Kids, Cats and Concepts: Toward a Grand Unified Theory of Thinking Geoff Norman The Goal To link research in three domains: Dual processing models of thinking Exemplar and prototype models of categorization / concept formation Expertise and clinical reasoning to a greater understanding of human information processing How I got there Distant Studies of clinical problem-solving Intermediate Role of experience in clinical reasoning Recent Diagnostic errors and “dual processing” A Difficult Diagnostic Task An Easy Diagnostic Task The rule is insufficient for the classification task But we can do the task quickly, accurately, and effortlessly HOW? The Role of Similarity DUAL PROCESSING Two basic strategies System 1 Based on holistic similarity to prior examples “Exemplar theory” (more later) System 2 Based on underlying conceptual characteristics “Causal models” System 2 thinking Playing by the rules Analytic View of Expertise “The matters that set experts apart from beginners are symbolic, inferential, and rooted in experiential knowledge…Experts build up a repertory of working rules of thumb or “heuristics” that, combined with book knowledge, make them expert practitioners.” E. Feigenbaum. The fifth generation: artificial intelligence and Japan's computer challenge to the world. 1983 System 1 thinking I’ve seen it before and here it comes again Successful categorization From 2-D abstract representation without analysis of features without language Successful generalization To other 2 D abstraction in atypical orientation The Non- Analytic View “We must be prepared to abandon the traditional view that runs from Plato to Piaget and Chomsky that a beginner starts with specific cases and… abstracts and interiorizes more and more sophisticated rules.It might turn out that skill acquisition moves in just the opposite direction; from abstract rules to particular cases.” H.L. Dreyfus, 2002 Outline Dual processing Concept formation and categorization Expertise and Clinical Reasoning Applications -- Implications Levels of Processing Perception Transfer Aging and reasoning Intelligence Three Literatures Dual Processing Concept formation (Thinking) (categorization) Stanovich, Evans, Kahnemann Medin,Brooks Clinical Reasoning Norman,Schmidt Dual Processing System 1 Rapid, unconscious, based on concrete similarity, “just” pattern recognition System 2 Slow, logical, conceptual, energyintensive, CHARACTERISTICS System 1 System 2 Unconscious Conscious Implicit Explicit Automatic Controlled Effortless Effortful Rapid Slow Holistic, Analytic Old (evolution) New (evolution) Contextualized Abstract Neuroanatomy of System 1,2 System 1 “right inferior prefrontal cortex” Evans, 2008 “Involves hippocampus” Smith & DeCoster, 2000 System 2 “ventral medial prefrontal cortex” Neurophysiology of System 1,2 Glucose dose (vs. Placebo) Shift of processing strategy toward System 2 (more energy demand) with glucose load (Attraction effect - 17% vs. 47%) (Masicampo & Baumeister, 2008) Mental representations SYSTEM 1 Abstract concepts Feature list, probability, causal mechanism, process SYSTEM 2 ??????? Categorization / Concept Formation Exemplar Theory Medin, Brooks Categories consist of a collection of prior instances identification of category membership based on availability of similar instances Retrieval process is “non-analytic (unaware), hence can result from objectively irrelevant features Retrieval process is not deliberate, not available to introspection (Like System 1) Dual Processing in Medicine From Process to Knowledge (Analytical and Experiential) The beginnings clinical reasoning as a process “Hypothetico-deductive method” (Elstein, Shulman, Sprafka, 1977) Expert (and novice) clinicians generate multiple diagnostic hypotheses early in the encounter then gather data to confirm (usually) these hypotheses Does hypothesis predict accurate solution? 100 90 80 70 60 Correct 50 on chart 40 30 20 10 0 Present Absent Correct hypothesis? Barrows, Neufeld, Norman, 1981 Where do hypotheses come from? “Medical experts differed from novices in that they generated better hypotheses…… and we don’t know why!” A. Elstein Dx Error Conference May 31, 2008 Expert Physicians and Dual Processing To what extent does the: formal knowledge of medical school vs. experiential knowledge of practice contribute to expertise Schmidt & Norman, 1991 Novice Intermediate Expert Basic Science Mechanisms Clinical Rules Examples System 2 Basic Science Mechanisms System 1 Clinical Rules Basic Science Mechanisms Who do you pick? Dr. JW completed the specialty exam last year and stood 14th in the country. Dr. WS completed the specialty exam 6 years ago. At the time, she was in the top 1/3 of all candidates. The Conundrum Why do we prefer the candidate with apparently less “competence” but much more experience? What did she get from 10 years of experience? 10 years of experiences (System 1 knowledge) BUT Every measure of formal (System 2) knowledge decays right after graduation Day and Norcini, 1988 540 520 500 480 460 440 420 <20 21-24 25-29 Years since Graduation 30-34 35-39 Evidence of System 1 in Diagnostic Reasoning Visual Diagnosis and Response Time STUDY 100 slides in 20 categories Students, clerks, residents, GPs, Dermatologist Accuracy and Response Time Accuracy by Educational Level 100 90 70 60 50 40 30 20 10 og is t er m to l GP Re si de nt Cl er k 0 St ud en t % correct 80 Response time by Educational Level 30 20 Correct Incorrect DK 15 10 5 og is t er m to l GP Re si de nt Cl er k 0 St ud en t Response Time 25 Evidence of Exemplars Effect of Similarity (Allen, Brooks, Norman, 1992) 24 medical students, 6 conditions Learn Rules Practice rules Train Set A (6 x 4) x 5 Train Set B (6 x 4) x 5 Test (9 / 30) Accuracy by Bias Condition 90 80 70 60 50 40 Correct 30 Other Incorrect 20 10 0 Bias Corr Bias Incorr Is it just visual similarity? If it’s “non-analytic” does it apply to objectively irrelevant features? ECG Interpretation Hatala et al, 1999 Medical students/ Fam Med residents PRACTICE (4/4 + 7 filler) middle aged banker with chest pain OR elderly woman with chest pain Anterior M I TEST ( 4 critical + 3 filler) Middle aged banker with chest pain Left Bundle Branch Block RESULTS Percent of Diagnoses by Condition 50 Percent mentioning 40 30 Bias 20 No bias 10 0 Correct Prior Diagnosis CONCLUSIONS Medical Diagnosis and Dual Processing Experiential knowledge is a major contributor to diagnostic expertise Categories and concepts are based on our specific experience with the world These specific experiences are accessed and used without awareness When do experts use system 2? Analytic reasoning and Diagnosis Invoked for confirmation in all Dx encounters Analytic knowledge of many forms: Illness scripts Symptom-disease probabilities Semantic axes Feature lists (e.g. DSM 4) Where Do Clinicians Use Basic Science? Most use basic science rarely? Observational studies (Schmidt, Patel) Some use basic science some of the time Difficult problems in nephrology Some use physiology ALL the time Intensivists, anesthesiology Most use it rarely (Patel, Schmidt) Clinicians rarely use basic science explanation in routine practice. While they may possess the knowledge, it remains “encapsulated” until mobilized for specific goals (to solve specific problems) (Schmidt, HG) Some Use it with Difficult Cases (Norman, Brooks, Trott, Smith) When experts are confronted with difficult cases, do they revert to causal reasoning? Experimental Design R1 --GP n=4 R2 -- IM n=4 Clinical Cases k=8 Explain and Diagnose Nephrol n=4 Diagnostic Accuracy 1 0.8 0.6 0.4 0.2 0 R1-FM R2-IM Nephrol Causal Explanations 2.5 2 1.5 1 0.5 0 R1-FM R2-IM Nephrol No of Diagnoses / Investigations 6 5 4 3 2 1 0 R1-FM R2-IM Nephrol Conclusions Use of Basic Science In difficult diagnostic situations, clinicians use causal physiological knowledge and analytic reasoning Expertise associated with more coherent explanations, better diagnosis Dual Processing and Experience With increasing experience, do people rely more or less on System 1 -- Nonanalytic reasoning? Studies of Relative Experts (Moruzi, Brooks, Norman, 2003) Dermatologists/ GPs / residents 36 slides (typical / atypical) Condition A Verbal description of slide (verbal) then photo (visual + verbal) Condition B Photo only (visual) Diagnostic Accuracy 90 80 70 60 50 Resident 40 30 20 10 0 Verbal Verbal+Visual Visual Diagnostic Accuracy 90 80 70 60 50 40 30 G.P. Dermatol Resident 20 10 0 Verbal Verbal+Visual Visual Hatala et al. ECG Diagnosis Prior match / unmatch history Postgraduate residents and med students RESULTS Percent mentioning Percent of Diagnoses by Condition 50 45 40 35 30 25 20 15 10 5 0 Bias No bias Correct Prior Diagnosis Medical Students Dual Processing and Instruction Role of Instruction in reasoning Since NA (System 1) reasoning occurs at all levels, is effective, is “automatic” You can’t: tell student to not do it tell student to beware of biases tell student to think of better diagnoses Does a coordinated strategy improve accuracy? Norman, Brooks, Colle (ECG) Schmidt and Mamede (Gen Medicine) Ark & Eva, (ECG) Norman, Brooks Colle, 2000 Contrast instructions to: Think of the first thing that comes to mind, then consider features vs. Gather all the data then arrive at diagnosis 32 Undergrad Psychology students 11 disorders, rules + examples Test -- 10 new ECG’s Diagnostic Accuracy 70 60 50 40 30 20 10 0 Pattern + Rules Rules Schmidt & Mamede, 2005 42 I.M. residents 16 written cases --- simple / complex Within subject/case design Instructions: “First thing that comes to mind” vs. “Hypotheses, findings for/against, differential, ….” Diagnostic Accuracy 80 70 60 50 Sim ple 40 Comp lex 30 20 10 0 Exemp lars Rule s ECG Diagnosis - Ark & Eva 48 undergrad psychology students 8 ECG diagnoses (A/A’, B/B’, C/C’,D/D’) Instructions • Compare and contrast vs. Sequential • Combined Analytical/Non-analytical vs. usual approach Test 20 ECG’s (10 old, 10 new) Immediate / 1 week later Effect of Examples and Instructions on New Cases after One Week 80 75 70 65 60 55 50 45 40 35 30 Combined No Instruction Compare Ark & Eva, 2005 No Compare Conclusions - Dual Processing and Diagnosis Evidence that clinicians access both kinds of knowledge/ use both processes Evidence that with increasing experience, greater reliance on system 1 Evidence that students benefit from explicit instruction to use both Dual Processing and Thinking DP and levels of processing DP and perception DP and transfer DP and aging Dual Processing and Levels of Processing Are conceptual structures and deep processing an underpinning to development of System 1 (nonanalytic) knowledge? (Dreyfus) Role of Basic Science in Novice Reasoning (Woods, Brooks, Norman, 2003) 4 neurology / muscular diseases 36 medical students Basic Science or Symptom/Disease probability Measurement Diagnostic Test 15 cases, 4-6 features Administered at 0, 7 days Score on Dx Test 55 50 45 Feature List Basic Sci 40 35 30 Immediate 1 Week Score on Dx Test 55 50 45 Feature List Basic Sci 40 35 30 Immediate 1 Week Score on Dx Test 55 50 45 Feature List Basic Sci 40 35 30 Immediate 1 Week Dual Processing and Perception Word Superiority Effect Higher - level concepts (words) in memory facilitate recognition of elements of words and pseudo-words: RINK BINK NRIK - possibly because of rapid (top-down) then slow (bottom up) processing Influence on Feature Interpretation Diagnostic hypotheses arise from pattern recognition processes based on similarity to prior examples In situations of feature ambiguity, hypotheses may influence what is seen Influence of Diagnosis on Feature Perception (LeBlanc et al) 20 residents, 20 final year students 8 photos of classical signs from clinical diagnosis textbooks Correct history and diagnosis vs. Incorrect history and diagnosis RESULTS Diagnostic Accuracy by Bias 90 80 70 60 50 Student Resident 40 30 20 10 0 Correct Alternate No. of Features RESULTS Number of Features of Correct Diagnosis by Condition 0.5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 Student Resident Correct Alternate Diagnosis RESULTS Number of Features of Alternate Diagnosis by Condition No. of Features 0.25 0.2 0.15 Student Resident 0.1 0.05 0 Correct Alternate Diagnosis Dual Processing and Transfer Although medical (and other) study is directed at conceptual learning, use of conceptual knowledge to solve problems (transfer) is rare and difficult. WHY???? Spontaneous Transfer 8 high performing undergrad (Health Sciences) students. 3 principles (Laplace, Poiseuille, Starling) 12 test cases Score 0= wrong answer, 1= right answer, wrong explanation 2 = right answer, right but poor explanation 3= right answer, good explanation Laplace’s Law In a cylindrical vessel, the wall tension is proportional to the radius and pressure exerted by the vessel contents. This can be expressed as T = PR where T is wall tension, P is pressure exerted by the contents, and R is the radius of the vessel. A 72 year old female has been diagnosed with an aneurysm (dilatation) of the aorta. The doctor tells her that if it grows to 5 cm in diameter she will need surgery to prevent bleeding. Explain why the increasing diameter is a problem. Average Score 15.2/36 = 42% “…during early learning, the principle is only understood in terms of the earlier example… the principle and example are bound together. Even if learners are given the principle or formula, they would use the details of the earlier problem in figuring out how to apply that principle to the current problem” (Ross, 1987) Why are the examples so seductive? System 1 Fast, unconscious, contextualized, concrete System 2 Slow, logical, abstract Transfer amounts to overriding System 1 to utilize abstract, conceptual information DP and Age Evidence from psychology that with increasing age, we rely more on System 1 thinking Eva & Cunnington, 2006 15 family docs, 7 < 60 yr., 8>60 yr. 8 cases: 2 diagnoses, 4 conditions Generated, Provided, Privileged, Extreme --------------->>>>>> weight on second diagnosis Diff (Dx 1 - Dx2) 50 40 30 20 Young Old 10 0 -10 -20 -30 Gen Prov Priv Extreme Some Last Words (from other people) {the expert} does not solve problems. He does not even think. He just does what normally works, and, of course, it normally works…. The expert is simply not following any rules! He is… discriminating thousands of special cases. H Dreyfus “In general, to preserve expertise we must foster intuition at all levels of decision-making, otherwise wisdom will become an endangered species of knowledge.” H. Dreyfus First and Last Word on Expertise “It is a profoundly erroneous truism, repeated by all copy-books and by eminent people making speeches, that we should cultivate the habit of thinking about what we are doing. The precise opposite is the case. Civilization advances by extending the number of operations which we can perform without thinking about them. Operations of thought are like cavalry charges in a battle -they are strictly limited in number, they require fresh horses, and must only be made at decisive moments.” A.N. Whitehead, 1911 (in J Bargh, 1999) Bibliography Dreyfus HL From Socrates to expert systems: The limits and dangers of calculative rationality. http://socrates.berkeley.edu Evans J St BT. In two minds: dual - process accounts of reasoning. Trends in Cognitive Science 2003; 7: 454-459 Evans J StBT. Dual processing accounts of reasoning, judgment and social cognition. Ann Rev Psychol 2008;59: 255-78.