Noise #3: Its Constituents, and Ways to Minimize it

What are the common sources of noise in decision making? Well, people have selective attention, and selective recall of the information, write the authors of Noise. Information is rarely coherent, it is often conflicting, and so people pick and drop parts. Therefore, what goes into the decision varies across individuals.

 

People are different. Some are strict, others lenient; some are optimistic, others pessimistic. You get the idea. If that sounds like bias, you’re right – it is actually both. It’s noise too because different people have different biases. This is called “level noise” – different people have different base levels.

 

People makes exceptions to their own rules. They also react differently if the same problem is presented in a different context. All these variations within the same person are called “pattern noise”.

 

Within pattern noise, there is a subcategory called the “occasion noise” – the mood of the person can produce different decisions. The “when” matters more than the “what” at such times.

 

The authors then switch to finding steps to reduce noise. We’ll look at a few of them in this blog.

 

More heads are better than one. That’s true, but only under certain conditions: (1) people should have different specialties and/or preferences, and (2) they should hear each other’s views only after they have put theirs down on paper first. But all too often, these conditions are ignored, in which case even if more people were involved in the decision, they aren’t independent of each other and thus don’t help reduce the noise:

“Independence is a prerequisite for the wisdom of the crowds.”

 

Unintuitively, a formula that is based on an individual’s decisions in the past, even a crude one that is “almost a caricature”, still “outpredicted the professional on which it was based”!

“The ersatz was better than the original product.”

This is hard to digest. After all, most of us consider judgment to be “complex, rich, and interesting precisely because it does not fit simple rules”. The simplified model, by definition, cannot add anything, it can only “subtract and simplify”. What does this say about those intangible, instinctive aspects to judgment?

“You may believe that you are subtler, more insightful, and more nuanced than the linear caricature of your thinking. But in fact, you are mostly noisier.”

 

Another unintuitive thing studies have found is that adding more rules doesn’t improve the quality of decisions. In fact, more rules usually make things worse. How can that be? It turns out most rules aren’t independent of each other; they are correlated with each other. So when one rule is wrong, the lack of independence means the other rules just compounds the error. But that doesn’t fit with how AI/machine-learning algorithms work. They use a lot more data points to decide, not less. What’s going on? They work better because they are able to notice relationships that humans fail to see.

 

If all this is true, why don’t people switch to rule-based decision making, whether by following it themselves or handing control to AI? Partly, because it is very unsatisfying to leave everything to cold, impersonal rules. But also, the algorithm isn’t perfect; it’s just better than us, a distinction we fail to remember:

“(People) stop trusting it (algorithm) as soon as they see that it makes mistakes.”

 

Another reason lies in the scales used by different people. The lack of standardized scales means everyone assigns different values, often wildly different. A solution might therefore be to show similar cases as “anchors” – use that for reference, as a comparative item, and then decide.

 

None of the above is intuitive or easy, which is why noise is so pervasive.