Complexity

The Devil’s Advocate™ →Go uptown A dancing landscape, yesterday. Complex systems present as “wicked problems”. They are dynamic, unbounded, incomplete, contradictory and constantly changing. They comprise an indefinite set of subcomponents that interact with each other and the environment in unexpected, non-linear ways. They are thus unpredictable, chaotic and “insoluble” — no algorithm can predict how they will behave in all circumstances. Probabilistic models may work passably well most of the time, but the times where statistical models fail may be exactly the times you really wish they didn’t, as Long Term Capital Management would tell you. Complex systems may comprise many other simple, complicated and indeed complex systems, but their interaction with each other will be a whole other thing. So while you may manage the simple and complicated sub-systems effectively with algorithms, checklists, and playbooks — and may manage tthe system on normal times, you remain at risk to “tail events” in abnormal circumstances. You cannot eliminate this risk: accidents in complex systems are inevitable — hence “normal”, in Charles Perrow’s argot. However well you manage a complex system it remains innately unpredictable. In which the curmudgeonly old sod puts the world to rights. Index — Click ᐅ to expand: The Devil’s Advocate Tell me more Sign up for our newsletter — or just get in touch: for ½ a weekly 🍺 you get to consult JC. Ask about it here.

All other things being equal, a bummer — if you have in mind risk — and a boon — if you have in mind reward. A violation of Occam’s razor; a source of confusion, a time-sink, a material contributor to catastrophic normal accidents; a waste — yet in a distributed network of autonomous bodies, an inevitability. The more sophisticated the group of individuals, the greater the rate of complexification.

Complexity as a bummer

• Chernobyl
• Space Shuttle Challenger
• Any financial calamity you care to mention

Complexity as a boon

• T’Internet
• The gradual disintermediation of information from its substrate
• Jacquard loom

Complicated versus complex

Things can be merely complicated without being complex. Complicated problems are naturally difficult, but you can solve them with rules and algorithms. The systems and processes, by which The Man commands and controls employees can manage this kind of complicatedness.

Algorithms, systems and processes don’t work for complex problems, however. Complex problems involve independent, autonomous agents interacting in un-anticipatable ways. No pre-defined rule-set can anticipate the interactions. Black swans, technology disruptions but also interlocking complicated systems (nuclear power plants, space shuttles, air-traffic control systems are complex.

Identify your systems

Is your system simple, complicated or complex?

Simple systems: simple systems are situations where essentially inanimate objects interact with each other in ways that are fully understood. Lego is a simple system. So is a cake recipe, or a bungee jump. The components of a simple system don’t fight back. Simple systems are therefore predictable. They can only go wrong if components fail or you don’t follow instructions. In either case they fail in predictable ways. As such, simple systems are suitable for checklists, recipes etc, where algorithms can overcome the hubris that will surely rain down on the heads of those who treat simple processes as trivial. Disinfecting your instruments before performing heart surgery, for example, is a simple step to take, but not a trivial one.

Complicated systems require interaction with autonomous agents whose specific behaviour is beyond the observer’s control, and might be intended to defeat the observer’s objective, but whose range of behaviour is deterministic, rule-bound and known and can be predicted in advance, and where the observer’s observing behaviour does not itself interfere with the essential equilibrium of the system.

You know you have a complicated system when it cleaves to a comprehensive set of axioms and rules, and thus it is a matter of making sure that the proper models are being used for the situation at hand. Chess and Alpha Go are complicated, but not complex, systems. So are most sports. You can “force-solve” them, at least in theory.

Complicated systems benefit from skilled management and some expertise to operate: a good chess player will do better than a poor one, and clearly a skilled, fit footballer can execute a plan better than a wheezy novice — but in the right hands and given good instructions even a mediocre player can usually manage without catastrophe. While success will be partly a function of user’s skill and expertise, a bad player with a good plan may defeat a skilled player with a bad one.

Given enough processing power, complicated systems are predictable, determinative and calculable. They’re tame, not wicked problems.

Complex systems present as “wicked problems”. They are dynamic, unbounded, incomplete, contradictory and constantly changing. They comprise an indefinite set of subcomponents that interact with each other and the environment in unexpected, non-linear ways. They are thus unpredictable, chaotic and “insoluble” — no algorithm can predict how they will behave in all circumstances. Probabilistic models may work passably well most of the time, but the times where statistical models fail may be exactly the times you really wish they didn’t, as Long Term Capital Management would tell you. Complex systems may comprise many other simple, complicated and indeed complex systems, but their interaction with each other will be a whole other thing. So while you may manage the simple and complicated sub-systems effectively with algorithms, checklists, and playbooks — and may manage tthe system on normal times, you remain at risk to “tail events” in abnormal circumstances. You cannot eliminate this risk: accidents in complex systems are inevitable — hence “normal”, in Charles Perrow’s argot. However well you manage a complex system it remains innately unpredictable.

Complexity and reductionism

The JC has encountered reductionists who see complexity as an emergent property of an algorithm. On this view, even something as simple as Conway’s Game of Life is, if you let it go long enough, complex, as it spawns sub-systems, gliders, glider guns, and these interact with each other in marvellous and unpredictable ways. There is a tacit assumption here that real life — you know, the offworld — is really just a scaled-up version of the Game of Life, being just an implementation of Darwin’s Dangerous Idea, after all.

This is reductionism viewed from the other end of the telescope. Rather than taking the rich tapestry of modern life and boiling it down to basic rules of cause and effect, we start with those basic rules, and scale them up. What prevents us from getting from one end of this spectrum to the other is only an absence of data (from the rich tapestry end) and a want of processing power (from the initial conditions). The universe is nonetheless fully determined at all levels of abstraction.

This undermines the distinction between simple, complicated and complex — they are now just points along a continuum, without hard boundaries between them — and undermines the explanatory power of complexity theory. It is really just saying, “well, in this complex system, something will happen; we don’t know what, but as and when it does we will be able to rationalise it as a function of our rules, by deducing what the missing data must have been.”

Ex-post facto rationalisation to comply with your rules is rather like the work normal scientists do in a research programme, of course. It is a form of narratisation.

See also

• Normal distribution
• Barnacles
• Normal Accidents: Living with High-Risk Technologies by the magnificent Charles Perrow.