Noisy people

(featured image: Mysid/Wikimedia)

Our decision-making is biased, but an even bigger limitation is that it is noisy. So what?

The loudest and longest standing criticism of neoclassical economics is that it assumes us meatbags are rational, self-interested, utility maximizing individuals, while in reality we’re subject to a raft of biases and fallacies. What’s more, the deviations from rational behaviour that characterizes us would seem to be systematic. The title of one of the books by Dan Ariely, Predictably Irrational captures that viewpoint well.

But it appears there is more to it. (There is always more to it. Always.)

At a conference on the Economics of Artificial Intelligence a few weeks ago, behavioural economist par excellence Daniel Kahneman made a remark that has been going round my head for several days now. He observes that “people are very noisy: you show them the same stimulus twice, they don’t give you the same response twice.” He goes on to state that the main limitation on human performance is not bias – it is just noise.

Artificial intelligence (AI) has been proposed as a way to overcome our irrational judgements. AI doesn’t have problems with self-control, can’t be fooled with astute framing, and are not afflicted by confirmation or optimism bias. Unfortunately, both the algorithms for AI (which are ultimately written by humans) and the learning regime for an AI can introduce prejudices. In 2015 one of Google’s AI algorithms apparently tagged two black people as gorillas. The firm itself explains how this kind of thing can happen in this short video. But it is by no means out of the woods yet. Joanna Bryson, a computer scientist at Bath University in the UK found that Google Translate converts gender-neutral pronouns from other languages to “he” when referring to a doctor, and “she” when referring to a nurse.


“I’m sorry, Dave. I’m afraid I am less noisy than you.” – source

But if it is true, as Kahneman says, that the larger reason of our suboptimal decision-making is noise, and not irrationality, prejudice and bias, then AI certainly would have the edge. An algorithm, when given the same stimuli, will always respond in the same way.

That raises two questions: how noisy are we really, and is our noisiness really a problem?

How noisy are we really?

We often do seem to react differently when we are confronted with the same stimulus – or apparently the same stimulus. In laboratory conditions it is possible to pretty much control ‘all else’ and make sure it is ‘equal’, and so present identical stimuli. But what about real life?

In an insightful Behavioral Scientist article, Jason Collins points out how easy it is to consider someone’s behaviour as irrational if we don’t know what their motives and objectives are. Might we not making the same mistake when we label behaviour as noisy? Perhaps the variability in decision making that we observe actually conceals the complexity of influences that we have incorrectly simplified away.

If you were to track what a person has for breakfast every morning, their choices might appear noisy. Some days they have porridge, other days there’s a fried egg, muesli or fruit on the menu. It might look unpredictable (although for sure we can predict that they’re not likely to suddenly opt for grass or a tin of petfood, let alone paperclips or wood shavings). But can we truly say that the same stimulus (getting up and feeling hungry) produces different responses? Even when we ignore ad hoc preferences, they may have run out of one particular option, they may have just heard somebody on the news warning people of eating too much egg, or they may be short for time.  If we don’t know the other elements which may consciously or unconsciously influence the breakfast choice, we cannot know for sure how much of it is due to noise.

And if this is the case for something relatively inconsequential as breakfast, it is probably also true for more momentous judgements. We take into account subtle cues that may not be apparent to an observer – that may even not be apparent to ourselves. Sometimes these cues might lead to poorer decisions, but sometimes they might add beneficial nuance, or indeed a crucial insight that changes everything. That might be a challenge for AI: shutting out all the noise is easy, but if we don’t know which subtle influences enhance our decision making, they risk being shut out as well.

Is it a problem?

Still, there is probably a good deal of real noise in the process. But is such noisiness a bug, or is it a feature?

People pay large amounts of money – multiple times the price of a recording – to see artists play their music live. Why? Because of the noise. No, not (only) because of the elevated sound levels at a concert, but because of the numerous small, surprising deviations from our expectations. The specific live interaction between musicians (and indeed between them and the audience) here and now means that no two instances are the same. That is noise.

Imagine the artists played a 100% identical performance every time – every single note in the saxophonist’s solo a perfect copy of what was played yesterday or the day before, every vibrato in the soprano’s high note exactly the same, time after time. How boring would that be?

symphony orchestra

Lots of noise, and never ever the same performance twice – photo: Quincena Musical

Or imagine your favourite dish in your favourite restaurant was precisely the same very time – identical seasoning, the same number of carrots arranged exactly the same on the plate. You might as well have a microwave meal.

There are certainly situations where noise can be detrimental, and where precision and consistency are to be preferred. Autopilots do a great job in keeping airliners from crashing into each other, because their decisions are not noisy like those of pilots might. Rigid procedures in operating theatres ensure no instruments are left behind in patients after they’ve been stitched up – a problem that could be caused by the surgeon’s noisy behaviour.

And there are plenty other cases where noise may well seriously limit human performance. Delegating judgement to systems that are not prone to noise can undoubtedly enhance our wellbeing.

But noise is also what makes us human. When we greet a colleague or a friend, we don’t mechanistically use exactly the same words, at the same pitch every time – there is variation, there is noise in the process. We each have our own, unique noisy signature, just like famous or less famous musicians and chefs.

Let us celebrate that noisiness. And even more importantly, for goodness’ sake, let us not get rid of all the noise.

About koenfucius

Wisdom or koenfusion? Maybe the difference is not that big.
This entry was posted in Behavioural economics, Cognitive biases and fallacies, Economics and tagged , , . Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s