Clouds and Clocks
Human societies are fundamentally more complex and more difficult to predict than orderly physical systems. But new research provides evidence that long-range forecasting is possible in certain circumstances.
Karl Popper made a distinction between two archetypal types of physical systems: clouds and clocks.1 Clocks are regular, orderly, and predictable. The solar system is a clock because its exact state follows “like clockwork” from its state at the previous time. Clouds are irregular, disorder, and unpredictable. Popper gives the example of a cloud of gnats that can shift in seemingly random ways.
This distinction isn’t firm, but a matter of degree. It may actually be—as I think—that all clouds are fundamentally just complicated clocks in the sense that the state of even the most irregular system follows deterministically from its previous state. It may also be—as Popper argued—that all clocks are clouds in the sense that even the most precise clockwork has imperfections that can cause it to behave in seemingly random ways. The key distinction between clouds and clocks is that clouds are more complex and don’t obey simple, easy-to-generalize rules.
Human societies are not clocks. History doesn’t follow a predictable course like the planets in orbit around the sun; people are not billiard balls governed by simple mechanical rules. We are more like Popper’s cloud of gnats, each adapting in idiosyncratic ways to the movements of the others. Patterns of behavior can endure for while before some minor perturbation—that triggers a bank run or leads to the collapse of a government—tips them into a new equilibrium state. One reason the social sciences don’t produce the kind of reliable scientific laws that characterize the “hard” sciences—and why forecasting human events is more art than science—is that human social behavior doesn’t necessarily obey any limited set of simple rules.2
Nevertheless, cloud-like systems are not completely chaotic. Our ability to function in society depends on our having a good idea—at least in normal times—what others are likely to be doing at a given place and time. The success of probabilistic forecasters in predicting political events a couple of years in advance shows that even if our behavior defies easy generalization we can discern some patterns in the cloud of human behavior. Some broad phenomena—like growth in the global population—can be predicted even further in advance. There’s no real doubt that we should make our plans on the basis of what we expect to happen in the future. The question is where the line between useful prediction and pure speculation is.
Philip Tetlock recently led a new paper that suggests forecasters can make at least some meaningful judgments about risk 25 years in advance.3 The authors draw on a handful of forecasts that were originally made in 1988 and 1997 as part of the research that formed the basis for Tetlock's Expert Political Judgment.4 The authors found that people with subject matter knowledge of nuclear proliferation issues—even though they weren't necessarily the most skilled forecasters—were able to produce more accurate and more coherent long-range forecasts than both people without subject matter expertise and simple base-rate extrapolation methods. The authors acknowledge that their dataset is too limited to draw any firm conclusions, but the paper provides at least tentative proof of the concept of long-range forecasting. Importantly, however, they also found that subject matter experts weren't able to produce more accurate long-range forecasts about changes to national borders. In other words, long-range forecasting may be possible in some cases, but it may be hard to know exactly what those are.
The truth is that we shouldn’t have much confidence in long-range predictions about anything but the most precise clocks. Even in areas we’ve success forecasting before there’s some unknown chance the future won’t conform to the same patterns as the past. The cloudiness of human events means that we need to plan for a wide range of possible outcomes. It doesn’t mean we shouldn’t make long-range forecasts, but simply that we shouldn’t stake too much on those forecasts being right. As uncertain as human society is, the present does provide valuable information about the future, if we can use it judiciously.
Sorry that I had to take another long hiatus. I spent most of the last three weeks dealing with a difficult family matter, but I plan to get back to writing and forecasting regularly again. Because I believe artificial intelligence is likely to continue to develop rapidly—that’s one of the things I want to write about soon—I added my name to the Future of Life Institute’s open letter calling for a pause in the training of powerful AI systems until we implement better AI safety protocols. I previously wrote a paper with Seth Baum on the governance of AI risk titled Collective Action on Artificial Intelligence. Finally, Good Judgment posted a fascinating interview with former military analyst and superforecaster Alice Dorman, and Dawna Coutant, another accomplished superforecaster, came out with a new mystery novel, Evil Alice and the Borzoi. As always, if you enjoyed this, please help me out by sharing it!
It seems possible to pretty confidently predict that our civilization will collapse at some point, given that (to my knowledge) every human civilization ever created has done so. The bottom line seems pretty easy, but the why, how, when impossible.