The study of animal innovation is in its infancy. Notwithstanding recent interest in the topic, fuelled by the observation that a species' innovativeness correlates positively with brain size (Lefebvre et al. 1997; Reader & Laland Reference Reader2002), the only book on this topic was published in 2003 (Reader & Laland Reference Reader and Laland2003b; cf. Reader & Laland Reference Reader, Reader and Laland2003a), and the number of published articles dedicated to it remains less than 30. These observations, when juxtaposed against the thousands of articles and tens of books on animal social learning, reveal animal innovation to be a fledgling and understudied domain. We welcome Ramsey et al.'s stimulating article, and enthusiastically endorse their goal of operationalizing animal innovation. Although we express reservations over aspects of their scheme, we applaud their efforts and encourage others to engage in the exercise.

We agree with Ramsey et al. about the need for new methods and concepts in the study of innovation, but begin by maintaining that the Reader and Laland (Reference Reader, Reader and Laland2003a) definitions are not as problematic as claimed. They were explicitly operational: “The above considerations lead us to the following two operational definitions” (Reader & Laland, Reference Reader, Reader and Laland2003a, p. 14, emphasis in original). The definitions were deliberately broad and encompassing, they eschewed reference to particular psychological processes, and were designed to be practical to implement in both the laboratory and the field. The aspect of Reader and Laland's definitions to which Ramsey et al. object, the requirement that innovations introduce novel behavioural variants into a population's repertoire, is a central part of that operationalization, in that an innovation can be designated when first observed, with no requirement that the original innovator be identified. This has left innovations (sensu product) easy to quantify, and extensive databanks have been collated on innovation frequencies in both primates and birds. True, such numbers are biased in a variety of ways, but researchers have devised statistical procedures to account for these (Lefebvre et al. Reference Lefebvre, Reader and Sol2004; Reader & Laland Reference Reader2002). Contrary to Ramsey et al.'s suggestion, the approach has facilitated comparisons in innovativeness between species, and allowed the relative rates of innovation in sex, age, and social rank categories to be investigated (Kendal et al. Reference Kendal, Coe and Laland2005; Lefebvre et al. 1997; Reader & Laland Reference Reader and Laland2001; Reference Reader2002). Nor is innovation, by Reader and Laland's definition, necessarily a function of population size: group and population size are not correlated with innovation rate in primates and birds, respectively (Nicolakakis & Lefebvre Reference Nicolakakis and Lefebvre2000; Reader Reference Reader, Preuss and Kaas2006). We do not pretend that Reader and Laland's definitions are perfect, or cannot be improved upon, but nonetheless they exhibit considerable currency.

In contrast, we are concerned that the Ramsey et al. operationalization is overly reliant on subjective judgements of individuals or coworkers. Judgements are required about how frequently or reliably the novel behaviour appears in the population (i.e., enabling exclusion of environmental induction), how rapidly they increase in frequency (i.e., deemed indicative of social/asocial learning), the reliability of “indicators of social learning,” the identification of ecological correlates of innovation, and so forth, all of which will vary among researchers. Ramsey et al.'s key is only tenable if each decision could be made reliably and objectively – but could they? There are an infinite number of ecological correlates to a behaviour: how many should a researcher consider before moving to the right in the key? Little consideration is given to the problem of identifying social learning within a population: There is currently no available metric with which to evaluate the decision point “The behavior is known to propagate via social learning” – this is never known with certainty (Laland et al., in press). We fear that criteria will be assessed differently by different investigators, leaving the identification of innovations somewhat arbitrary. Of course, there are concerns over the subjectivity of existing innovation databases, but at least the Reader and Laland criteria are clear, objective, and measurable.

Leaving aside its subjectivity, there are other features of Ramsey et al.'s approach that worry us. We disagree that a behaviour pattern should be precluded from categorisation as an innovation because it is seen to be a response to environmental change (see also Reader commentary). Considerable available evidence indicates that much innovation is governed by the adage “necessity is the mother of invention” (Reader & Laland Reference Reader, Reader and Laland2003a). In a search of primate journals, approximately half of recorded innovations, where data were available, were prompted by ecological challenges such as periods of food shortage, dry seasons, or habitat degradation (Reader & Laland Reference Reader and Laland2001). Similarly, many innovations are induced by the social environment, for example, one in which young/subordinate animals are forced to innovate in order to obtain food or mates monopolised by older/higher-status individuals. This is an awful lot of data to throw away, and Ramsey et al. leave themselves vulnerable to false negatives. Just as social learning is expected to be adaptive and tied to the environment, so is innovation. Ramsey et al.'s claim, that behaviour patterns with strong fitness consequences are unlikely to be innovations, is entirely unproven.

Some aspects of Ramsey et al.'s approach also appear contradictory. We applaud the authors' use of captive animal resources in their method. The presentation of natural or captive populations with novel puzzles or challenges and recording the characteristics of innovators and contextual variables, as recommended by Kummer and Goodall (1985), has proven an effective experimental approach to the study of innovation (e.g., Kendal et al. Reference Kendal, Coe and Laland2005; Laland & Reader Reference Laland and Reader1999; Morand-Ferron et al. Reference Morand-Ferron, Lefebvre, Reader, Sol and Elvin2004). However, a logical extension of Ramsey et al.'s line would be to dismiss this experimental approach completely – such behaviour patterns are, after all, “environmentally induced,” so should we therefore conclude that we are not studying innovation?

The authors acknowledge that their method bears similarity to the ethnographic (“geographic”) method, which has been used to infer the presence of cultural traits in primates and cetaceans, but this method is a contentious model on which to base their scheme (Laland & Janik Reference Laland and Janik2006; Reader commentary). We welcome Ramsey et al.'s emphasis on “rigorous application,” but they will acknowledge that the ethnographic method has yet to be applied to culture with sufficient rigor to satisfy all. Thus far, readers have had to content themselves with the opinion of the authors that no ecological correlates were present: they were rarely listed or subject to statistical analyses. This approach to science, in which verdicts are given by a jury of authoritative experts, leaves us uncomfortable. There is much to commend in Ramsey et al.'s approach to innovation, but objectivism must not be a casualty of its operationalization.