The Next Frontier for Naturalistic Decision Making: A Call to Action

John Schmitt suggests a new frontier for applying NDM principles and provides a landscape of current NDM tools that may be relevant to this emerging area. We are actively seeking input from the community about tools for addressing wicked problems!

In a previous essay, I suggested that breakthrough solutions are usually the result of being able to turn the problem “on its head”—that is, to look at the problem in a fundamentally different way that points to a new way of solving it.  This is exactly what the naturalistic decision making (NDM) movement did starting in the 1980s in studying human decision making.  The research up until that point had studied decision making in laboratory settings, using undergraduate test subjects performing menial decision tasks under controlled conditions.  The result was the Rational Choice Theory (RCT), which posited that decision making consisted of generating and comparing multiple courses of action (COAs) according to some predetermined set of evaluation criteria to arrive at the optimal solution.  RCT became the accepted theory of decision making.  The example par excellence of RCT is the Military Decision Making Process (MDMP), a highly regimented staff planning procedure of steps, sub-steps, and sub-sub-steps in which the required output of each step is the necessary input for the next step.  Some variation of the MDMP is taught at war colleges around the world.  

The Rise of Naturalistic Decision Making

But starting in the 1980s, a group of skeptics started looking at decision making in a very different way.  Instead of studying novice decision makers in highly controlled settings, they decided to study practitioners in the field.  They embraced uncertainty, time pressure, changing conditions, noise, shifting and competing interests and goals, and potential danger and other stresses as central to the problem of making decisions in the world.  They studied firefighters, police officers, military commanders, emergency room nurses, and cockpit crews.  They later went on to study petroleum plant operators, nuclear power plant crisis managers, child protective service workers, stock traders, hedge fund managers, and technology developers.  In the process, they developed a radically different understanding of how humans make decisions.  NDM was born.  The adherents of RCT have been fighting a rear-guard action ever since to protect what remains of their eviscerated theory.  Although there is little remaining doubt in the research community that NDM is a more accurate description of human decision making, RCT continues to be taught in many places as the way decisions ought to be made, even if they rarely are.

Gary Klein’s Recognition-Primed Decision (RPD) model is the best known of the NDM models.  According to RPD, a practitioner recognizes unfolding patterns from past experience and uses that experience to generate a COA.  Again relying on experience, the practitioner runs a mental simulation of the COA to see how it might work in the present situation.  If it is workable, she executes it.  Research shows that experienced practitioners generally conceive a satisfactory first COA, so there often is no need to compare multiple COAs.  However, if the mental simulation reveals that the first COA is not satisfactory, the practitioner will consider another.  But here’s the important distinction:  she will not compare multiple COAs against each other; she will run each COA against the situation until she finds the first one that she believes will work.

Recognition-Primed Decision Model

Most of the research that led Klein to the RPD examined decisions that were challenging and complex but were generally constrained in time and space.  For example: 

  • A fire chief sizing up an apartment fire.
  • A police officer deciding how to handle a domestic violence call.
  • A military commander reacting to an enemy tank battalion suddenly appearing on his flank.
  • An airline pilot who has just lost power in both engines from a bird strike shortly after takeoff.
  • An emergency room nurse whose overdose patient’s respiration has suddenly dropped to zero.
  • A social worker determining whether a child’s substance-addicted mother is currently a safe caregiver.

These are challenging, often life-or-death situations that must be solved with a comprehensive solution at a particular time and place:

  • “Company 12:  External attack of the exterior stairwell with the deck gun [i.e., water cannon] for 90 seconds, followed by internal attack with the one-and-three-quarter-inch hose.  Company 13:  Search and rescue.”
  • Wait for backup, then tactical approach using cover.  Listen for sounds of struggle and try to locate the subjects inside the house.  Call the house phone to see who answers.
  • “Artillery:  Immediate suppression on the enemy tanks. Alpha Company:  Return fire, fix the enemy in place.  Bravo Company, reinforced by Charlie:  Envelop left via Checkpoint 23 and catch the enemy in the flank.”
  • Make an emergency landing on the Hudson River.
  • Signal a Code Blue; insert a nasal trumpet to open an air passage.
  • Find a reliable relative known to the child who will take temporary custody while you come up with a treatment plan for the mother.

Wicked Problems: A New Frontier for NDM

One of the fallbacks of the RCT supporters is to concede that RPD may be superior for these kinds of in-the-moment, seat-of-the-pants decisions. But they insist that RCT is still superior for large-scale, deliberative types of problems.  So what about large-scale societal problems that cut across domains and disciplines for which no one actor has the necessary authority, responsibility, or resources?  Problems that are truly novel and so interactively complex that no expert has the requisite experience to simply recognize what to do?  The kinds of problems that you must reason your way through because their scale and complexity swamp human intuition?  The kinds of problems that cannot be comprehensively solved but must be continuously managed over time?  The kinds of problems that generally involve entire staffs rather than single decision makers?  (A retired three-star general I know is fond of saying, “Never confuse staff activity with actual progress.”)  Rittel and Webber famously called these wicked problems.  Read their ground-breaking paper here.  Wicked problems are all around us.  Examples include gun violence in American cities, income disparity, childhood obesity, pandemic response, financial crises, racial injustice, access to quality education, and terrorism, to name a few.

Where RPD provides an experience-based method for solving bounded problems, wicked problems instead require a method for gaining insight in the face of extreme scale and complexity.  While RPD may not address these kinds of societal problems, that does not mean that RCT is the answer either.  The evidence is pretty clear that it is not.  These kinds of problems must be reasoned through, yes, but I believe there are better deliberative methods than RCT—naturalistic methods that are more compatible with the way humans actually think.  NDM can and should address wicked problems.  In fact, I suggest this may be the next frontier for Naturalistic Decision Making.  There is no equivalent to RPD for wicked problems, but there should be.

Individual tools already exist:

  • Cognitive task analysis to better understand the kinds of decisions that must be made in a given domain and what makes them difficult.
  • PreMortem exercises to anticipate the potential failure points in plans.
  • Design thinking through the use of the dialectical method to integrate multiple perspectives.
  • Systems thinking as a disciplined approach to problem formulation (including Peter Checkland’s Soft Systems Methodology as an example).
  • Matrix games to understand the complex dynamics of situations involving multiple stakeholder interests.
  • Concept mapping to organize knowledge of complex subjects. 
  • Abductive reasoning methods, such as the use of metaphors and analogies, to generate inferences from limited data.

None of these tools is the answer, but I suggest that they could be incorporated into a broad naturalistic framework for better addressing the problems that plague us as a society.  It is an important and worthy endeavor.

What’s next?

If you have ideas and would like to join us in considering how existing or new NDM tools could help solve wicked problems, we would love to hear them!