to people’s health, comfort, satisfaction, and productivity. But such conditions must fulfill the efficiency criterion, mainly in terms of energy use and environmental impact. According to this view, buildings are expected to provide a high degree of habitability (associated with the effectiveness criterion) in an energy and resource-saving manner (relevant to the efficiency criterion). BPS typically involves both of these criteria. It can support the design of buildings that are effective, i.e. offer appropriate indoor-environmental conditions. It can also support meeting the performance expectations in an efficient manner concerning energy and resource deployment as well as environmental impact.

The evaluation of buildings’ effectiveness specifically necessitates a deep understanding of human requirements. The representation of the human dimension in BIM and BPS, which could be referred to as Human-Information-Modelling (HIM), must rely on research findings regarding occupants’ perception of and behaviour in built environments. Arguably, the value and importance of computational methods and associated formalisms lies in their support of an efficient building design, construction, and operation process, once the effectiveness criteria are settled. However, it seems at times as if it is believed that computational methods and formalisms on their own (or based on limited, simplistic, and superficial content-related model input assumptions) can yield optimal solutions. Whereas efficient formalisms are necessary and essential, formalism-centric thinking may divert from the appreciation of the indispensable significance of dependable domain knowledge.

2. About nail-seeking hammers

Prior to taking a closer look at advanced computational formalisms (specifically, agent-based modelling) for representing occupants in BPS, it may be instructive to reflect on the relationship between computational formalisms on the one side and the phenomena or processes they aim to frame, capture, and describe on the other side. At a general level, a formalism (involving, for instance, a tool, an instrument, or a technique) can be thought of as a kind of abstract structure projected onto observed target of observation and representation. In this context, occasionally, the following problem arises: Fixation on the formalism can not only lead to inaccurate or mistaken projections, but also end up supplanting the essence of the circumstances or phenomena that were meant to be understood and represented in the first place. In this context, one can speak of formalisms searching for applications, or, metaphorically speaking, hammers seeking for nails. This phenomenon may motivationally stem from a well-known general trait of human cognition system, namely the tendency to seek for meaningful patterns in the observational field (Mattson 2014). Arguably, such a trait is, from the evolutionary standpoint, of considerable survival value: The consequences of falsely recognizing a non-existent threat may be inconvenient, but failing to recognize the pattern associated with a real threat may be fatal.

Note that, abstract formalisms, such as theories in mathematics (e.g. topology) can be the subject of queries and probing independent of physical realities they may be deployed to describe. Such formalisms may exist long before they are found to be relevant for developing physically relevant models. For instance, seventeenth century progress in description of the planetary movements could refer back to prior – much older (i.e. ancient Greece) and independently established – investigations of conic sections (Howard 1990). Likewise, the formulation of Einstein’s general relativity benefited from Grossmann’s input regarding the previously established understanding in non-Euclidean geometry, differential geometry, and tensor calculus (Einstein and Grossmann 1913, 1914; Sauer 2015). On the other hand, explorations in physics might trigger new mathematical formalisms. For instance, Newton’s ‘fluxional’ is suggested to have been motivated by his prior investigations into the behaviour of physical systems such as the acceleration of falling objects. Another remarkable instance of physics-to-mathematics cross-fertilization pertains to the delta function. In his book, ‘The Principles of Quantum Mechanics’, Paul Dirac introduced this function as a continuous analogue of the Kronecker delta (the discrete version of the delta function) (Dirac 1930). Ever since, this mathematical structure for the representation of point objects (e.g. mass, charge) is widely applied in quantum physics. Later, the mathematician Laurent-Moïse Schwartz developed the theory of distributions (Schwartz 1966), which provided a solid interpretational basis for objects such as the Dirac delta function.

As such, the dialectic of mathematical formalism and descriptions of physical phenomena has been highly fruitful. But overly formalism-centric approaches can end up diverting or even derailing research progress. The following brief references to a few instances of such formalism-centric attitudes are meant to illustrate the kinds of misconceptions that may arise when an inquiry starts with a cherished formalism (or a pattern, tool, technique, algorithm, etc.) searching for factual (or observational, empirical, etc.) subject to be projected upon:

• Numerous studies (some scholarly, some amateurish) point to real and putative geometric patterns and proportions behind objects in nature, arts, and architecture. Well-known instances of such patterns involve the golden section, Fibonacci series, and harmonic intervals (Naredi-Rainer 1982). Interestingly, even while discussing the one and the same subject (for instance, the geometric features of a cathedral), there are disputes as to which of such patterns are the ones truly underlying the compositions (Haase 1988). This reinforces the impression, that once a certain interest is aroused concerning the significance of a specific pattern, scheme, or proportional system, there is a tendency to look for – and claim to have demonstrated – its prevalence. It is of course quite possible, particularly in case of artifacts of art and architecture, that designers and creators have indeed consciously drawn on such patterns. But in many other cases, there is a notable degree of arbitrariness in the alleged discoveries.

• Fascination with formalisms and the urge to widely project them onto all kinds of objects, entities, and phenomena appears at times to follow certain cycles and fashion-like tendencies. One such formalism – highly popular in the last decades of the previous century emerged from an extension of the classical geometry, namely fractal geometry. Here again, the claim to interpretative potency goes beyond the initially postulated ‘Geometry of Nature’ (Mandelbrot 1982) to cover ‘ … carpets, bricks, and much else besides’ (Barnsley 1993), not to forget about applications in architecture (Iasef Md Rian et al. 2007).

• Another relatively recent instance of the proverbial nail-seeking hammers may be recognized in the popularity of the so-called shape grammars (Stiny and Gips 1972) among some researchers in the design computing field in the eighties and nineties of the last century. Aside from its purported applicability to compositional design, shape grammars were also suggested to provide means of analyzing various design languages (Chiou and Krishnamurti 1995). It has been suggested that a sense of preeminence in matters of design interpretation (almost a universality claim) accompanied shape grammar discourse. However, convincing instances of practical applications in the building design and construction domain did not emerge.

• In multiple instances in the past, an entire scientific discipline or theory has been adopted in terms of formal explanatory vehicles in architectural domain. Notable instances of such systems may be found, for instance, in semiotics, cybernetics, and information theory (Mahdavi 2020). Semiotics is a theory of signs, how they function, and how they acquire meaning (Buchler 1955; Eco 1978). Information theory concerns the measurement and transmission of information (Shannon 1971). Cybernetics is not only about regulatory processes in biological and technological systems, but has been applied also in cognitive, and social systems (Ashby 1956; Wiener 1948). Notwithstanding the actual value and utility of these disciplines, once again aspects of the aforementioned nail-seeking hammer symptom are recognizable, including an initially enthusiastic adoption in architectural discourse, a tendency to claim wide applicability despite insufficient evidence, and a slow fading process, typically followed up by another fashionable paradigm.

• The potential and value of parametric simulation – particularly in design development and optimization scenarios is well established. However, in the context of research in relevant areas of building science, parametric simulation is of very limited use, if it is carried out solely based on an empirically untested computational model (Samuelson et al. 2016). Carrying simulation-aided virtual experiments using an unavoidably reduced model of reality may be useful in certain situations, but cannot substitute independent empirical scrutiny of the respective findings. The tendency to throw parametric simulation runs at fundamental research questions seems like yet another instance of tools and techniques aiming to demonstrate their applicability independent of the availability of empirically-based proof for the validity of the underlying models.

• The above observations are arguably relevant to a number of contemporary formalism-centric approaches that are of interest to this contribution’s, namely the inclusion of the occupant models in BIM and BPS. A few relevant critical reflections concerning some of these approaches may be formulated as follows:

• The majority of the developmental efforts toward predictive models of occupants’ behaviour have been based on rather limited sets of empirical data. This applies also to the efforts to capture the dynamics and probabilistic appearance of occupancy-related processes using stochastic methods (involving, for example, generalized linear mixed models) (Mahdavi and Tahmasebi 2019). Common instances of such processes include patterns of occupants’ arrival in an office building or the frequency of window opening actions in a residential building. There is as such nothing wrong with the deployment of probabilistic formalisms in the context of occupants’ control-oriented behaviour in buildings. They may indeed represent highly fitting instruments in a number of application scenarios. A problem arises though, when the deployed formalism is claimed to have explanatory or predictive power beyond the scope of data underlying its development. Such claims have been falsified by a number of relevant studies using independent empirical information (Mahdavi 2015; Mahdavi and Tahmasebi 2016, 2017, 2019). To repeat, there is no reason to disregard data-driven stochastic formalisms and their potential to generate realistic time series of occupants’ actions. But it is not the specific genus of a formalism as such that establishes its credibility. Rather, as always in science, conformance to observational data remains the ultimate validity criterion.

• A second cluster of ongoing research and development efforts (Linder et al. 2017) is motivated by the perceived success of the so-called ‘big data’ approaches in other domains (Mayer-Schönberger and Cukier 2014). It is assumed that the application of data mining techniques to data streams from multiple sources (e.g. monitoring systems, building automation, occupants’ phones and other personal devices) can support operational optimization of buildings’ energy performance and maintaining optimal indoor-environmental conditions. AI-based tools and techniques such as data mining and pattern recognition have been indeed demonstrated to facilitate computational solutions in a number of domains (such as those requiring powerful search algorithms) without relying on white-box (e.g. explicit, causal or first-principles based) models. Such solutions may be also applicable to and effective in the behavioural modelling field, but this has not been conclusively established. The degree of the effectiveness of inherently black-box type methods – particularly in design support applications – toward a deeper understanding of causal and motivational factors behind occupants’ behaviour remains an open question.

• A further class of computational tools and techniques are informed by the – principally reasonable – premise that no model can be ever considered to be fully correct. Rather, there will always be some level of uncertainty associated with both model input assumptions and the resulting computational output. Sensitivity analysis, uncertainty analysis, and Bayesian reasoning (Gelman 2013; Tian et al. 2018) are geared toward operationalization of uncertainties. These could be the result of insufficient or low-quality input data or a consequence of knowledge gaps in the models’ underlying representations of real-world phenomena and processes. Such techniques and methods can be undoubtedly useful – for instance when evaluating the reliability of computationally derived building performance indicator values. However, the utility of these types of application scenarios is a direct function of the empirical cogency of assumed uncertainty distributions of pertinent model variables. In the absence of such empirical grounding, the mere generation of potentially arbitrary uncertainty ranges would be of little use. Likewise, the utility of Bayesian reasoning remains limited, as long as assumptions concerning the pertinent priors are entirely arbitrary. In such instances, we may end up with models that generate behavioural pseudo-data (such as patterns of building occupants’ interactions with buildings’ environmental control systems) with very little or no empirical backing.

To venture into a bit of meta-critical reflection, even search and finding instances of confirmation bias itself may be an instance of confirmation bias at work. Nonetheless, the preceding critical remarks point to a specific kind of confirmation bias: Having developed or adopted certain tools, methods, and formalisms, we may become invested in demonstrating their broad utility and effectiveness. Whereas developing and tweaking formalisms is something we can pursue and control, validity and applicability can only be established based on – frequently cumbersome – empirical data collection and analysis. It thus should not come as surprise, when in building design and operation domain, formalisms have been frequently granted credibility prematurely – that is without a thorough evidence-based examination (Mahdavi 2015; Mahdavi and Tahmasebi 2017, 2019).

The reliability and utility of formalisms meant to capture people’s behavioural patterns depends on the depth and credibility of the underlying empirical understanding of people’s perceptual and behavioural processes. This applies also when we look at the potential of ABM toward the implementation of knowledge-based human-centric building design and operation approaches. The following brief treatment of this technique can shed further light on the dialectic of formalisms and domain knowledge.

3. The case of ABM

ABM is being increasingly applied in multiple diverse domains (Bonabeau 2002; Macal and North 2008; Wilensky and Rand 2015), including also those relevant to the built environment (Alfakara and Croxford 2014; Andrews et al. 2011; Azar and Menassa 2010). Specifically, ABM appears to provide an adequate vehicle for the detailed and dynamic representation of building occupants in BIM and BPS. As such, occupants’ behavioural tendencies – for instance, their decision-making processes with regard to building controls – can be coded in terms of a set of rules. Hence, as a modelling technique, ABM has the potential to cope with the formally complex and dynamic aspects of occupant-related processes in the built environment. Moreover, given reliable underlying empirical information, ABM can facilitate the representation of socially and culturally relevant inter-occupant behavioural influences and learning phenomena. However, we can reasonably suggest that the previous discussion regarding formalism-centric approaches is also relevant to ABM and its application in BPS. As such, aside from coding integrity and usability issues, ABM’s usefulness in building design and operation directly depends on the robustness of the underlying assumptions regarding agents’ behavioural tendencies and repertoire.

A recent comprehensive review of related scientific publications (Berger and Mahdavi 2020) explored the sources of knowledge tapped into in order to generate agents’ behavioural patterns in building-related ABM applications. The review focused specifically on publications concerned with the simulation of buildings’ energy and indoor-environmental performance. The result of this review indicates that only a minority of the research efforts (about 30%) drew on concrete underlying behavioural theories. In fact, about 40% of the reviewed publications rely exclusively on basic general material included in codes and standards, meaning they neither referred to a theoretical foundation nor included any original empirically-based information on occupants’ behaviour. Some form of empirical data (e.g. interviews, observations) could be located in roughly 40% of the efforts. However, such data cannot be assumed to be generally representative or applicable, as none of the contributions included empirical data obtained from multiple buildings or locations. A simultaneous inclusion of theoretical underpinnings, standard-based information, and at least a modicum of empirical data could be found only in 13% of the reported building-related ABM development efforts (Berger and Mahdavi 2020).

We can derive, from this review exercise concerning ABM applications in BPS, a couple of important conclusions. It is evident that, from the purely technical point of view of computational implementation, the ABM-based emulation of occupants’ presence and behaviour in buildings has become increasingly feasible. However, the actual utility of ABM-based building design and operation instruments remains limited. This is due, on the one hand, to insufficiently developed explanatory theories of human behaviour as individuals and as members of socially relevant groups and, on the other hand, to the paucity of sufficiently detailed and representative empirical data.

4. Understanding human perception and behaviour

The main point of our arguments so far has been to underline the indispensable role of domain knowledge (empirical observation and explanatory theories) for the development and validation of formalisms in general and ABM in particular. But at the same time, we should not create the impression that a set of clearly outlined and empirically tested behavioural theories exist out there and that could have been easily tapped into by the BIM and BPS communities. Ideally, explanatory theories of building occupants’ perception of and behaviour could provide the foundation for an evidence-based approach (together with related codes, standards, guidelines, and computational tools) to building design and operation. However, despite much past research (including recent studies in psychology and neuroscience), one needs to go back to intellectual currents in the twentieth century, if not the late nineteenth century, to locate high-level explanatory theories of human perception and behaviour (as proposed, for instance, in psychophysics, human ecology, ecological psychology, and cybernetics) with general and transparent characteristics (Mahdavi 2020; Mahdavi, Teufl, and Berger 2021). A detailed discussion of these theories and their more contemporary instances cannot be offered here. However, a few of their key aspects are briefly outlined in the following, based on the findings of a previous review (see Mahdavi 2020 for a more detailed treatment).

One recurrent trend pertains to models of perception. Such models aim at operationalization of conjectured correlations between the strength of physical stimuli (e.g. temperature, luminance, sound level) and the intensity of people’s subjective perception of warmth, brightness, loudness, etc. These kinds of correlations provide the basis of psychophysical scales adopted in standards, and guidelines which map measurable environmental stimuli onto corresponding physiological and psychological responses of occupants. However, the utility of psychophysical scales – as well as other similar or derivative approaches – is limited due to a number of factors, including: (i) the non-linear nature of perceptual responses to multi-dimensional (i.e. thermal, visual, olfactory, auditory) exposure situations; (ii) the difference between immediate perceptual correlates of physical stimuli and the higher-level assignment of values to such perceptions; (iii) people’s attitude toward and prior experience with the sources of stress as well as their true and perceived level of control over such sources; and (iv) people’ diversity, as evidenced in inter-individual differences in age, gender, cultural background, physical and mental health, attitudes, expectations, and preferences.

A further trend discernible in some behavioural models views survival as the most fundamental priority (value) of an organism and the requisite of the organism to maintain its life-critical internal state in an adequate range. In case of organisms with high-level environmental mapping capability such as human beings, both non-conscious homeostatic processes and conscious behaviour (triggered by feelings of pain and pleasure) bring about state transitions toward the homeostatic range beneficial to survival and well-being. This very insight has informed the development of comfort models. For instance, conventional thermal comfort models typically highlight the human body’s need to maintain heat balance on the long-term basis (Fanger 1972). Protracted departure from physiological equilibrium states associated with heat balance is assumed to trigger thermal discomfort and encourage compensatory (homeostatic) processes.

Nevertheless, the extensive research and standardization work in the thermal comfort domain has not resulted in definitive instances of related models. Alterations to the conventional thermal comfort models suggested by the adaptive thermal comfort theory (Humphreys, Nicol, and Roaf 2015) may represent progress, but they too do not conclusively resolve thermal comfort modelling challenges. A key problem is the relatively large number of suspected independent variables comfort models must consider. Obtaining dependable values for the multitude of variables involved would be a persistent challenge, even if we could construct a multi-layered (physical, physiological, psychological, social) causal model of comfort or behaviour.

These challenges are further complicated due to the fact that both indoor-environmental quality judgments and adaptive behaviour are influenced by occupants’ views and attitudes. This circumstance can be exemplified by occupants’ tolerance (the ‘so-called forgiveness factor’) toward occasional departure from comfort conditions due to inhabitants’ social and environmental attitudes (Leaman and Bordass 2007). Furthermore, comfort and behavioural models face difficulties in accounting for the sizable diversity amongst the population of buildings’ occupants. Besides, in building design scenarios, the detailed composition of future occupants is rarely known. This makes it difficult to predict future patterns of occupants’ perception and behaviour. Hence, uncertainties resulting from inter-individual diversity of building occupants can outpace the differences between predictions of rival models.

Yet another challenge results from the fact that occupants perceive (and act in) a multi-dimensional exposure field, including – amongst other things – thermal, visual, auditory, and olfactory factors. Perception of comfort and disposition to control-oriented behaviour with regard to a specific environmental factor (such as noise) is not independent of exposure parameters pertaining to other factors (such as glare). But currently available comfort models do not capture such interdependency. This circumstance is briefly addressed in the next section.

5. In search of theories of multi-domain exposure

In principle, the development of robust theories in all fields of inquiry requires a sufficient density of available factual data. Here is the field of behavioural research relevant to building occupants no exception. Despite a long research tradition, health and comfort standards still address thermal, visual, olfactory, and auditory factors in isolation. This may be consistent with the traditional divide et impera approach in many scientific inquires. But single-domain models fall short of accounting for real indoor-environmental exposure situations involving multiple kinds of stimuli (Mahdavi et al. 2020; Berger and Mahdavi 2021). The results of past multi-physical perceptual research remain rather inconclusive, and in part even contradictory. The related studies have been mostly short-term with frequently limited number of (non-representative) participants. Moreover, they have been conducted in rather artificial settings and have been susceptible to the Hawthorne effect. Frequently, researchers do not apply standard research designs, data collection strategies, metrics, and statistical analysis techniques. This makes it thus very challenging – if not infeasible – to conduct meta-analyses of the subject. To increase the credibility of occupant-centric computational formalisms for building design and operation support, and to truly bring HIM to BPS, we need a robust, sufficiently operationalized, and empirically tested the theory of occupants’ perceptual and behavioural processes under realistic situations in built environments. Formalisms can help us look for the answers, but formalisms are not the answer.

6. Concluding remark

The analysis offered in this contribution was motivated by a basic observation: If applied ICT-based methods, tools, and platforms are to support effective occupant-centric building design and operation, they must incorporate substantiated knowledge on human perception and behaviour. Thereby, representations of building occupants cannot be approached solely in terms of a series of alternative formalisms (e.g. stochastic methods, ABM). Rather, they must be based on solid knowledge derived from pertinent domains in human sciences. The tendency to approach computational representations of buildings’ occupants from a formalism-centric angle may be understandable, but is problematic. As far as the academic research community is concerned, this tendency might have been motivated in part by the pressures of funding acquisition. Formalisms and abstract models can be played around with extensively (resulting, perhaps, in academically exploitable output) without the need to engage deeply in the unforgivingly laborious empirical research into the nature of human cognition, perception, and behaviour. But, without a deep empirically-based understanding of human perception and behaviour, computational occupant representations will remain superficial, and so too their incorporation in BPS.

Acknowledgements

The author gratefully acknowledges TU Wien Bibliothek for financial support through its Open Access Funding Programme.

Disclosure statement

No potential conflict of interest was reported by the author(s).