The value of experts

Experts are useful but they work best in a narrow set of circumstances. Where do they work and where don’t they?

Let me illustrate with a 2X2 matrix adapted from the book Think Twice Consider:

Range of Outcomes along the x-axis (narrow to wide)
Decision along the y-axis (rules-based to probabilistic)

👉A rules-based decision with a narrow range of outcomes could be a simple medical diagnosis based on an X-ray or a biomarker. An expert has domain knowledge, but a machine, fed enough good data, can predict the outcome more accurately than any expert. Machine is best.

👉A rules-based decision with a wide range of outcomes is harder to get right. Chess offers several possible board positions and determining which one is best requires expertise (= the ability to retrieve from stored patterns and make a match). Still, enough data and processing power got Deep Blue over the human line. Expert + Machines are best.

As information goes missing and complexity of the system being analyzed increases, rule-based answers become a luxury. Both man and machine need help.

👉A probabilistic solution with a narrow range of outcomes means that several answers are possible and all can be identified. Imagine formulating company strategy for the next year. Your variables could be last year’s business performance, leading market trends, competitors, customer feedback, funds available, et cetera. Your game plan is based on your unique knowledge of the company. This directs your intuition. Computers can boost your predictive accuracy with data. Expert + Machines work best.

👉A probabilistic solution with a wide range of outcomes means that several answers are possible but not all can be identified. No one can predict stock market movements, geopolitical relationships, or supermarket sales. No one person’s a domain expert. These are complex systems where consequences of component interactions are impossible to predict. Here, wisdom of the crowd rules.

Why do non-expert collectives tend to do better than the best individual expert? Because in a group individual error is compensated for by prediction diversity. This works only if there’s diversity in the crowd.

So what can we do to correct our over-reliance on experts and under-reliance on computers and crowds?

✔Properly classify the problem — don’t have a default approach for all problems
✔Adopt technology — use machine power where possible
✔Seek diversity — reduce blind spots in individual-led decision-making

We still need experts: to build systems that will help them decide, to troubleshoot, and to work with people. And experts are good at eliminating bad options and often that’s good enough.

Most tellingly, experts offer psychological safety. We feel more at ease with a high-paid professional with years under their belt than with a nameless and faceless computer or collective. We’re human after all.