Bayes Theorem: We are Prediction Machines

Uncertainty

Few of us thrive on uncertainty. It tends to make us anxious, nervous, and generally feel out of place. Despite this, we swim in a sea of risk and uncertainty from one moment to the next, and our brains have adapted in a way to help us survive in a world full of risky and uncertain outcomes. We strive for knowledge, for certainty. To assist in this effort, our brains make up compensations to aid in perception and decision-making, by using mental models and shortcuts (or heuristics, as they’re referred to).

There are two ways to think about unknowns, and sometimes we confound the two. The first relates to known events that have an uncertain likelihood, or probability of occurrence. We’ll refer to this as risk, and an example would be a volcano eruption. Volcanoes do not work in predictable ways, but we do know they erupt at some point. These eruptions do not follow predictable patterns, so we really don’t know the exact likelihood of it occurring in a month, year, decades, etc. The second way relates to unknown events where the likelihood of occurrence can’t be assessed, and we’ll refer to this as uncertainty. For example...and bear with me on this one...you might wake up to a snowy day where your car is blocked in by the work of a snowplow. In this instance, and especially if you are new to driving or coming from a different climate, you might never have even known of that actual event, so therefore you could not possibly assess the likelihood of it happening.?

We now know volcanoes erupt at some point, but what about the very first people to ever come across one? They had no awareness of the event, the likelihood of it occurring, or the consequences that follow. This is uncertainty.?

With that said, uncertainty could very well be something where we don’t yet understand the risk or likelihood of occurrence, but with additional knowledge, we could. Moreover, in uncertain circumstances, we may draw on past similar experiences as a guide to assessing the risk. Going back to the car example, you would be more aware of that risk after spending a few winters in that location. So, the next time the forecast calls for heavy snow overnight, you might choose to position your car in a different spot. Another example. What’s the likelihood (risk) of surviving on Mars? Right now, it’s uncertain and we can’t assess the probability. But with more knowledge and experience in traveling to the planet, navigating it, and living on it, we can make a better assessment of the risks. In the meantime, we could use the experience of traveling to the Moon as a guide.?

Predictions

But none of that stops us from making predictions – predictions about the likelihood of outcomes. Consider that we are constantly making predictions. This is more profound a statement than it may seem at first, but through our senses and our brain’s decision-making abilities, we navigate our day making constant predictions and assessments to guide us in all our activities. Our brains are prediction machines. Most of these actions occur without our conscious awareness.?

Just getting out of bed or reaching for the morning coffee mug requires considerable predictive capabilities, which we take for granted. Our brains do a good job of making predictions to accomplish these tasks on our behalf, sometimes compensating for the limitations in our senses or memories. The predictions we make are steeped in either risk or uncertainty (but not both). In the case of risk, there is a range of possible outcomes, while in the case of uncertainty, we take actions despite not knowing the risk of taking those actions. We don’t have any ‘priors’ in the Bayesian sense. In other words, we don’t have a sense of the risk in the action we’re about to take.

Bayes Theorem

Bayes' theorem essentially concerns itself with how to reassess the likelihood (risk) of an outcome given our prior knowledge when presented with additional information. It is posited that the brain works in a similar way. More recent neuroscience considers that the mind works based on what is rereferred to as predictive coding. We develop what are perceptual models of the world for various circumstances, environments, and situations. We then seek to reduce prediction error when our expectations don’t meet reality. We do this by calling on our higher-level perceptions and making a ‘best guess of what we’re experiencing. It brings up the size/weight illusion. We mistake a large metallic-looking object as being heavier than it is and utilize much more force than needed to pick it up from the shelf, not realizing it is hollow and made of plastic. That’s a prediction error at work.?

So, what’s wrong with the modern manner in which we attempt to make decisions? Well to start, we suffer from information and sensory overload. We’re straining our attention and overwhelming our senses with noise. We're simply going the wrong way when it comes to improving our decision-making. Our critical thinking capabilities are eroding. We’re in front of ever more information but making less and less sense of it.?

Ultimately, better decision-making comes down to making better predictions, under conditions of risk or uncertainty. As for making better predictions, that comes down to closing the gap between our expectations and outcomes (reducing prediction error). Our expectations, to a large extent, are a function of the quality of information (input) we’re using to make the decision.?

What this comes down to is that we need tools and approaches to assist us in reducing prediction errors so we can make better decisions. We may not need to implant devices in our brains to foster better decision-making; improving the quality of the input that drives our expectations may well suffice. Effective crowdsourced wisdom is one such tool, and it would also help for us to enhance our approaches to assessing the inputs we receive, via our mental models and the inferences we make from the information we receive. Let’s muse about that next time...

Want to benefit from the quality insight that can drive better decision-making? Take a look at macrovo.com