The term “complex problem” and the concept behind it became an essential part of scientific discourse during the 1980s. By that time, most researchers strongly agreed that human problem-solving capabilities could not be fully scientifically understood based on studies of simple problems only. Laboratory data obtained in such studies significantly lacked ecological validity because of several reasons. First, common tasks, used in laboratory experiments on thinking and decision-making, had little to do with problems that people faced and had to solve in real life. Second, it became more and more evident that problem-solving in real life was not only connected with the actualization of different mental processes (rational thinking) but also regulated by personality and psychological factors, such as motivation, self-esteem, personal involvement in the problem, etc., [1
From the very beginning, several research traditions emerged in that field, that had little in common [1
]. Among currently existing approaches toward complex problem-solving, the more prominent are Naturalistic Decision Making (NDM), Dynamic Decision Making (DDM), Implicit Learning in System Control, Expert Adaptive Decision Making, Complex Problem Solving (European tradition), and so forth [2
]. Research approaches varied from field studies involving real-life situations in NDM to the development of simulation scenarios with more or less complex structure. It is possible to say that all the traditions used complex, dynamic situations to research problem-solving on a new level, unreachable by earlier approaches [2
]. However (or, specifically, because of that pluralism of approaches), no universally accepted definition for the complex problem and complexity, in general, was established, thus making it still difficult to compare and integrate different branches of complex problem-solving studies.
Earlier attempts in that direction were mainly revolving around the identification of objective characteristics of the “complex problem” task. Campbell [4
] conducted an extensive literature analysis and marked three main approaches to understanding the “complexity” of a task. According to him, complexity can be understood first in the context of the person’s subjective experience; second, as a characteristic of the human-task interaction, and third, as the objective characteristic of the task itself. Campbell’s position was consistent with the latter approach, and he suggested that complexity could and should be determined objectively. He identified four primary sources of complexity in the task: (1) The presence of many ways to achieve the goal; (2) the presence of many goals; (3) the presence of a contradiction between different goals; and (4) probabilistic relationship between the methods and the goals.
On the contrary, recent works usually attempt to describe complexity from the subjective experience approach. For example, Osman considered problem-solving as “the process by which people transform the unknown to the known” [5
] (p. 52). In this case, complexity represented the level of subjective uncertainty. There can be different sources of uncertainty, such as characteristics of the problem itself (random fluctuations, probabilistic relationships between cause and effect, nonlinearity), or psychological factors (inaccurate representation of the problem’s structure; biased assessment of probabilistic relationships, etc.,). Regardless of those sources, the subjective feeling of uncertainty can be considered the criterion of equivalence for various complex problems.
] pointed out that complex problems had one central requirement for the person, trying to solve those problems: the demand for the personal ability to tolerate uncertainty. In their 2017 article, Dörner and Funke [7
] emphasized the role of mentalizing in eliminating the uncertainty. In this case, mentalizing was understood in the broader sense, including both discursive (reasoning) and non-discursive (creative, intuitive) thinking processes. According to those authors, complex problem-solving includes cognitive, emotional, and motivational aspects [7
] (p. 6).
Among subjective factors influencing complex problem-solving effectiveness, cognitive abilities are the most researched ones [8
]. Early studies found no correlation between psychometric intelligence and successful complex problem-solving [10
]. The low ecological validity of standard intelligence tests was one of the probable explanations [11
]. The requirements of complex problems and intelligence tests widely varied, indeed. Usually, complex problem-solving involves goal setting and planning processes actualization, active search, and information selection, hypothesizing, structuring and restructuring knowledge acquired in the process, and feedback-based action-management. Traditional intelligence tests cannot grasp most of these aspects. To explain human ability to use their cognitive abilities and skills, Dörner [6
] proposed the term “operative intelligence.” Thus, human interaction with complex dynamical systems should not be reduced to some generalized scheme, which could be solved by some universal algorithms or optimal strategies, such as Newell and Simon’s general problem solver [12
]. Problem-solving strategies and methods are only adequate for specific conditions of the task or situation. Thus, planning, goal setting, reflection, and self-reflection play an essential role in complex problems and should be taken into consideration.
Yet, the “classical” complex problems (such as LOHHAUSEN, MORO, etc., [6
]) were often criticized for low reliability and weak foundation of the complex problem-solving success measurements. Various complex problems indeed approached performance assessment differently. Those assessments could include scores on a scale or a set of scales, genuine system properties understanding, finding the strategy that would be closest to the optimal one, etc. The diversity and heterogeneity of those tasks and measures made it hard to compare performance in different tasks in a meaningful way. These considerations served as one of the origins of the psychometric approach in complex problem-solving research [7
]. This approach aimed to develop reliable diagnostic methods for complex problem-solving assessment [14
]. The paradigm of minimal complex systems was proposed as an alternative to microworlds with many variables and connections between them [14
]. Currently, within this approach, MicroDYN and MicroFIN scenarios are the most used as general models of a complex problem.
A recent meta-analysis [16
], which included 60 studies from 1982 to 2014, showed that the correlations between successful complex problem-solving and intelligence varied from 0.339 to 0.585, depending on the type of the complex problem. The weakest correlation (0.339) was obtained for “classical” complex problem tasks (LOHHAUSEN, MORO, TAILORSHOP, etc.,) [6
], while the strongest (0.585)–for minimal complex systems tasks (MicroDYN, MicroFIN, etc.,) [16
As was already mentioned, Dörner [6
] considered psychometric intelligence to be a poor predictor of success in complex problem-solving (supported by the correlations above). Comparing the “bad” and “good” “burgomasters” in the LOHHAUSEN simulation, he pointed out the following differences: “good” participants acted more comprehensively, considered more aspects of the system, tested hypotheses through questioning to clarify the causal relationships of the variables, generally organized, and structured their behavior better. On the opposite, “bad burgomasters” tended to switch from one problem to another when faced with obstacles, were more often distracted by current stimuli, and sometimes demonstrated “encapsulated” behavior, spending a lot of effort on solving insignificant problems. Thereby, the difference between the participants in this task was more about personality and less about pure intelligence or knowledge.
Among non-cognitive variables, the most studied ones are motivation and the process of goal setting [17
]. Recent studies in MicroDYN scenarios showed the contribution of the “Big Five” personality traits in complex problem-solving [20
], however, with only relatively small effect sizes. To our knowledge, personality traits related to attitudes toward uncertainty were not previously researched in the context of complex problem-solving.
Dörner suggested that the ability to cope with uncertainty was a significant predictor of complex problem-solving success [6
]. Thus, in the current study, we found it necessary to focus on personality traits known to contribute to decision-making under uncertainty. Iowa Gambling Task-based studies [21
] showed tolerance for uncertainty to regulate risk propensity during the initial stages of decision-making, influencing exploratory learning. Intolerance for uncertainty, on the other hand, was shown to regulate risk propensity after a failed trial. According to the authors, intolerance for uncertainty, thus, potentially constrained learning under uncertainty through risk aversion [21
Among other personality traits, mediating decision-making under uncertainty, psychologists mention impulsiveness [22
], intuition [23
], risk-readiness, and rationality [24
]. Janis and Mann [25
] argued that strategies to cope with decisional conflict (arising in uncertain situations) also influenced decision-making significantly. Our previous study showed that people with different forecasting strategies had a significant difference in their intolerance for uncertainty levels [26
]. Other studies showed that tolerance/intolerance for uncertainty traits ratio and levels of personal risk-readiness formed stable latent profiles, explaining personal attitude toward uncertain situations [24
It is worth noticing that our interest in the current research revolved around the connection of those personality traits with human activity while interacting with a complex dynamic system. Thus, we wanted to know whether the personality traits mentioned above can predict the parameters of such activity and were not deliberately focusing on success or failure in complex problem-solving.
For minimal complex systems such as MicroDYN and MicroFIN, two basic strategies of interaction with complex systems were described [17
], with different research behavior patterns aimed to test hypotheses. The first basic strategy requires systematically changing only one exogenous variable at a time to observe cause-and-effect relationships for that variable. This strategy is called VOTAT (Vary One Thing at a Time). The opposite strategy is the unsystematic change of all exogenous variables (CA, Change All). The combination of those strategies in exploratory behavior usually referred to as Heterogenous Testing (HT). Individuals using the VOTAT strategy generally perform better in minimal complex systems [17
]. In recent works PULSE strategy also was described for minimal complex systems with eigendynamic [27
At the same time, “classical” complex problems have no general theoretical framework that allows the identification and classification of problem-solving strategies. However, it is possible to point out two approaches to the analysis of complex problem-solving strategies. These are: by the analysis of aloud reasoning protocols and questions from the participants to the researcher (as done in Dörner’s studies), or by allowing the participants to gather information about the task and its structure by themselves. Both ways had their advantages and disadvantages.
The first way allowed the researcher to analyze the individual dynamics of the decision-making, such as choosing the goal, characteristics of hypotheses, and methods of testing them. However, the use of quantitative methods was limited because of the difficulties in the individual protocol comparison.
The second way allowed quantitative evaluation of the participant’s activity, fixated by the computerized task’s log files, and further statistical analysis. For example, in the FIRECHIEF scenario, used by Cañas et al. [28
], the log-files analysis led to the identification of three distinguishable strategies with different ratios of assumed basic “quenching” and “control” strategies in them. In the later study using the same simulation [29
], different quantitative parameters were used, such as the distribution of the resources (tractors and helicopters) in different moments. Based on those logs, the authors identified such parameters of problem-solving strategies as strategical planning, tactical planning, strategy flexibility, etc. However, the log-files analysis has weaknesses as well. For instance, it is not able to identify internal psychological mechanisms, mediating seemingly similar decisions.
The studies of decision-making strategies in sequential choice problems [21
] demonstrated tolerance/intolerance for uncertainty and risk-readiness‘ contribution to the participant’s‘ activity regulation. On this basis, we suggested that those (and some related) variables could be involved in the regulation of activity when solving complex problems as well.
Thus, we assumed the following general hypotheses:
In complex problem-solving, tolerance/intolerance for uncertainty and rationality are associated with a preference for a certain level of awareness of the state of the system.
The willingness to vary this level of awareness during the interaction with the system is associated with uncertainty coping strategies use (specifically, buck-passing and procrastination levels) and personal risk-readiness.
The general variability of actions while interacting with a complex system and the tendency to experiment with the variables are associated with intuition, problem-solving self-efficacy, tolerance for uncertainty, and risk-readiness.
The tendency to perform significant abrupt changes in decisions and actions while interacting with the complex system is associated with impulsiveness.
To test those hypotheses and with the chosen approach to use a “classical” complex problem, we needed to create a suitable task allowing us to record the parameters of the subject’s activity of interest in the log files in the first place. Especially for this study, we developed a computerized Java-based task called “The Anthill.” The participants were asked to manage a simplified model of an anthill. The management process included decision-making about resource distribution to ensure the survival of the anthill. There were two resources in the task: the ants and the food. The ants could be distributed freely by the player between three classes with different roles and functionality—the scouts, the workers, and the soldiers. The task is described in detail in Section 2.2.1
of this article.
We chose the quantitative (log-files-based approach) to identify general strategies indicators (such as variability of actions, depth of orientation) and their connections with the participants’ personal specifics. That was the reason to introduce the scouts as a possible ant-class for the player to spend resources on. Scouts helped the participants to gather information about the riskiness and potentials gains of certain decisions and actions. Based on scouts and other variables-related decisions, made by the participants, several parameters were identified during the log-files analysis, further explained in Section 2.2.1
of this article.
In three out of five empirical hypotheses (numbered 1, 2, and 4), we expected to find relationships between the parameters of complex problem-solving strategies and the scales of the LFR-21 questionnaire—rationality and personal risk-readiness. According to the authors of the questionnaire, those scales describe the personal properties of self-regulation in decision-making and actions under uncertainty [30
]. It means that risk-readiness and rationality here are more than personal dispositions, but somewhat generalized characteristics of how people can find their way out of uncertain situations.
By assuming the connection between risk-readiness, rationality, and complex problem-solving strategies, we, thus, hold the idea that those uncertainty-dealing general personal characteristic would manifest in choosing and varying the preferred levels of orientation, as well as when varying the distribution of resources in the “The Anthill” task (as a representation of a complex problem in our study). We were unable to confirm those assumptions.
We did not find evidence that risk-readiness and rationality are directly included in complex problem-solving, at least at the planning level. More likely, those two traits characterize the style of action in uncertain situations. The same applies to the imaginative capability and problem-solving self-efficacy, measured by the Subjective Risk Intelligence Scale [37
]. Additional research is needed to clarify this new assumption. It will also require the inclusion of new parameters of complex problem-solving strategies.
According to our data, the preferred orientational level in the task negatively correlated with tolerance for uncertainty (r = −0.291 at p < 0.05). Those, who are more tolerant toward incompleteness and inconsistency of the information at their disposal, and prefer new and complex tasks, require less information (or less accurate information) while trying to solve a complex problem. On the other hand, we did not find any significant connection between intolerance for uncertainty or rationality with the POL parameter, which means that both the first empirical and general hypotheses can only be accepted partially.
The second empirical hypothesis of our study (matching the second general hypothesis) is also partially accepted. Of the two unproductive copings (buck-passing and procrastination), only the buck-passing showed correlations with the flexibility of the orientation level (higher OLV) with r = −0.293 at p
< 0.05. Buck-passing is described as a refusal to make decisions independently [35
]. As a pattern for coping with decisional conflict in complex problem-solving, it can manifest itself as a refusal to change some parameters of the strategy, the desire to act “in a usual way,” consistently with previous turns. Additionally, we found significant correlations between higher orientational level variability and the negative attitude toward uncertainty (r = −0.274 at p
< 0.05), as well as with venturesomeness (r = 0.296 at p
< 0.05). Like the buck-passing, the negative attitude toward uncertainty is associated with the rejection of attempts to change one’s research behavior in uncertain situations when the effect of this change is unknown and unpredictable [37
]. The assessment of that effect would appear as a separate task, which apparently would probably make the situation even more subjectively uncertain. On the contrary, the venturesomeness manifests itself as a desire to try something new, to look for new experience and thrills; it pushes a person to experiment—in case of “The Anthill” task—with resource distribution throughout the game. Risk-readiness, on the contrary, did not demonstrate a significant correlation with orientational level variability.
The third empirical hypothesis is partially accepted, as well. The class quotas‘ range parameter reflects the boundaries of the search area for a successful strategy. It shows to what extent a participant is ready to experiment with input variables in complex problem-solving. In a situation of an opaque task with a high degree of uncertainty, this activity cannot be regulated only by discursive reasoning. Hypothesizing and verifying those hypotheses in the absence of sufficient information is primarily based on conjecture and intuition. We found a significant correlation between the class quotas‘ range and intuitive ability (r = 0.292 at p < 0.05), but no correlation with intuitive engagement was found. No significant correlations were found between the class quotas‘ range with the imaginative capability and problem-solving self-efficacy as well.
The fourth empirical hypothesis was not confirmed. It was built on the assumption that the mean and median shifts in quotas distribution would reflect the degree of the participant’s caution. In a situation of consistent decision-making, cautious people would perform minimal modifications of their previous decisions. However, there are no relationships with risk-readiness and tolerance for uncertainty, which, among other things, reflect the personal willingness to unexpected consequences of their decisions and actions. Nevertheless, a negative correlation was found between median quotas shift and negative attitude toward uncertainty (r = −0.309 at p < 0.05), which fits well with our interpretation of the general class quotas‘ range parameter. A wide range of changes in input variables reflects the willingness of the participant to experiment, and any pronounced change in input leads to a significant difference in output, which in turn adds new information that needs to be taken into account when making decisions. New information in an opaque situation with many hidden variables and relationships increases subjective uncertainty, rather than clarifying the situation.
For the abrupt changes of strategy indicator, no correlations were found. The fifth empirical hypothesis and the fourth general hypothesis are not accepted.
Regression analysis revealed the predictors of three complex problem-solving strategies‘ parameters. Procrastination demonstrated a statistically significant effect on the orientational level variability and the median quotas shift. Both reflect a person’s willingness to experiment with input variables in the process of solving the problem. Postponing decision-making, as a way of coping with uncertainty in complex problem-solving, can manifest itself as a decrease in the variability of the strategy in general, or between single turns.
Additionally, risk-readiness had a negative effect on the median quotas shift. That result was somewhat unexpected for us, as risk-readiness is generally associated with the ability to make decisions and act under uncertainty. We expected that higher risk-readiness would lead to bigger shifts in resource distribution from turn to turn.
However, another interpretation is possible, since procrastination is the second predictor of the variability of shifts. Risk-readiness, in this case, can enhance the effect of procrastination, since the rejection to choose includes the adoption of the consequences of this rejection as well, which seems to be easier for people with higher risk-readiness. Still, this requires further research on larger and more balanced sample.
For the preferred orientational level, the only significant predictor was tolerance for uncertainty. We explained the correlation of these indicators at the beginning of this section.
For all three regression models, the percentage of explained variance was not remarkably high (no more than 15 percent). Nevertheless, we were able to show some contribution of personality factors to the regulation of complex problem-solving strategies.
Thus, some of our initial assumptions about the relationships between personality traits and parameters of people’s activity when interacting with a complex dynamic system were confirmed. A decrease in tolerance for uncertainty was associated with a desire for more comprehensive information about the system’s state. The use of buck-passing and procrastination as coping strategies reduces the variability of the preferred level of awareness in the progression of the game. Intuitive ability is associated with a greater tendency to experiment with variables.
Limitations. In the current study, the sample was small and unbalanced by gender (female participants prevailed) and age. Those characteristics limit the possibility of generalizing the obtained results.
There is no generally accepted conceptualization of the strategy concerning complex problem-solving, just as there is no generally accepted formalized model of this process. Therefore, it is rather difficult to evaluate the extent to which the parameters of strategies that we have identified describe human activity in complex problem-solving. The analysis of those strategies did not consider the influence of feedback on the participants’ actions, which limits our abilities to understand their activity fully. Refinement of the script to include feedback display in log-file at each turn would allow us to monitor how the participants change their actions accordingly. In the current state of the program, this information is exceedingly difficult to work with, as it requires step-by-step reproduction of the participants’ actions following the log-file.
The obtained correlations need to be checked on larger samples. A larger sample would also open the possibility for the use of multivariate statistical methods. That would allow us to identify stable patterns of the ratio of fixed indicators of strategies and give a more general description of strategies.
Further research requires the construction of conceptual models of human activity in complex problem-solving. It will probably make it possible to operationalize the concept of strategy concerning complex problem-solving in the future.