In an insightful paper almost two decades ago, Allen and McGlade (1986) stressed that the key elements of fishing (like ancestral hunting) are discovery and exploitation, not simply the latter as is currently the case in western agriculture. As well as a desire to avoid congestion (like in the bar problem), fishing involves communication, adaptive learning and the emergence of non-uniform tactics and role-playing among fishing agents. Within such a socio-ecological system, key questions on our relationship with nature and the general problem of the management of a complex adaptive system can be explored.

Largely hidden from human view, the marine ecosystem is complex with respect to its species composition as well as to the processes occurring in it. As exploitation increases, Beddington and May (1977) showed that such an ecosystem moves towards instability and greater risk of collapse. Human responses tend to amplify random fluctuations, thereby further increasing the risk of collapse.

Ignoring the complex issue of ecosystem dynamics for the moment, let us concentrate on the strategies of the owners of the fishing vessels. From an owner’s selfish viewpoint, if there are no institutional rules imposed on him, he aims to maximise his own vessel’s catch. Obviously, the best thing for him to do will depend on what the other fishing vessels in the vicinity of his vessel are doing. Thus our fishing problem is a self-referential one.

Allen and McGlade (1986) developed dynamic, multi-species, multi-fleet models to explore the implications of different fishing strategies and information flows among fishing vessels in a Nova Scotia fishery. They identified two fishing strategies, calling them Cartesians and Stochasts. Stochasts search randomly for better sites or use their own intuition, without resorting to information that is shared between associates. They are risk-takers, seeking higher returns commensurate with the higher risks they take. At the other extreme we have the Cartesians, skippers who (for various reasons) are unwilling to take any risk and who only go to the zone promising the best known return. It does not matter how long it takes for a Cartesian to reach their chosen site, as long as the final catch is guaranteed.

Allen and McGlade found that less information exchange among the fishing vessels ensures a more random response on the part of boats (i.e. more Stochasts among the fleet). Because the Stochasts continue to explore less visited parts of the system, the fishery as a whole has a greater chance of survival. If boats refuse to take risks and go only to where they know there are fish, the end result can be disastrous for everyone. Discovery involves risk, but to abandon it invites disaster.

There is a powerful message here that goes beyond the bounds of fisheries alone. In any mobile population, for example, we find some risk-takers and some who are risk-averse. Like the number of fishing boats turning up at the same site, the number of vehicles turning up on a specific road each day is unpredictable. Of interest are the adaptive strategies of drivers exposed to regular traffic jams. Anthony Downs (1962) identified two behavioural classes of driver: those with a low propensity to change their mode or route strategy, called sheep, and those with a propensity to change, called explorers. Explorers search for alternative options to save time. They are quick to learn and hold several heuristics in mind simultaneously. Sheep are more conservative and prone to following the same option. Empirical work in North America has confirmed the presence of sheep and explorer behaviour in real traffic (Conquest et al. 1993).

The parallels between Cartesian or Stochast fishing strategies and sheep or explorer driving strategies may seem striking. Yet they are less surprising if thought of as symptomatic of a more general phenomenon. In the world of technology, risk-averters and risk-takers appear under different guises: imitators and innovators. If we allow for the coexistence of imitative and innovative mechanisms in a population, both must be treated as co-evolutionary variables dependent on the unfolding of events.

Is the coexistence of Cartesian (imitative) and Stochast (innovative) strategies in a human population an Evolutionarily Stable Strategy (ESS)? An ESS is a strategy (or mix of strategies) which, if most members of a population adopt it, cannot be bettered by an alternative strategy. The reason is that the fitness of individuals adopting an ESS strategy is higher than the fitness of individuals adopting other strategies (Maynard-Smith 1982). Another way of putting it is that the best strategy for any individual depends on what the majority of the population is doing (Dawkins 1976). This last perspective helps us to grasp the idea that an ESS may be an important property of self-referential systems.

There are three classes of ESS: pure, mixed and conditional. A pure ESS is a strategy that is consistently exhibited by individuals throughout their lifetimes. A mixed ESS is a complex of two or more strategies varying either within or among individuals over time. In a conditional ESS, each individual's strategy varies under different conditions in the social or physical environment. If the evolutionary payoff for each strategy depends on what other individuals are doing and decreases as a greater proportion of the population adopts that strategy, other strategies may also be an ESS (frequency-dependent selection).

To apply the ESS idea to our fishery problem, consider the following variant of one of Maynard Smith’s simplest hypothetical cases (hawks and doves). Here we draw extensively on the simulation results in Allen and McGlade (1986). Suppose that the only two types of fishing strategy in a fleet of vessels are Stochasts and Cartesians. We want to know whether Pure-Stochast or Pure-Cartesian is an evolutionarily stable strategy. As Stochasts and Cartesians compete for returns, we must estimate payoffs to the different fleet strategies in order to find an ESS.

Serving as the eyes of the fleet, Stochasts search for fishing zones of high return. A fleet with a high proportion of risk-taking Stochasts will discover successive zones of high return, eventually fishing out the high-return zones in a patchwork manner. But Stochasts ignore fish from nearby zones of intermediate returns. As it takes time for the high-return zones to recover, pure Stochast strategies tend to lead to boom-and-bust series of good and bad years. However, this patchwork approach guarantees that the fishery system survives rather than running the risk of collapsing.

Based on the results in Allen and McGlade (1986), we can assign a Pure-Stochast a cumulative payoff of 76 points (equivalent to 6 large circles plus 8 small ones) over a ten-year fishing period. This includes several bad years during which the stocks must recuperate.

Cartesians refuse to take risks and rely solely on the information passed on by the Stochasts. Thus they go only to zones where they know in advance that there are some fish. A fishing fleet dominated by risk-averse Cartesians will simply direct all the boats in the short term to a few of the best zones for fishing. Because the information about which are the best zones is generated by a small number of Stochasts, the tendency for Cartesians is to lock onto one particular location as the best. In the long run, this leads to the fleet exploiting a single location instead of the whole area, leading to a small catch and a small fleet. Based on figures found in Allen and McGlade, a Pure-Cartesian strategy yields a payoff of 45 points over the same ten-year period.

The above figures confirm our suspicions that Pure-Stochasts do better than Pure-Cartesians over time. They are rewarded for taking more risks. Pure-Stochasts achieve better returns because they can maintain fishing activities over the whole area, obtaining a higher catch in the good years and maintaining a larger fishing industry overall. Pure-Cartesians become trapped in a few locations instead of spreading out across the whole area. They are limited each year to a smaller catch and a declining fishing fleet and industry over time.

But is a Pure-Stochast strategy evolutionarily stable on its own? To answer this question, let us suppose that we have a fleet consisting only of Pure-Stochasts. Despite the bad years, they seem to do very nicely, earning the payoff given in Table 4.1. Now suppose that a mutant Cartesian arises in the fleet. Being the only Cartesian, must he compete aggressively with the Pure-Stochast? No. Instead he could follow the Pure-Stochasts, letting them show him where it is best to fish. Then, by fishing at the edges of sites discovered by Pure-Stochasts, he does not irritate them too much. Remaining mostly unnoticed, he can free-ride, enjoying above-average returns for very little search effort.

As long as there are sites offering intermediate returns that are left untouched by Stochasts, there are niche opportunities for Cartesians to fill. Based on figures in Allen and McGlade (1986), 10-year payoffs to a mixed fleet of Stochasts and Cartesians under various levels of information exchange are given in Table 4.2.

When information is shared equally between both groups, the payoff to each is higher than to a Pure-Stochast or Pure-Cartesian fleet (see Table 4.1). Thus a Pure-Stochast strategy is not evolutionarily stable, since it can be bettered by a strategy in which Stochasts and Cartesians communicate and cooperate with each other.

What happens if either group cheats by not transmitting reliable information about their catches? As Allen and McGlade noted, there is an interesting asymmetry. If Cartesians fail to inform Stochasts, there is very little effect (see Table 4.2). But if the information possessed by Stochasts is not transmitted to Cartesians, the latter perish (or are compelled to become Stochasts). Thus Cartesians will try to obtain catch information and Stochasts will try to withhold it. Such effects have evolved in real fishing fleets. For example, Vignaux (1996) found that trawlers fishing for hoki in waters off New Zealand’s coastline do not share catch information, instead basing their own decisions partly on watching where other vessels fish. This amounts to spying. Other tactics include listening in on radios, lying about catches and spreading other misleading information.

Using agent-based simulation, a richer suite of possibilities can be explored. Recent Agent-Based Modelling (ABM) work involving Bayesian belief networks has shown that various kinds of information flow among fishing vessels has important effects on the dynamics and resource exploitation of a simulated fishery (Little et al. 2004). As stated earlier, information flow tends to benefit the Cartesians at the expense of the Stochasts. This asymmetry confirms the importance of realistically representing the rich variety of possible fisher behaviour in any modelling framework that aims to assist with integrated management of fishery resources.

The superior payoff achieved when catch information is shared reliably is in line with observed properties of the minority game. If all agents in a fleet have access to public information about catches and zones of highest return over a period of time, then the agents interact only through this public information and the system has a mean-field character (in the sense that no short-range interactions exist). Self-organisation in such a system is achieved by allowing each agent to have several strategies from which he selects the one that seems best (to him).

In the literature on the minority game, there is a second order phase transition between a symmetric phase (in which no information is available to agents) and an asymmetric phase (in which information is available). It is easy to see why this phase change might occur in our fisheries context. First, if reliable catch information is passed between Stochasts and Cartesians, this may ensure that a boom-and-bust series of good and bad years is avoided. Stochasts can direct some Cartesians to high-return zones and others to medium-return zones, in such a way that no high-return zone will be in great danger of over-fishing. Second, the two strategies are complementary. Cartesians make good use of information, while Stochasts generate it. Together they can exploit the resource more efficiently because of their complementarity.

Ideally, each fleet needs some Stochasts (researchers) and Cartesians (producers) who cooperate within a fleet but not with competing fleets. To survive, each needs the other. What, then, is the ideal ratio of Stochasts to Cartesians? This is a difficult question to answer without considering the dynamics of the fish population and other environmental factors. The higher payoff that a Stochast may enjoy is tempered by the higher risks involved, whereas the risks taken by a Cartesian are minimal.

In a highly dynamic world like fisheries, one may expect the ratio of Stochasts to Cartesians in a fleet to vary in response to the search success of the Stochasts and their willingness to inform their Cartesian partners. Like music lovers at the El Farol bar, the ratio of Stochasts to Cartesians can oscillate forever. But, in this case, there may be no emergent regularity as the congestion level is a dynamic variable. The situation is a self-defeating one, since the best thing for each vessel to do depends on what everyone else is (thinking of) doing.