Perception, Cognition and Thought: Part III: Reasoning, Judgement and Decision-Making

A human being is a part of the whole called by us universe, a part limited in time and space. He experiences himself, his thoughts and feelings as something separated from the rest, a kind of optical delusion of his consciousness.

- Albert Einstein

Introduction

In the first two segments of this series, we have seen that perception, cognition and thought along with conceptual and symbolic processing characteristic of higher human abilities represent both a blessing and a curse.^1,2 They are a blessing and, indeed, a necessity in successfully interacting, surviving and thriving within the environment by recognizing, manipulating and building on concepts and symbols that are inherited at birth and further developed over time with experience by attending to certain characteristics of our environment and ignoring others that might distract from responses necessary for survival. At the same time, concepts, language and thoughts that are imprinted and otherwise imbedded in us through evolution, development, education and experience including socialization, limit our ability to recognize and understand other aspects of our environment and, in a sense, bias or even blind us to much of what represents the underlying reality of our existence and the universe. For most of us, it is only the rare occasion, if at all, that we gain brief glimpses of that underlying reality hidden from us by virtue of how we perceive, think, and carry out our daily lives. Now, we take another step into our understanding of the nature and effects of human cognition and understanding which enable us to mentally and physically act upon the world around us.

Origins

As early life emerged from the water, it no longer faced only a world, food and predators within a few hundred feet or less requiring a rapid response to a threatening or appetizing stimulus. Suddenly living things on land could see or otherwise sense a much wider world often considerable distances away providing the opportunity or even the necessity of considering more complex options and anticipate opportunities. Therefore, rather than a rapid response, life was faced with even greater choices and options requiring or encouraging a decision to be made. Thus, the ability to make decisions and, in fact, to make wise decisions developed because of its importance to surviving and even thriving in this new environment. In humans, thoughtful consideration of the available options and reasoned decision-making became fundamental to our existence and conscious awareness. Routine, repetitive choices soon became unconscious acts while recognizing new challenges or opportunities became and remain central to our conscious awareness and a sign of intelligence. At the same time, we have learned that decision-making in man is fraught with many challenges or biases in their own right that can lead to faulty decision-making including many possible logical fallacies, cognitive biases, innumeracy etc.

Formal approaches

More formal approaches to decision-making have emerged based on the principles of logic and critical thinking in an effort to avoid many of these pitfalls and to encourage more logical and error-free reasoning. These approaches are based on an assumption that there is one or more optimal decision based on the primary goal(s) of the decision maker, the probability of the possible outcomes as well as a quantitative assessment of value or utility of the outcomes in achieving the primary goal(s). Often the outcome is a composite of individual outcomes reflecting generally mutually exclusive goals. However, not all human goals and values (utilities) can be objectively quantified and often the probabilities among the various outcomes are subjective at best and unknown at worst. Nevertheless, such an approach has served as a ‘check’ on the more common subjective intuitive or ‘gut’ feeling approach to decision making. Most often, the probabilities and utilities are gathered from previous experience, research, publications or expert opinion all of which have potentially serious limitations of bias, availability and accuracy and may not reflect the primary goals and values of the individual impacted by the decision.

Limitations of decision-making

Clearly, decision-making in the setting of uncertainty is a complex process of generating hypotheses, acquiring and evaluating often complicated information, and ultimately a decision prompting some action.^3,4 Decisions are premised on a knowledge base derived from a collection of facts gathered over years of formal education, training and experience which will vary considerably between individuals and over time. At the same time, rational decisions require a set of rules to be applied to these facts to guide how to gather information and evaluate it in order to arrive at the best decision from a specified perspective. While the fundamental rules applied to clinical decisions change little, there is clearly considerable variation in reasoning skills between individuals with differing training and experience.⁵ Reasoning abilities vary from individual to individual with only a few hypotheses capable of being considered at a single time.⁶ New information, as well as past experience, is often utilized incorrectly with errors in reasoning occurring at several points in the process, including problem formulation, probability estimation, utility assessment, and problem evaluation and extrapolation.^7,8 Studies of skilled clinicians suggest that probability estimation is among the most vulnerable features of clinical judgment.⁹ While the representativeness heuristic, in which we estimate the probability of an event based on how similar it is to a known situation or those generally found, is often employed. Errors in this approach include attention to nonspecific findings, over-attention to redundant predictors, misinterpretation of regression to the mean where more extreme events are often followed by more moderate ones, limited experience with the disease in question, and inattention to the prior probability of disease.^3,7,9,10 Clearly, a rare disease is still unlikely even if it manifests typical characteristics while a common disease may be more likely even if presenting with unusual features.¹¹ Another commonly employed heuristic is that of anchoring or making rapid assessments based on the initial information or probability estimate received. Subsequent adjustment of the probability estimate considering other specific factors appears to result in consistent overestimation of the risk associated with a positive test result.^12,13 As Groopman points out in his book “How Doctors Think”, physicians often make diagnostic and treatment decisions very early into their interaction with a patient, sometimes within seconds.¹⁰ They often interrupt the patient even before they have had the opportunity to fully describe their symptoms. While sometimes such decisions turn out to be correct, all too often they are not and may have catastrophic consequences as a result. As Groopman notes, such an approach may result in a misdiagnosis as often as 20%.¹⁰

Determinism versus free will

In the end, our consideration of human decision making cannot be separated from the age-old consideration of determinism versus free will. If all decisions are ultimately linked at their deepest underpinnings, to our genetics, biology, training and experience, then there is little to distinguish them from a deterministic universe. Proponents of determinism note that if a decision-maker is queried as to why they made a specific decision or choice, they will almost always indicate one (or sometimes a litany) of factors that ‘justified’ their choice and presumably influenced their decision. Considering themselves rational thinkers, the decision maker seeks to justify the choice they made in terms of reasonable causal factors, experiences or situational factors. Of course, taken to its natural inference, the determinist argues that this is evidence that the decision was, in fact, ‘determined’ by these factors at least in part if not entirely. Those favoring free will, on the other hand, argue that, in the end, despite the various factors impacting on our actions, the final decision of a conscious individual is made freely and could just as well be different. Many and undoubtedly most of our daily actions that might be considered ‘decisions’ are based on prior experience and responses to situations and are made subconsciously such as driving to a certain destination. However, these do not appear to be true conscious decisions, but reflex responses based on prior experience and training. A truly new conscious decision or a decision to do something different than we have in the past, when all else is the same, appears to represent a conscious decision made freely.

Probabilities versus certainties

Either way, in the current prevailing view of underlying reality as a probabilistic universe, the choices, options and future events, in general, are not considered determined in the traditional sense of the word but merely probabilities until results are observed or decisions are made by the observer or conscious individual. Based on the prevailing view of reality, supported by considerable science, we are led to believe that our underlying reality, including our conscious awareness, is probabilistic and uncertain until observed or acted upon leaving room for free will within our decision-making. The choices or actions and eventual outcomes available to the decision-maker are seen as having certain probabilities of happening until such time as an actual decision is made collapsing the probabilities of available choices into the certainty of the specific choice/action. This does not alter the probabilistic nature of subsequent outcomes or events but does resolve the specific step of the decision at hand. Decision-making, as with other conscious actions, is often part of a sequence of many probabilistic events, each of which influences the initial probabilities considered in subsequent decisions or actions.

The probabilistic reality of the world, however, does not change the underlying dilemma as to whether conscious decisions are freely made or not. A third view of reality related to decisions is that of conditional free will. While not entirely resolving the apparent conflict between free will and determinism, such a view imposes an intermediate step in the process that can be referred to as the desire, wish or goal of the decision-maker. The argument, made by Locke, Hume and others, is such that decisions are made freely conditional on the desires or wishes of the decision maker. This view does highlight the likely importance of goals or the ‘search for meaning’ in our decisions and focuses our past experiences, education and upbringing on the development of one or more over arching life goals that drive many of our decisions. Others argue that all of this simply ‘pushes the can down the road’ in that desires and wishes are, in large part, if not entirely, determined by past events, experiences, education etc and not entirely free. Nevertheless, it does provide the opportunity for greater insight into the dynamics of decision-making and how ‘free will’ is allowed or even required under certain conditions. Finally, thinking about decisions must always consider the context in which a decision is to be made. Are there really alternative choices that a rational thinker can make? Can different choices appear to achieve the same goals and therefore approach equipoise where individual choice might not be influenced one way or the other by prior events or knowledge? Most individuals believe intuitively that there is some degree of free will in the decisions that we consciously make regardless of whether that free will is absolute or is conditional upon our wishes and desires. However, virtually all of us realize that there may be serious limitations to making good decisions due to our susceptibility to biases, logical fallacies and other errors in optimal reasoning. The nearly limitless scope of such logical fallacies and erroneous reasoning have been extensively discussed by other authors and will not be further elaborated upon here.^3,4,6,10,14 Leave it to say that we have all made bad decisions and decisions that flew in the face of logic and common sense.

Clinical decision-making

We recognize that two different individuals faced with the same decision in very similar circumstances and with similar backgrounds will often make very different decisions. For these reasons, in many critical disciplines, including medicine, a more rigorous approach to decision-making is often sought and applied.^15,16 Most commonly, such formal decision-making is based on Bayes’ theorem which considers the subject’s baseline or pretest probability along with the known test performance characteristics of sensitivity and specificity in order to provide a revised, posttest or conditional probability given the test results. Such a decision framework requires a structuring of the question, knowledge of the performance of methods of information gathering such as tests, and an understanding of the likelihood and value of various outcomes.^5,7,9,17 Reasoning by clinicians and nonclinicians alike often fairs poorly compared to more formal methods under controlled conditions.^11,18 Any difference between human estimates and those generated by more formal algorithmic methods may be related to either our incorrect use of the pretest probability or incorrect revision of those probabilities based on presumed test performance and results.^3,11 While, physicians generally estimate test performance characteristics accurately, they consistently overestimate the effect of positive test results on the probability of disease.^11,12 In fact, the estimation of the probability of events is one of the most vulnerable features of clinical judgement.^9,14 At the same time, equipoise or the uncertainty principle requires sufficient uncertainty with a choice to presume that no likely difference in outcomes remains. This principle remains an ethical prerogative in the design and conduct of a randomized controlled trials.¹⁹ If there is insufficient uncertainty about the equality of outcome from such a trial, it is considered unethical to proceed.

Algorithmic decision-making

In order to most accurately estimate the probability of an event, it is essential that any a priori or baseline probability estimate, usually derived from published studies for a general population be adjusted for a specific situation based on additional information often in the form of observation or formal testing. This provides a conditional or posttest estimate of the probability of an event (p) in the specific situation. Statistically this can be formally presented as the posttest odds (p/(1-p) estimated by the pretest odds multiplied by the likelihood ratio consisting of the probably of a given result when the event, eg, disease, is present divided by the probability of the result when the event is not present. This process is readily programmable if the baseline probabilities are known and the effect or performance of all observations (measurements) including testing are known. However, the impact of observations is often limited given the complexity and range of observations or measurements that may be available as well as those that are unknown or unobserved. Therefore, while algorithmic approaches to decision-making may avoid many of the logical fallacies that human observers are vulnerable to, the ability of a seasoned or highly experienced observer to detect subtle or even subconscious cues that may impact on the estimation of event probabilities is difficult to capture. It has been argued that any difference in probability estimates between individuals and formal methods based on Bayesian analysis could be due to either the incorrect use of the pretest probabilities or the incorrect revision of these probabilities by specific test results.¹¹ Potentially, with future developments in artificial intelligence resulting in self-learning or strong AI applications, some of these challenges may be overcome with greater success than efforts to improve human clinician performance and reduce the frequency of logical fallacies even in experienced observers. At present, we appear to be left with the important role of the experienced human observer aided by the programmed assessment of available observations.

Expected values and utilities

In clinical applications, overall survival is often considered the gold standard and, when appropriate, may be very objectively and accurately estimated. However, often the overall goals of the decision to be made are uncertain or, at best, subjective with considerable variation across different observers or participants. In most decision settings the goals are multifactorial based on various parameters used to assess the ‘value’ of the ultimate outcome.²⁰ The expected value represents a weighted sum of the expected values of all possible paths where the weights are represented by the probabilities of various chance events. The expected value, in turn, is the outcome for the immediately preceding decision eventually culminating in an expected outcome for each decision choice considered. In real-world settings, the variance of the expected value can be estimated if the distribution of the individual probabilities and outcome values are known. In turn, univariable and multivariable sensitivity analyses can be performed varying assumptions about the probabilities and outcome values or even the decision choices. Under idealized settings, the optimal decision is that which is associated with the greatest expected value in a given setting.

Other major considerations in decision-making relate to the utilities or values of the outcomes associated with the various choices in order to achieve specified goals. Increasingly these consideration include the role of shared decision making between the decision-maker and the subject of the decision as well as consideration of ethical and moral factors.^21,22 In clinical settings, outcomes may involve not only survival or life expectancy but some composite outcome adjusting for toxicities, costs, social and functional impairment or overall adjustment for some measure of ‘quality-of-life’. There has been increasing recognition that conventional value measures of these outcomes fail to capture the actual patient experience and patient-specific values. As a result, this has prompted the current focus on capturing patient reported outcomes (PROs) in an effort to more fully understand the actual value of the available choices.²³ While often valuable for experienced clinicians dealing with very complex clinical scenarios or settings of profound public health importance, formal methods can also be helpful in training early decision-makers to apply formal logical approaches and hopefully minimize the frequency and impact of common errors in reasoning.

Conclusion

Ultimately, as we continued to be awed by the complexity and philosophical and neurophysiological issues around human reasoning and decision-making, we will continue to seek ways in which we can minimize our biases and improve reasoning ability. In the meantime, formal methods of reasoning and decision-making are available for both training and application in practice and will likely continue to improve through the use of Bayesian and various artificial intelligence applications. It is important, however, that we remain vigilant of the limitations of these methods and the need for them to augment or amplify rather than replace the critical role of human intelligence and awareness wherein lies important ethical, moral and intuitive abilities capable of constraining the potential abuse and adverse impact of such applications. For the foreseeable future, conscious, deliberate and ethical/moral human controls will be needed to assure that the ultimate objective of optimal decisions for individuals as well as for the betterment of mankind and our environment are sought and ultimately achieved.

Declaration of Interest

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the article.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.