Diagnostic Reasoning as Artificial Intelligence Emerges: A Distributed Cognition Framework

---

This is an invited guest post by Dr. Cory Rohlfson (@CoryRohlfsen) based on an interesting Twitter thread of his from a few months back.

---

After almost despairing over how artificial intelligence (AI) would one day replace us, I realized that it might not, but not for the reasons commonly cited (such as empathy, relationships , sympathy). Many have imagined what the future of medicine will look like with AI as a co-pilot. It’s just a matter of time The Turing Test Patient results prove the superiority of AI when in the driver’s seat. From then on, will doctors be in the cockpit too or will we be left peering into the proverbial dust of the runway as the AI ​​takes flight? What generation of doctors can expect this to happen in their lifetime?

Because much of our identity as physicians is rooted in the cognitive domain, it is normal to wrestle with these questions, especially as our roles and responsibilities evolve. If you’re reading this as an experienced doctor, imagine the disproportionate impact such disruptive technology could have on a trainee whose identity is being actively faked.¹ As an internist who practices both primary care and hospital medicine, I’ve been through all the stages of grief, negotiating a few downward doomsday thoughts on my way to acceptance. Eventually, a more optimistic view emerged – a view that I am thrilled to share with you today.

Certainly, doctors will always have the upper hand in our physical, healing presence. But that’s not what this post is about. We also have a certain unique cognitive trait to offer – one that took over 10,000 years to develop.

Since the applications of AI in medicine are numerous, the scope of this post will be limited. appraisal. Although this proposed framework can also be applied to a wide range of managerial reasoning and clinical decision-making (including ethics, judgment, uncertainty, etc.), let’s start small. Evaluative reasoning – A strictly academic domain.

Let’s first acknowledge that our analytical brains appreciate hypothesis-based reasoning and problem solving. We are attracted to what we can measure, study and improve. But in doing so, we ignore our greatest contribution to medicine.

Yes, AI will outperform average doctors. An information-rich environment.² It will process high volumes of data with high speed, high fidelity, and relentless capacity. But only a part of the assessment is revealed in this domain. Most diagnoses come from this. Information deserts.

Remember that ~80% of diagnoses are caught during history.³ Even if the diagnosis cannot be confirmed until an exam, lab, biopsy, or imaging study is completed, the biggest breakthrough in the presumptive inquiry usually occurs in the patient interview. It also happens to be where “humanity” prevails and shines brightest.

Since the discovery of mirror neurons,⁴ Humans do not yet know what this “super power” is called. Some call it experience, pattern recognition, situational awareness, emotional intelligence, or action learning. For the rest of this post, I’ll call it “Realized perception“

Whatever it is, we have a superior. find.⁵ Works in information deserts. With greater sensitivity for recognizing socially encoded cues and patterns (separating signal from noise), we are programmed to navigate these deserts even if we don’t know we’re doing it. are

Remember those mall maps back in the day? The ones you went to when you were lost and needed to find the nearest exit to your car. The good guys signaled “You are here” with a big red star.

What was harder for the mall maps that didn’t have that big red star?

• Task 1) Find yourself on the map (locate) or…
• Task 2) Once oriented, map your route to the destination

We know the task of finding this elusive star is very taxing! Once you’re oriented, it’s not hard to find your way around. This search function represents an information desert and we all navigate it differently. Novices can scan “left to right” (taking in every detail to avoid missing a star). But experts usually have a strategy – even if it’s an unconscious one.

Search strategies may include:

1. Finding the center of the map (a classic histology trick on a standardized test)
2. Scanning a key (looking for symbols) or…
3. Heading towards a well-known landmark (eg a food court)

The thing is that a Scientific research Differs from analytical reasoning because it is context-dependent, situated in a unique time and place, and often unconscious (split-second, reflexive judgment cues). When adding humans to the mix, it’s even more variable and “fuzzy” – this is where “situated perception” becomes very important.⁵

As tacit data is exposed, a pattern emerges and only then is the data codified into information. Only then can hypothesis generation begin (as the spontaneous problem shines brightly). Like getting lost in a mall, a map only helps you if you know where you are. Our superpower as humans is turning to this problem – a human problem in all its biosocial wonder.

For medical professionals or medical specialists who have given up on history to reduce it to any skill and sensation, I hope this post is re-invigorating because every medical issue begins as an information desert. , like a mall map without a star. Novices will search with inferior strategies, but experts will benefit from their situational awareness and experience. A preclinical student might learn to memorize a “comprehensive review of systems” while a third-year student learns that some questions are more appropriate than others. Similarly, a fourth-year medical student aims like a postgraduate trainee to be hypothesis-driven in his data collection. But there is a higher level of patient interviewing, which I call “local search and hypothesis refinement.”

If targeted history taking is trained within a hypothesized deduction framework, we will be perfectly capable of achieving most diagnoses. The missing piece to becoming a complete expert is a feedback cycle of natural, situational interview cues that inform (and refine) a hypothesis-driven line of questioning. Simply put, the skill of interviewing a patient involves sensitivity to subtle, nonverbal cues – something we evolved to be really good at.

A period of immunity.

A hint of sarcasm.

A look of annoyance or pain.

or loss of eye contact when the history begins to blur.

Each of these moments represents a window of opportunity to explore. When the evaluator says, “I let out a big sigh after mentioning ____, can you tell me more about it?”

Expert appraisers will find an oasis in the desert. Through feedback cycles of “sensitive hypothetical inquiry,” they pinpoint the problem where no one else is looking.

Since this SEARCH function is rarely talked about, it can help to contrast it with an analytical mind. Since analytics is a form of information processing (1 + 2 = 3), it will work well in information-rich environments with high signal:noise ratios and Limited possibilities. AI is already showing great promise here. On the other hand, situated cognition (“sit cog”) thrives in information deserts with “fuzzy” signal(s) and Unlimited possibilities.

A set cog knows “how” even if it doesn’t know “what” or “why.”⁵Thus, it will take decades for AI to compete with humans in this domain.⁶ Thankfully, it’s not a competition (and it shouldn’t be). Combining these two highly developed forms of cognition results in a complimentary problem-solving system that is superior to using either one in isolation.^7-8

In other words, from a “Distributed Cognition” approach, it is valuable to capture the “net sum” of the signal, regardless of how it is collected or encoded.⁷ Traditionally, our imaginations have been tickled by how AI will compliment humans in diagnostic reasoning to achieve the best possible patient care. But have we fully envisioned how humans will define AI on this journey?⁸

Yes, our professional identities will evolve as some cognitive tasks are “offloaded.” True, AI will outgrow and outperform our analytical capabilities. That said, our ability to search through information deserts is a uniquely human trait that will prove to be a “superpower” for calibrating AI for generations to come.

In summary, human cognition has evolved over thousands of years to understand socially encoded information including nonverbal cues and subtle nuances in speech, tone, or behavior. These elements are important in accurately defining medical problems, in which the patient’s history, perceptions, and environment can provide essential insights. However, as Representation of the problem As it matures, and more information becomes available, the role of AI becomes prominent. AI’s ability to process vast datasets and provide statistical analyzes enables clinicians to identify less intuitive patterns or be sensitive to underlying disease rates or anomalies that may elude human perception. Simultaneously, real-world cognition of humans working with AI’s analytical capabilities will be a powerful engine for diagnostic reasoning.

Now ask yourself… 30 years from now what will be the rate limiting step in eliminating diagnostic errors or mistakes? I’m betting on humanity. Why? Because human problems (especially the unspoken ones) require that humans be helped.

references

1. Jussupow E, Spohrer K, Heinzl A. Identity threats as a cause of resistance to artificial intelligence: a survey study with medical students and professionals. JMIR Form Res. 2022; 6(3).
2. Bergerum C, Petersson C, Thor J, Wolmesjö M. ‘We are data rich but information poor’: how do patient-reported measures stimulate patient involvement in quality improvement in Swedish hospital departments? BMJ Open Qual. 2022; 11(3).
3. Cooke, G. A is for aphorism: Is it true that “a careful history will lead to a diagnosis 80% of the time”? Australian Family Physician. 2012; 41(7).
4. Heyes C, Catmur C. What happened to mirror neurons? Prospect Psychol Sci. 2022; 17(1):153-168.
5. Kirsh D. Problem solving and situated cognition. The Cambridge Handbook of Situated Cognition. 2009; 264-306.
6. Krishna R, Donsuk L, Fei-Fei L, Bernstein, MS. Socially situated artificial intelligence enables learning from human interaction. PNAS. 2022; 119(39).
7. Merkebu J, Battistone M, McMains K, McOwen K, Witkop C, Konopasky A, Torre D, Holmboe E, Durning SJ. Situation: A family of social cognitive theories for understanding clinical reasoning and diagnostic error. Diagnosis (Brill). 2020; 7(3):169-176.
8. Rajkumar A, Dhaliwal G. Improving diagnostic reasoning to improve patient safety. Perm J. 2011; 15(3):68-73.

---

You can find more First10EM guest posts here.