Predator inspection in fish

Probably one of the most often analysed phenomena in terms of the IPD, but also the most debated one is predator inspection in shoaling fish. When a predator is stalking a shoal of minnows (Phoxinus phoxinus), guppies (Poecilia reticulata) or sticklebacks (Gasterosteus aculeatus), individuals will separate from the shoal, swim tentatively towards the predator until only a few body lengths away, wait there for a few seconds and then slowly return to the shoal. It has been suggested that in such visits the fish can gather information about the identity, precise location and current motivational state of the predator (Milinski et al. 1990, Reboreda and Kacelnik 1990, Dugatkin and Alfieri 1991, Dugatkin 1991a, b, Turner and Robinson 1992).
Two fish will approach the predator more closely than will single fish, so they both obtain the same benefit b and share the cost c(R=b-c/2). If one fish lags behind the other, he will get the same information without the cost (T=b) while his sucker takes all the risks (S=b-c; Dugatkin and Alfieri 1992). If neither of them inspects, c and b are zero (P=0). If c>b one condition for the Prisoner's Dilemma is satisfied: T>R>P>S (Dugatkin and Alfieri 1991). Moreover, it is suggested that guppies (e.g. Dugatkin and Alfieri 1991) and sticklebacks (e.g. Milinski 1990) are capable of recognising previous defectors/cooperators, thus fulfilling another condition of the IPD.
Yet, there is no case study proving that all conditions are satisfied (Milinski 1992) and hence all experimental results are intensely debated (e.g. Milinski 1990, Lazarus and Metcalfe 1990, Reboreda and Kacelnik 1990, Dugatkin 1991b, Turner and Robinson 1992, Milinski 1992). An alternative way would be to assume the conditions for an IPD are fulfilled and predict the fishes' behaviour as specifically as possible. Assuming the fish use TFT, they must display the three characteristics associated with TFT: nicety, retaliation and forgiveness. Despite the problems translating these features into measurable, behavioural traits, Dugatkin (1991a) claims that his data support the use of TFT in guppies. His results indicate that guppies are 'nice' by beginning their initial inspection at about the same point in time, 'retaliatory' by turning back and making a fish that lagged to far behind (i.e. defected) the closest to the predator (i.e. defecting the defector) and 'forgiving' by sticking close to the former defector if he in turn now is approaching the predator. (See Dugatkin and Alfieri 1992 for similar results).
However, as already pointed out, the outcomes are debatable, since there is a lack of sufficient unambiguous evidence that fish do prefer to inspect in pairs (Turner and Robinson 1992). Moreover, if the knowledge gathered by inspecting fish is transferred to their non-inspecting shoalmates, as is commonly agreed upon, why should inspectors bear the cost of inspection but share the benefits (Dugatkin 1992)? How related are shoalmates with each other? Further experiments are needed to decide whether the IPD is an applicable model for predator inspection in fish.