# Article : Capturing Clinical Expertise an Analysis of Knowledge "mining" through expert system development Paula S Nurius. Anne E. Nicoll. University of Washington Clinical Psychology Review, Vol 12. Pp-705-717, 1992 Read [[24-10-2022]] - It's not just about "WHAT" information is important, but "HOW" to synthesise information and applying it effectively that's important. - It is hard for practitioners to articulate how they integrated certain information or how they arrived at certain conclusion - [[Tacit knowledge]] - [[Knowledge Management]] - This is related to how human memory work - Recent models of human memory indicate that knowledge of how to do things (procedural knowledge) is stored in a different form than knowledge about concepts and basic facts (declarative knowledge: Anderson, 1976, 1983). **Declarative Knowledge** - Facts, concepts, **Procedural Knowledge** - Include behavioural knowledge about how to respond. - "If-Then" statements - IF a certain situation exists, THEN perform a certain overt or covert action (Linville & Clark, 1989; Nasby & Kihfstrom, 1986). All new knowledge are declarative knowledge, with much practice, it get automatic. It is hard to extract once it become unconscious/automatic. ## Expert Systems as a vehicle for exploration. - As a way to operationalise core declarative and procedural knowledge and strengthen current methods of eliciting this knowledge. ## Initial Specification of the Expert's Mental Schema Need to specify and establish knowledge domain that are relevant to the problem. - What knowledge is needed? - What is the basic framework in which this knowledge is organised? - What are the clusters of information can the expert consciously articulate as being common and important? 1. Describe a typical counselling session with a client, and 2. explain the course of her reasoning as she moved through initial interview 1. Think Aloud - what associations come to mind. 2. Prompts - 1. What questions would be universally important to ask in the first session and why? 2. What kinds of decisions commonly need to made? 3. What goals might a practitioner generally or universally have with that kind of problems? 4. What assessment factors were important to consider? 3. Differentiate observations or conclusion from 1. Observable facts - from the session/environment 2. Inferable facts - data largely from practitioner interpretations. 1. Such inferences may or may not be valid 2. These knowledge structures are activated by cues. 4. Interesting : an inexperienced practitioner would likely make an accurate assessment, but experts can reach these conclusions with fewer questions, through relying less on observable facts and more on ==information composities, inferences, and experience-driven probabilistic reasoning==. ![[Screenshot 2022-10-24 at 10.43.25 AM.png]] Figure 1 is an example of how expert organises his mental schema. ?What is the mental schema for case managers? ## Specifying the Scope of the Prototype - What i gathered from this passage is that, experts rely on wide scope of information; may not be directly related to the cases/issue at hand, it is hard to differentiate, to say what is important or not. - What makes expert an expert is the experience and wider connections he is able to make. ## Building the knowledge base - The expert and knowledge engineer walk through case examples, - expert identified major parameters or factors important in the assessment, - Usually these categories are inline with training received (aka what is usually taught), - However, tacit knowledge or common sense are harder to explain, - e.g The assessment (e.g how risky the case is) could be influenced by contextual factors - e.g organisational constraints, availability of resources, caseload factors - all these could influence him at any given point of time. - This is where potential sources of individual variability that cannot be dealt with in knowledge acquisition and representation. - At this point, expert try to DECODE how he know, while knowledge engineer ENCODE them in models. ## Articulating the Inference Flow **The following giving me idea about how to design [[Digital Interventions]] in the project. How case managers have a mental schema for assessments, and each factors having indicators and interventions are recommended based on an continuum of intensity.** - To develop rules, a system of indicators was specified for each dimension (Severity, imminence, intervention), thresholds then could be operationalised to differentiate imminence of danger. - Yes vs No. - Intervention options were not viewed by the practitioner as discrete but on a continuum of intensity. - ![[Screenshot 2022-10-24 at 10.59.52 AM.png]] ==The above graph, is like a model. What if i translate case management into such a graph?== - One of the difficulties the expert encounter is to mange "it depends" terms. (Because we know human issues are not so clear cut) - That is because expert are more aware of the different conditions, (more differentiated), not just a overall, global "IF-THEN". Experts see many "IF-THEN" conditions. - Thus can go into very fine subsets. - "...The expert was in a position of having to interpret her own knowledge base: to extract the general principles that at one time she reflected upon very deliberately but had since replaced with a net of finer, more context-specific rules."... - ==But it is not clear at this level, is the model representing what "is in there" vs expert theory of himself. (thus a question of validity) - Will this model make sense universally? == - I guess that's why we need more than one expert, Big data, to build robust models. #Biases [[Book - Machine Learning for Dummies]] ## Sobering Realities and Promising Possibilities - Limitations, due to computer technology, or software features, sometimes, the way knowledge are represented is limited by how the software work. - Rather then software working like human brain... human adapt to machine. - Difficulties in capturing expertise when trying to determine how many, under what conditions, and to what extend factors need to be incorporated .. just too many variables. --- This article provided some good overview of this process of "mining" expert domain knowledge. I am curious about this topic, hope that it will help me with improving digital interventions designs. Here, more articles citied this article. https://scholar.google.com/scholar?cites=14735597413437097305&as_sdt=2005&sciodt=0,5&hl=en