This article is taken from the March 2022 issue of The Critic. To get the full magazine why not subscribe? Right now we’re offering five issue for just £10.

---

As anyone who has spoken in front of a mic knows, noise can amplify itself out of control. The right thing to do is to quickly stop the feedback signal, turn down the volume, or simply cut the cord.

When in April 2020, Francis Collins, the director of the world’s largest biomedical agency, the National Institutes of Health (NIH), sent a letter to Antoni Fauci demanding that an anti-lockdown proposal from “three fringe epidemiologists” be met with “a quick and devastating published take-down of its premises” on the grounds that it “seems to be getting a lot of attention”, presumably he had noise elimination on his mind.

Daniel Kahneman and his co-authors Olivier Sibony and Cass Sunstein, however, won’t have it. In their new book, Noise — A Flaw in Human Judgment (2021), noise is no longer simply an unwanted signal. Rather, noise is defined as variability in a signal.

Imagine two judges who, given the same case and the exact same information, choose two different verdicts. If one judge decides to hand out six years in jail to the accused, and another judge two years, it averages out in the long run. But the convicted felon who gets six years in jail, instead of two or four, does not give too much thought to averages.

Similarly, an insurance company where two underwriters facing the same candidate and information choose two different risk profiles, will perform worse than an insurance company where such variability is not present. On average, if one underwriter sets a premium at $9,500, another quotes $16,700 — five times as large as expected by company executives.

Might our institutions and societies be such non-linear, complex systems?

It should be clear that Kahneman and his colleagues do not think all noise is bad. The variability in the examples above is bad because it is happening to judgment within systems. And “in what we call matters of judgment … system noise is always a problem”.

Within markets, diversity thrives. In fact, diversity is what makes markets. Within organisations, however, variability in judgment worsens performance. To this, the authors propose a solution. Individuals, institutions, and societies — in short, systems — must erect a set of decision hygiene algorithms to eliminate faulty measurements of minds and become noise-free. After all, there’s no point in circling noisily around the bullseye when you can hit it squarely on.

It’s obvious that system noise is indeed often a liability. Before the metric revolution, weavers measured yarn by stretching the material from thumb to elbow. Any buyer ordering two elbow-thumb lengths of yarn, received slightly differing amounts of yarn each time and so might prefer that weavers devise a more precise measurement practice.

The only question is if the logic holds generally. In 2006, professor of electrical engineering Bart Kosko published a book simply titled Noise. Kosko showed that many non-linear systems in fact improve when injected with a little noise. The intuition is that such noise injections bump the system out of a local optimum and onto the path of a global optimum.

Might our institutions and societies be such non-linear, complex systems? Indeed, when things get complex, we often don’t know where the bullseye is, nor how to find it. Until we do, a certain amount of noise might enhance exploration and ultimately improve the search. To Kosko, noise elimination is not the end-all-answer to a clear signal. Rather, we should learn to “see more noise patterns as signals whether or not we like those signals”.

What was the signal in the noise when, in a Bloomberg column from February 2020, Cass Sunstein wrote that “many people are far more scared than they have reason to be”, and a month later he wrote that “this time we can’t be too careful”? Was it a signal that this pandemic is a complex, non-predictable system? Was Sunstein noisily circling around an already known optimal strategy, or did he in fact not have any clue where the true bullseye of covid strategies was?

Scott Page, a professor of Complexity at the University of Michigan, has shown that a diverse population of individuals, each with very weak, noisy heuristics, will outperform a single individual with strong heuristics, and that’s especially true in complex environments. “Diversity trumps ability,” is Page’s way of saying it.

They could be the people who lead us out of the valley of the inadequate equilibrium

Consider the following thought experiment from Scott Alexander, author of the popular blog Astral Codex Ten. Imagine a number of fishermen fishing in a lake. Eventually, overfishing and pollution dooms the lake and the fishermen lose their income. But a contractor develops a water filter. Given the cost of the filter, the fishermen now earn a little less than before. Peter, a regular dissenter, thinks that if everyone uses the filter, he can simply do without it and earn a bit more than the others. Peter is smart and uninstalls the filter.

Jealous and angry, the others follow suit. At some point, enough of them will have uninstalled their filter with the result that Peter earns less than if he had not uninstalled his filter in the first place. To Peter’s fortune, however, the fishermen sign a pact demanding that everyone reinstalls the filter. Once again, Peter earns more than all the others. The others cannot uninstall the filter without breaking the pact. Peter, similarly, has no interest in signing the pact as long as everyone else does.

On the face of it, Peter seems selfish, noisy, and irrational. But perhaps the fishermen would never have come up with the life-saving contract were it not for Peter.

To take the logic to its extreme, underwriters who, given the same candidate, choose different risk profiles, or countries facing the same pandemic situation which choose different alleviation strategies, might not be bad actors with a cognitive measurement fault. Rather, they could be the people who lead us out of the valley of the inadequate equilibrium.

The question is whether systems such as organisations are, in fact, markets. For instance, the email sent from Francis Collins need not represent the belief that the NIH’s strategy was on target. Instead, it could simply mean that for any strategy to yield interpretable results within a system, it needs to be followed consistently for at least a period of time. But even the NIH changes its strategy from time to time, and when it does so, it shifts to a market logic.

If a pro-mask government wants to explore the efficacy of masks, should they aim at 100 per cent compliance in mask wearing or wait for the results from a non-mask-wearing country to surface? Just how noisy should a hypothetical world government be when assigning mask-strategies to each country to optimise mask-knowledge? If that is the kind of discussion that the authors foster, then their book will succeed in shifting decision-making closer to the target.