What really makes us smart is not our ability to pull facts from documents or decipher statistical patterns in arrays of data. It’s our ability to make sense of things, to weave the knowledge we draw from observation and experience, from living, into a rich and fluid understanding of the world that we can then apply to any task or challenge. The Glass Cage: Automation and Us Nicholas Carr
What is human intelligence? It seems to be a miraculous facility that enables man to make complex decisions based on intuition and feelings. We don’t usually consider all the options options, calculate probability of success or really think about whether we really want what we’re going to get? How is it that we are such incredibly sophisticated, rule-based decision making machines?
Is it even possible to get better at decision making or has evolution done all of the work already?
Years ago, I began with the premise that formal decision tools could help make more rational, more successful decisions. I learned the concepts of decision making under conditions of uncertainty from consultants and practitioners in the field of Decision Theory. The basic tools like decision trees and simulation were presented as simple consequences of statistical principles. I worked through a basic textbook, Robert Clemens’ Making Hard Decisions: An Introduction to Decision Analysis and began writing about making decisions here at ODB.
After a few years, I realized that no one outside of the world of management consulting and corporate strategy offices really used these tools. Eventually I came to a fundamental reimagining of decision making based on belief models and my training in Neuroscience. The tools of decision analysis, mathematical models, and simulation seemed to best characterized as tools to augment human imagination and understanding. They were useful fictions to guide thinking, merely simple representations of the complex world. I think it’s definitely possible to improve decision making, but the task is as much improving the mental decision making machinery as it is understanding the mathematical tools.
The Subjective Nature of Probability
I’m not alone in reaching this conclusion. There have always been similar lines of argument, going back to some of the original works in the field of Decision Theory and modern statistics. In Statistical Rethinking, the book I’ worked through to learn Bayesian statistical methods, Richard McElreath provides a wonderful introduction to the Bayesian interpretation of probability. Decision theory is predicated on a specific interpretation of probability. Probability is seen as the subjective likelihood of a particular future event, whether it’s the outcome of a coin flip or the choice of nominee by a political party. These things happen only once, so probability is prediction.
There has been a decades long discussion about whether the probability of heads in a coin flip and the probability of a complex event in the real world can really be the same kind of probability. McElreath presents the formulation of Jimmy Savage, one of the foundational thinkers in the field. Savage proposed that there is a difference between the “small world” of the coin flip which can be accurately reduced to a simple mathematical model and the “large world”, where simple models don’t necessarily hold.
The Brain Is Not a Small World, Artificial Intelligence Is
The brain is complex and unpredictable and so is by Savage’s term, a large world. Because of our ability for symbolic reasoning, it can contain within it small worlds, models of part of the world that are sometimes explicit but most often implicit. We use these internal models for decision making.
The brain can easily imagine the mechanics of the Bernoulli distribution. But the brain can also contain the mental model of neurological disease and the potential effect of a new medicine. Perhaps machine and other algorithmic, mathematical “small world” systems can never match the “large world” human decision making brain. Why?The brain itself is part of the even larger real world, constantly exchanging information with it, something that we have yet to achieve with our machine intelligence.