It is hard to even fathom let alone comprehend the size, scale, and complexity of the universe. It takes light 91 billion years to traverse the diameter of the observable universe. That is approximately six times longer than the age of the universe itself. Similarly, the amount of information in the world vastly overwhelms the information that humans are capable of processing. For example, the brain receives between 12.5 gigabytes and 2.5 terabytes external information per second but has the capacity to focus on only 60 bits of information per second. That is data compression on the order of 10¹². In reality, we are biologically unequipped to even process the vast majority of that information. Visible light comprises approximately 0.0035% of the electromagnetic spectrum. The audible sound ranges between 20 and 20,000 Hz, therefore, anything less than 20 Hz and greater than 20,000 Hz is unheard by the human ear. The processable external information is further filtered, compressed, and synthesized in a process termed exteroception. This is not limited to the external world but internal information is also compressed, filtered, and synthesized in a process termed interoception. Bodily signals such as a racing heart, dry mouth, and pain are contextualized and interpreted by the brain. Therefore, external and internal realities can never mean anything more than a compressed model of an immensely complex external and internal real world. Therefore, it is not surprising that we not infrequently ”draw illogical conclusions from accurate assumptions or logical conclusions from inaccurate assumptions.” [Donella Meadows]
Models take many forms from the mental and the verbal to the mathematical and physical. Mental models are informal abstractions carried in our minds that are often limited to a few variables and are not directly accessible by others. They are informed by both objective and subjective experiences and often operate on the subconscious level with simple categorizations. In contrast, technology in tandem with the scientific method has enabled us to expand our perceptual abilities and create physical and mathematical models incorporating more variables. Ideally, through training, our informal mental models should interact with formal physical models to offer a more comprehensive view of reality. Health care providers routinely work at this intersection of formal and informal models with every patient encounter. Since we have an in-depth knowledge of human physiology, we can utilize appropriate models to make relatively accurate predictions. Nevertheless, even in a relatively limited domain of human physiology, the unknown vastly surpasses the known. Since models are only useful if they capture relevant aspects of the real world and leave out unnecessary details, the possibilities of errors in predictions are present at all levels of the model from the individual (the way we “see” the world and ourselves) to the level of science (theories). Let’s take an example of the former, explained through the relatively simple example of a patient presenting to the emergency department and tease out the potential for errors in modeling and prediction.
The patient experiences a specific set of symptoms. She utilizes her interoceptive capabilities to gauge and assess her symptoms. However, there is variability in the capacity of interoception. Some people are accurate (interoceptive accuracy) or have a metacognitive awareness of accuracy (interoceptive awareness), while others have a heightened tendency (interoceptive sensibility) in detecting bodily signals. These symptoms are interpreted or misinterpreted in the social, historical, emotional, and cognitive milieu of the individual. Based on this complicated background, a decision is made to go to the emergency department. In the emergency department, the symptoms are internally transferred the to brain modules responsible for communication. However, since language is a product of a complex set of factors such as education and culture, word choices are limited and often do not express interoceptive signals accurately. Therefore, communicating a set of symptoms is a process of modeling your internal world within the constraints of language. Once the symptoms are communicated to the nurse, he utilizes objective data (i.e vital signs)) and his training in tandem with his subjective conscious and subconscious experience (i.e. norms, categories, biases) to decode into a model the symptoms into the electronic health record. The physician thereafter is the presented with the synthesis of the multiple layers of models into a “reason for visit.” She incorporates that synthesis and layers her own model after her interaction with the patient.
At each step, in a complex interplay of subjective and objective experience, the modeler is capturing necessary data and filtering out unnecessary data. This is why models are useful and powerful – they can distill a complex and unfathomably large reality into a simplified but comprehensible version. However, therein lies the catch. The potential for filtering out what is necessary but deemed “unnecessary” and incorporating what is “unnecessary” but deemed “necessary” increases with every model. For example, the patient might not express her symptoms in a manner that the nurse or doctor comprehends. The nurse might be biased by her last patient who presented similarly and therefore, modeled the presentation inaccurately. The physician could be subjected to a range of cognitive biases based on the nurse’s model and incorporates unnecessary data and filters out needed data from the patient. Therefore, the potential for an error in prediction compounds at every step.
David Hume stated, “as to those impressions, which arise from the senses, their ultimate cause is…perfectly inexplicable by human reason, and ‘twill always be impossible to decide with certainty, whether they arise immediately from the object, or are produced by the creative power of the mind.” However, the human desire for certainty and causality combined with the illusory sense of “knowing the world” and “knowing ourselves” leads us away from the fact that the representation of the “real world” and even “ourselves” are models limited by coarse categorizations and cognitive biases. Science, technology, and computation have improved our ability to make models that can capture more of the complexities of the real world. Nevertheless, the fact remains that “our knowledge is amazing; [but] our ignorance even more so.” [Donella Meadows] An ideal decision maker is keenly aware of those facts and can inform her informal mental models with the formal models of science.