The Big-Top of Complexity

Many of the topics which now vie for attention within the complexity circus had their origin somewhat outside of the new field, or before the new tent was erected. Examples include artificial intelligence, which itself comes in a variety of incommensurate forms. These include expert systems, which quarry human experts such as doctors diagnosing a patient, to elicit the rules underlying their decision making. Others may depend on artificial neural nets (ANNs). This is a development in computing which explicitly uses a simple model of data processing in brains. Neurons are connected in nets, and ‘fire’ each other off, or not, according to specified thresholds. They ‘learn’ to replicate the same relationships between input data and the diagnoses made. The problem with the latter is that the relationships that link data with output cannot be isolated from the network: they are ‘black boxes’, in a sense truly emergent. It is then difficult to build confidence that the network will correctly identify novel cases. An apocryphal anecdote relates that the US Army trained a neural net to distinguish between Soviet and US tanks by displaying thousands of photographs of each. When it came to using the net in the field to distinguish at high speed between friendly and enemy tanks, all tanks were found to be friendly. It turned out that the net had originally distinguished the two types of tanks by whether the photography was clear (a US tank) or grainy (a surreptitious distant shot smuggled out of the Eastern Bloc). All real time images were clear, therefore all tanks were friendly. Not withstanding these limitations, ANNs have been used in geography in a variety of ways, though usually connected with physical and biogeography rather than human geography – for example, to classify ecosystems from remotely sensed satellite imagery.

A similar ‘blind’ and ‘self organizing’ tool is genetic programming, where thousands of arbitrary rules are assigned to solve a task, and those that are most successful in any generation are selected for the next iteration. The programs often involve random mating (swapping halves of code) and other variation as well. A program might be designed to learn how to play a card game like poker, or be applied to the prediction of energy use in a city.

Another popular development within the big top at the moment is the use of agent based modeling (ABMs). These are in principle rather more sophisticated versions of cellular automata. An application within geography is the modeling of cities, seeing how social segregation, or suburban sprawl, may result from interacting agents with differing behavioral rules. Studying the behavior that emerges, and testing new hypotheses and new rules, can help understand urban processes in a way which is radically different from old style classical spatial economics. These models can be linked to GIS systems so that the ‘board’ is a ‘real’ city. Fun, and useful, though these may be, complexity theory does little to guide the construction of such ABMs.

Developments in robotics may come within the circus tent, not so much for the sake of applications within industry, as for a better understanding of biomechanical principles. The construction of a mechanical dog which can run with the same gait as a real dog reveals that only some of the simulation needs to be in information processing. Much of the performance of the robot is better built into its own mechanical structures and material balances and tensions, so that the control is stimulating ‘inherent response’ harmonics of bouncing and stretching.

Perhaps the most far reaching of complexity’s current buzz words is ‘tipping point’. The implication that there are thresholds beyond which stability becomes instability, or leads to an irreversible, alternative, and undesirable state, has been given political legs, and is used almost regardless of whether there is a well founded model behind the statement or not. So in that sense, complexity has entered the popular imagination as a frightening metaphor. The climate will soon pass a tipping point, immigration will soon pass a tipping point, and the educational system will soon pass a tipping point. Perhaps complexity’s big top needs a little self or ganization and self regulation.