Recently, Jordan Greenhall drew a distinction between complex and complicated. The distinction matters. With complex problems, we need to lower our expectations about our ability to arrive at fully satisfactory solutions.
Greenhall wrote,
a complicated system is defined by a finite and bounded (unchanging) set of possible dynamic states, while a complex system is defined by an infinite and unbounded (growing, evolving) set of possible dynamic states.
. . .In the case of complication, the optimal choice is to become an “expert”. That is, to grasp the whole of the system such that one can make precise predictions about how it will respond to inputs.
In the case of complexity, the optimal choice goes in a very different direction: to become responsive. Because complex systems change, and by definition change unexpectedly, the only “best” approach is to seek to maximize your agentic capacity in general. In complication, one specializes. In complexity, one becomes more generally capable.
Greenhall offers this illustration: the behavior of a simple bumblebee is complex, because it has response mechanisms that we do not fully understand; but a Boeing 747 is merely complicated, because its behavioral range is limited by a design and structure that we understand and can model. He does note that there are unusual situations in which the Boeing 747 could exhibit complex behavior.
Greenhall’s essay concerns the challenge of dealing with social media. The apps deal with the complex problem of social interaction as if it were merely complicated. They train us to play the complicated game of attracting attention and accumulating “likes,” as opposed to displaying the complexity of our true creative expression.
Artificial Intelligence, board games, and the ah-pe-tor
The distinction between complicated and complex can help us to appreciate the challenges of artificial intelligence. AI can handle complication better than humans. But it is not so well suited to complexity. Hence, an AI may out-perform humans at complicated board games, such as Go; but this does not guarantee that AI will be successful at complex tasks, such as speech recognition.
Our grandson is just learning to talk. When he makes the sound “bpah-bpah,” he could be asking for his milk bottle, his pacifier, or for me. We can tell what he means by context. For example, if he is standing at the door of the refrigerator, then he wants his milk bottle. How would an AI interpret a tape of him saying “bpah-bpah”?
When we approach an elevator, our grandson can communicate to us that he wants to push the button. But if you played a tape of him saying “ah-pe-tor,” a speech-recognition program would be unlikely to recognize this as “elevator.”
Economics
When I was a graduate student in economics in the late 1970s, we were trained as if the economy is complicated, but not complex. We were told that if we learned enough mathematics and statistics and applied these tools, then eventually we could predict and control economic outcomes.
In fact, economic behavior is complex. There are too many causal factors, feedback loops, non-linear effects, and unprecedented phenomena involved to enable economists to control the economy precisely and reliably. Often, the best mathematical models are not even useful, as was dramatically shown a decade ago by the failure to anticipate the financial crisis and its aftermath.
In fact, complexity is a challenge in all of what we unfortunately call “the social sciences.” The very term social science gives the impression that human behavior is merely complicated, so that social outcomes can be predicted and managed by experts. With complex systems, as Greenhall points out, we are often better off with adaptive processes than expert decision-making.
Climate Complexity
Climate scientists use computer models, because the problems with which they deal are complicated. But there are multiple models, and they do not agree with one another. That tells me that the climate, like the economy, is complex. There are too many causal factors, feedback loops, non-linear processes, and unprecedented phenomena involved to enable precise and reliable prediction and control.
In contrast, landing a spacecraft on the moon is merely complicated. It is a very difficult problem, but we can arrive at a determinate solution.
But suppose we were trying to land on the moon and all we had were a collection of models that disagreed about whether a given trajectory would reach the moon, fall short, or slip past it. Even if the “average” of the models said that we would hit our target, I do not think that we would risk sending a human in a spacecraft to the moon on that basis.
For climate policy, we are offered a variety of models to choose from. In broad terms they agree, but in broad terms in 2007 all of the mainstream economic models agreed that there would be no major recession.
Given the complexity of the process, it seems inappropriate to pound one’s fist on the table and insist that the science is settled. Instead, the most we can claim is that the best guides for policy that we have are consensus forecasts for climate change and consensus estimates of the effect of human activity on the climate.
Treating the climate as complex means that we probably should not rely on any single forecast or any one policy lever. Above all, we need contingency plans. What if ice sheets start to melt very rapidly? What if we cannot reverse climate change before low-lying areas begin to suffer flooding?
We also need to deal with policy uncertainty. What if reducing manufacturing locally leads to more coal-fueled manufacturing somewhere else? What if, when the entire production cycle is considered, biofuels on net result in more atmospheric carbon dioxide, not less, than the fuels for which they substitute?
Just to be clear, I do not claim to have better answers. All I am suggesting is that we use a complexity framework to look at the problem.
Setting expectations
Many complicated problems have been solved by human beings and by our powerful computing tools. But I think this creates the expectation that we can solve complex problems as well. By understanding the difference between complication and complexity, we can take a more realistic view.
This is especially the case in biology and neuroscience. If cancer were merely complicated, then by now we would have won the “war on cancer.” If genetics were merely complicated, then the extravagant hopes that were raised during the race to complete the human genome would have been realized. If the brain were merely complicated, then we really could model the process of thought.
Instead, we face complexity: in speech recognition; in economics and other social sciences; in climate change; and in biology and neuroscience. In these complex fields, we should be humble about how much we know and cautious about predicting our ability to attain full understanding.