Why Analytic Decision Theory Needs Rethinking

This website is now about 18 years old. I’ve been writing about computers, photography, books and investment. But in the background is a big project I refer to as ODB. “On Deciding Better.”

ODB started with my introduction to analytic decision theory when I left academic medicine to work in a biotech. I had the privilege of working with a few visionaries in drug development, who advocated a more data driven approach to making decisions about how to turn a chemical into a commercial product that improved the lives of patients. It’s been clear for a long time that the pharmaceutical industry is wasteful, best characterized by a combination of wishful thinking and poor decision making.

My biggest achievement in implementing these ideas was in developing a new sedative based on propofol collaborating with experts in state of the art techniques of mathematical modeling and clinical trial simulation. I left the company before the final approval, watching from afar the introduction of the drug to the marketplace and then its withdrawal as a commercial failure.

As I moved on to other positions in the industry, I found very little enthusiasm for wide adoption of these approaches. There were pockets of use, but more broadly no one found them useful enough to pay for their use. Trials were designed using the same flawed methodology and decisions were made as they had always been. By argument and gut feeling.

I came to the realization over time there there was a disconnect between the principles of analytic decision theory and the way people actually made decisions. In the last few years I think I’ve figured out just why analytic decision theory doesn’t work to augment human decision making.

While the basic components of a decision are represented as objective, options, values and probability are all deeply subjective. When I recast them as imagination, emotion and belief, no one would even entertain the idea that this process constituted a rational approach to making a decision.

So get a group of decision makers together and they will see the right answer immediately. Or at least they see the outcome they want the most that they believe is achievable with acceptable risk of things going wrong. It’s a fact of life in organizations with human decision makers.

Fortunately, we’re entering a time where our computational power is great enough that we can augment these intuitions with algorithms powerful enough to capture some of the real world’s complexity and provide ways to support imagination rather than diminish it.

I think the essential rethinking has to how the theory of decision making and Baysian statistics are implemented by brains in our modern cultural context.

Leave a Reply