R1.1. Is all learning innovation?

Some of the commentators (e.g., Rendell, Hoppitt, & Kendal [Rendell et al.], Sargeant & Mann) have suggested that our definition of innovation might be so broad that all learning (or all individual learning) ends up falling under the rubric of innovation. We agree that this would be a problem if it were the case, but fortunately it is not. First of all, our definition explicitly excludes social learning from the process of innovation. By doing this we are not denying the possibility that social learning can prompt innovation or that innovations can spread via social learning. Instead, we are denying the identification of the process of social learning with the process of innovation. We are also not denying the importance of social learning. Pace Logan & Pepper, we take social learning to be central to the study of innovation, since social learning allows innovations to have effects beyond the individual (indeed, this is the reason we decided to examine the concept of innovation). Viciana & Claidiere's point that the processes of social learning and innovation can be intertwined is well taken. But the close connection between social learning and innovation does not make them the same process.

A further restriction in our definition is that innovation must be learned but not environmentally induced. As we discussed in the target article, the concept of environmental induction is a tricky one and is revisited in section R1.4. The point we would like to make here is that the restrictions we have placed on the concept of innovation, that it is behavior that is novel, learned (both in the sense of not innate and in the sense of modifying the organism's behavioral dispositions), not socially learned, and not environmentally induced, clearly mark innovative behavior as a subset of learned behavior.

R1.2. Is our definition tautological?

Giraldeau, Lefebvre, & Morand-Ferron (Giraldeau et al.) accuse us of “tautological thinking.” In order to understand their critique, let us first say a few words about the nature of scientific definitions and the reasons why we think it is important to develop a new definition of innovation. One role of definitions in science is to refer to new concepts or things. Generally, this is accomplished by introducing a novel term and stipulating its meaning. Such stipulative definitions are neither true nor false, and, although they may be useful, they are not very interesting. Our definition of innovation is not of this kind. Another common kind of definition is (merely) descriptive – the sort of definitions one finds in a dictionary. These definitions – which can be true or false – do little more than help one learn the meanings of terms. The kind of definition of innovation that we have produced is explicative; instead of merely stipulating or describing the use of the term innovation, we have attempted to explicate the concept of innovation, which heretofore has either been only nebulously defined or has been merely defined in operational terms. We feel that in constructing this definition we have been sensitive to the class of phenomena researchers have been picking out as innovations and not deciding “a priori on the characteristics that it should possess if it were to be recognized,” as Giraldeau et al. phrased it.

Are we guilty of tautological thinking? A tautology, of course, is a sentence that is true by virtue of its logical form (and is thus independent of matters of fact). Giraldeau et al. are not explicit in identifying such sentences. Instead they say that our “categories appear entirely tautological and far from being amenable to strong empirical study.” This statement is a bit puzzling since categories are not the sort of thing capable of being tautological, but we feel that what they mean is that the criteria we propose in our operationalization have been produced a priori, but are in fact based on a posteriori facts. For example, the key in the target article's Figure 2 is based on the putative fact that the near simultaneous appearance of a behavior in a population makes it improbable that it arose through innovation followed by social learning. This inference – from simultaneity of origin to the improbability of innovation – is not based on a priori assumptions about the rate of the spread of innovations. To use a fanciful example, if wild chimpanzees started using cell phones, the pattern of spread of innovations would be different than we would expect otherwise – innovations could appear within and between multiple, geographically isolated populations nearly simultaneously. Such changes in the world would necessitate changes in the key. Giraldeau et al. argue that, “The assumptions used by Ramsey et al. are perhaps correct, but it is far from self-evident and must be the object of scientific scrutiny.” We could not agree more. It is an empirical question whether the key is in fact successful in identifying innovations – as we have defined them – in the wild, and it is not immune to modification in light of empirical discoveries.

R1.3. Individual versus population level definitions

We defined innovation at the individual level, arguing that the process of innovation is an individual-level process. We recognized that this individual-level process would be difficult to observe, and this is why we used the operationalization section in the target article to connect these individual-level processes to more easily observable products and properties.

Sargeant & Mann suggest that defining innovations at the individual-level may be problematic. The concluding sentence in their Individual versus population section is: “Thus, we consider ‘innovation’ better defined as a subset of individual learning evidenced by behaviors new to a population, useful for studying conditions favoring the development, spread, or effects of new behaviors.” They thus agree with us that it is an individual-level process involving learning, but we disagree on the evidence that counts for marking the behavior as an innovation. They hold that novelty at the population level is the only such evidence. We hold that novelty at the population level is evidence of innovation, but that it is not the only such evidence. We consider other evidence, such as the distribution of the behavior within and between populations, to bear on the question of whether the behavior arose as an innovation.

Kendal, Dean, & Laland (Kendal et al.) also suggest that the individual-level definition of innovation is problematic and defend the Reader and Laland (Reference Reader, Reader and Laland2003a) population-level novelty definition. The two main reasons that they argue for the population-level definition is that it makes empirical work on innovation easier and that it is more objective. We answer the question of objectivity in section R2.2. But what we would like to point out here is that we are not denying that population-level novelty is a sign of innovation. We are critical of their approach because (among other reasons) we hold that it rules out the possibility of multiple individuals in the same population producing the same innovation – thus innovativeness will be underestimated. Also, despite the fact that Kendal et al. hold that the population-level novelty definition makes the identification of innovations easier, we feel that it is in fact more restrictive than our approach. Our operationalization does not require long-term studies (like those needed in order to say whether a behavior is truly novel for a population); estimates of innovations can be obtained from cross-sectional field studies with data on the distribution and/or properties of the behaviors.

Some of the commentators that argue against the individual-level definition end up implicitly endorsing it. Consider this quote from Rendell et al., for example: “First, we do not believe that defining an innovation as something new to the individual, rather than the population, makes sense. No doubt … it is possible that more than one individual in a population might produce the same innovation independently” (our emphasis). Rendall et al. do not seem to see the inconsistency in these statements. If innovation is defined as population-level novelty, then it follows from this that if an individual produces some innovation, I, then later independent productions of I by other members of the population will not count as innovation. For the latter to be possible, innovation must be defined at the individual level. In fact, it is considerations such as the fact that “more than one individual in a population might produce the same innovation independently” that motivated us to produce an individual-level definition.

A population-level definition of innovation conflates multiple, distinct processes. For example, Deleporte points out the importance of cultural loss in relation to innovation. If one were measuring innovativeness using a population-level definition, the innovation rate (and thus innovativeness) would be a function of the rate of cultural loss, because if a novelty is lost (from the population) and then created anew, the new creation (being novel at the population level) will be recorded as an innovation. On the other hand, if the novelty did not disappear through cultural loss, then independent creations of the behavior would not count as innovation. This seems to us highly problematic: Innovation and cultural loss are distinct processes that should not be conflated. Our individual-level definition does not possess such problems. An innovation is an innovation whether or not the same behavior exists in the population.

R1.4. Environmental induction versus innovative response to environmental novelty

The most common misunderstandings of our definition of innovation center around the distinction between environmental induction and innovative responses to environmental novelty. Many commentators have taken all environmental involvement in the production of novel learned behavior to be instances of environmental induction. For example, Reader cites the case of kingbirds feeding by street lights in several Caribbean islands and suggests that under the target article's classification system this would be considered environmental induction (and not innovation). But we would suggest that the mere fact that the behavior has arisen independently in multiple locations does not imply that it is environmentally induced. If all kingbirds in the presence of streetlights immediately begin to feed near them, this suggests environmental induction. But if the process is slower, if only one or a few kingbirds exhibit the behavior at first, then many of them begin to exhibit it – this suggests innovation. It is thus not inconsistent with our concept of innovation for the same behavior to arise multiple times independently. In fact, as discussed in the previous section, the possibility of multiple origins of the same behavior is one of the benefits of our individual-level definition of innovation.

Rendell et al. similarly argue that our category of environmentally induced novel behaviors should actually be considered innovations. They say that “Milk-bottle opening by blue tits in the United Kingdom is widely touted as an innovation, but without the environmental induction of soft-topped milk bottles, would never have occurred.” Is opening the milk bottles by blue tits an example of environmental induction? This seems doubtful, since this is not a behavior that arose throughout the United Kingdom shortly after the introduction of the foil-topped milk bottles. Instead, the behavior appears to have arisen by innovation in a limited number of cases and then spread.

Rendell et al. introduce the distinction between “passive” and “active” innovation as an alternative to our environmental induction–innovation distinction. We think their distinction is interesting, but do not see it as incompatible with ours. Instead of having “passive” innovation replace our concept of environmental induction, we see it as being intermediate between innovation and environmental induction, perhaps occupying part of the space that we marked off as “weak innovation.” We believe at least some of their disagreement with us stems from a misunderstanding of the distinction we draw between innovation and environmental induction, evidenced by the following passage. “We are unconvinced by the restriction on cases involving what the authors term ‘environmental induction’ […], and on cases with significant positive fitness impacts.” We did not suggest either that environmentally induced behaviors cannot have positive fitness impacts or that innovations must have a positive fitness impact. In fact, our concepts of innovation and environmental induction are independent of fitness considerations. (This is not to say that such considerations cannot come into play when trying to assess whether a behavior is an innovation – see sect. 4.2.2 in the target article for a discussion of how the fitness impact of a behavior can serve as evidence for whether it is an innovation.)

Khalil's conclusion that our definition of innovation is inconsistent also stems from the mistaken idea that innovations must be fitness enhancing. His argument is that in order for innovations to boost fitness, there must be an inducement from the environment. But, he concludes, if innovations are defined in terms of a lack of environmental induction, then the definition is inconsistent. It is quite clear that if innovations are not defined in terms of fitness enhancement – like the fitness-neutral definition in the target article – then his argument does not apply.

The more liberal definition of innovation advocated by our critics is in reasonable compliance with their operationalization, which often includes responses to human-induced environmental changes. We have raised the bar, and asked for more than mere flexibility or the ability to adjust to human-made changes in the environment. Some work suggests that curiosity and problem-solving abilities are not correlated across species (Mettke-Hoffmann), suggesting they are regulated by different factors. To mention just one obvious difference, exploratory tendencies may depend strongly on perceived risks (Gajdon) or on context (Mettke-Hoffman), which may be irrelevant, at least at the proximate level, to problem-solving abilities. In the end, this is an interesting empirical issue, and the answers may even vary by taxon. But we must note that most bird work to date has taken the liberal definition, which probably underlies some of the comments. As we point out later, there is a risk that our critics' approach has measured only this ability to adjust, for example, to human observers or to human-made objects and structures, and not the narrower category of what we have termed innovation, which is more then mere behavioral flexibility.

Gardner discusses the interesting case of cross-fostering. A human-raised chimpanzee, for example, will exhibit a host of novel behaviors. Are all such novelties cases of environmental induction? No, just because the captive individual is exhibiting a behavior not represented by wild individuals does not mean that they are environmentally induced. It could be the case that captivity provides more opportunities for innovation, or disposes individuals to innovate (see, e.g., Gajdon's kea example), and that many of the behaviors exhibited in such cases are innovations. Comparative studies can help resolve this issue. Comparative studies with cross-fostered individuals can also help us decide whether a behavior observed in the wild is an innovation. If cross-fostered individuals exhibit a behavior also exhibited by wild individuals, this decreases the probability that this behavior is an innovation.

R2.1. False positives and negatives

The operationalization we developed is deliberately conservative, and thus better at avoiding false positives than false negatives. Several commentators argue that we cannot properly deal with the role of ecology. Giraldeau et al. complain that we need to know a lot about the ecology of the species we study before we can decide whether a behavior is an innovation. Similarly, Sargeant & Mann point out that most foraging behaviors were correlated with ecological variation and maturation. We agree that our approach is likely to produce false negatives, because wherever there is a clear ecological correlate to the presence-absence pattern of a behavior across populations we conservatively assume that the more parsimonious alternative, that of a reliable interaction between innate predispositions and environmental releasers, explains the behavior's geographic variation. There are several ways of dealing with this problem. Perhaps the most practical one is to see if there is extensive peering or other indications that social learning may take place before a maturing individual adopts the behavior. Field experiments furnish another option, provided one can exclude neophobia as an explanation of negative results.

We suggested that comparisons with captivity, especially experimental attempts to elicit the behavior with captive animals by offering them the essential conditions, could be an important additional criterion for deciding whether a behavior is an innovation. Gajdon correctly points out that captive animals may be much less neophobic, and thus more exploratory, and hence more likely to find the solution. Ongoing work with orangutans (by S. Lehner, for example) points in a similar direction. That means that comparisons with captivity may falsely suggest that behaviors of wild animals are not innovations, thus increasing the number of false negatives.

Some commentators raised concerns about the key. Giraldeau et al. ask what we mean by a behavior being common. As pointed out earlier, there are no absolute criteria, and it is largely up to the researcher, so long as it remains consistent between populations or studies. Novelty at the population level is a statistical quantity. Our approach makes the criterion explicit and quantifiable, rather than relying on the intuitive guesswork of anonymous authors. Russon, Andrews, & Huss's (Russon et al.'s) discussion of leaf carrying in orangutans (the tendency to carry leafy twigs of a particular species around before selecting a site to build a nest) concludes that it is probably not an innovation. We are puzzled by this conclusion, since it is absent from numerous orangutan populations and it is difficult to design a solid ecological explanation for its absence at these sites given its presence at some others. That makes it a strong candidate for innovation. The key is not meant to deal with ex-captive animals that may be ecologically naïve, or had already acquired some part of the repertoire of their population before being moved into captivity, and are then released into a natural habitat that may or may not be the same as the one they came from. The key certainly does not deal with ecological induction in captivity and might therefore misfire when applied to released rehabilitants.

R2.2. Is our operationalization objective?

Kendal et al. charge that our way of identifying innovations is subjective, or at least less objective than that of Reader and Laland (Reference Reader, Reader and Laland2003a). They do not define what they mean by “objectivism,” but let us assume that they mean something like inter-subjective agreement – that if multiple researchers were told to do the same task, they would produce the same results. Or if different results were obtained, the difference would be due to errors of measurement, not different judgments of the researchers. A proper response to the charge made by Kendal et al., then, requires discovering two things. First, what is the task? And second, to what degree is this task objective?

One task would be to search for words like unusual, novel, or original in the animal behavior literature and then compare the numbers of occurrences of these terms with morphometric data such as brain size. We will refer to this as the anecdote compilation method. (This is the sort of task that Giraldeau et al. promote in their commentary.) Is this method objective? The compilation of the anecdotes may be an objective task, but it is far from clear that the anecdotes were produced with a standardized, objective method. If the anecdotes were not produced in a standardized, objective way, then the claims made by studies using the anecdote compilation approach are undermined. In our target article, we refrained from criticizing this approach. Now that we are criticized as lacking in objectivity, it is perhaps useful to review the differences between the two approaches.

We recognized two major ways of documenting innovations: (1) by recording the first instance in a population subject to long-term study (as per Reader & Laland's Reference Reader, Reader and Laland2003a definition), and (2) by recording properties of the behaviors, such as their rareness within a population or across all populations studied, the behavior's limitation to only some populations but not others, as well as other factors such as the relationship between the behavior and various environmental and social elements. We promoted the second approach as a practical complement to the first one, one that would not require long-term data and that might allow us to identify many more innovations than the first, provided we have data on multiple populations.

In assessing the objectivity of our method as well as the anecdote-compilation method, we should enquire into the method(s) by which the anecdotes were generated. While many anecdotes seem to use approach (1) – evidenced by the use of terms such as novel, unusual, first recorded instance, not noted, and so on – many of them were also likely to have relied on approach (2), for example, a behavior only seen in one place, not in others. It is unlikely that most reports were based on thorough long-term studies of single populations. Indeed, what looks novel or unusual to a mobile observer, visiting numerous sites, may simply be different from the average over multiple populations they have composed in their mind. When terms such as “not reported” or “previously unknown” are used, they are not generally meant to refer to individual populations, but to the knowledge of the species as a whole. Indeed, in many sites where these behaviors were observed, they may have been around for a long time. Inevitably, this means that a mix of approaches (1) and (2) probably underlies the majority of the anecdotes. To put it more bluntly, when commentators criticize our approach, they are criticizing the source of their own anecdotes (the source being “a jury of authoritative experts” [Kendall et al.]). Surely, all should therefore agree that finding a method to replace ad hoc anecdotes with more systematic data, as we tried to develop in our target article, is a priority if we are to make progress. If this reduces the number of acceptable innovations to a trickle (the complaint of Sargeant & Mann), then so be it: It would be very reassuring to know whether the small, but cleaned-up data set obtained this way still produces similar results.

When one relies on the opinions of others, one closes one's eyes to the risk that some of what the authors of the original papers identified as innovations may not be innovations if we use more explicit criteria. It is therefore important to ask whether the anecdote-compilation approach may lead to the inclusion of behaviors that should not be considered innovations or introduce biases in the innovations that get recorded this way. If we examine the list of examples in some of the papers (Lefebvre et al. Reference Lefebvre, Whittle, Lascaris and Finkelstein1997; Reference Lefebvre, Gaxiola, Dawson, Timmermans, Rosza and Kabai1998; Reference Lefebvre, Reader and Sol2004; see also Reader & Macdonald Reference Reader, MacDonald, Reader and Laland2003), we find many instances where all would agree that innovations are the most plausible interpretations, but also quite a few where this is not the case. Such cases may not be innovations in spite of the fact that they are “strange” or “interesting” or “worthy of attention.” First, many cases of unusual food items may simply indicate that the animals have broad diets (such as the many examples of consumption of vertebrate prey by animals that usually eat invertebrates, or of fruit by animals that mostly eat animal prey), or that the opportunity to eat some foods simply rarely arises, such as the meat of a stranded whale. Thus, the more we study a particular species, the more we might see these putative examples of innovation on a regular basis (seasonally or on a longer time scale, but predictably shown by the same individuals). In the worst case, then, all we have is an indication of the overall behavioral diversity of a species. Second, for others where some element of novelty or unusualness is clear, some of our other commentators (e.g., Russon et al.) would rightly question whether we deal with true novelty from the perspective of the behavioral decisions made by the animals and as intended by the definition in the target article. To give just a few examples: When house sparrows systematically search car radiator grilles for (squashed) insects, how different is this (to a house sparrow!) from searching any other rough substrate with crevices? Or, when birds follow tractors that flush prey items, how different is this from merely being attracted to moving prey (Lefebvre et al. Reference Lefebvre, Whittle, Lascaris and Finkelstein1997)? In most of these cases, the answer is not obvious, and detailed case studies, such as those done by Morand-Ferron et al. (Reference Morand-Ferron, Lefebvre, Reader, Sol and Elvin2004), are needed to assess the probability that the behavior is an innovation. Until these are done, many behaviors on the list require judgment calls similar to the ones we had to make in the case of applying the technique proposed in the target article (see van Schaik et al. Reference van Schaik, van Noordwijk and Wich2006), with the difference that we made them explicitly and suggested further work that could resolve their status. Similar concerns can be raised for the comparative work that uses primate anecdotes.

Even if all the anecdotes used in the comparative analyses are true innovations, one may still ask whether they represent an unbiased sample across species. As Healy and Rowe (Reference Healy and Rowe2007, p. 458) pointed out, “these accounts might be biased by the visibility of the species, the species' distribution in relation to human populations, and also the energetic needs of the species. Indeed, some of the innovations given as examples suggest hunger as a driving force (e.g., the first report of a rook eating frozen human vomit: Lefebvre et al. Reference Lefebvre, Whittle, Lascaris and Finkelstein1997).” Visibility to humans may be linked to size or to adaptability to human observation, which may well have neuroanatomical correlates. Mettke-Hofmann suggests that object and food-based innovations may be more likely to be recorded than spatial innovations: “Object- or food-related innovations are probably easier to identify than spatial innovations such as use of a different strata or movement technique, or the use of a novel habitat. In birds, only 13% of the reported innovations (augmented version of the database used by Lefebvre et al. Reference Lefebvre, Reader and Sol2004) can be classified as special innovations, although birds may be more likely than other taxa to show spatial innovations because of their three-dimensional use of the environment.” As we saw earlier, there is also a danger that the anecdotes merely capture omnivory or generalism, and it is not clear a priori that such feeding strategies are a direct result of accumulating innovations. During the analysis, further biases may have crept in, such as when the number of records for taxa is corrected for research effort by counting the number of publications on them. That number may be affected by factors such as geographic distribution, use in biomedical or other research, or, indeed, flexibility and thus tractability as research subjects.

These two problems suggest that the results of the anecdote-compilation are subject to bias of unknown magnitude, which may or may not be serious. More worryingly, there is the risk that they have merely shown us the brain correlates of a trait other than innovativeness. Still, the results of the comparative studies have been quite impressive and robust (Lefebvre et al. Reference Lefebvre, Reader and Sol2004; Reader & Laland Reference Reader2002; Sol et al. Reference Sol, Duncan, Blackburn, Cassey and Lefebvre2005a; Timmermans et al. Reference Timmermans, Lefebvre, Boire and Basu2000), suggesting clear correlations between brain size and the measures extracted through anecdote compilation. But it does not follow from the link between brain size and the anecdote compilations that the trait of innovativeness correlates with brain size. The anecdotes, instead of picking out the trait of innovativeness, may instead often be picking out a different trait, such as generalism. Although it would be an interesting discovery to find a positive correlation between brain size and generalism, this is quite distinct from finding such a correlation with innovativeness in the more specific sense used here.

Our comments are not meant to disparage the important comparative work, which we find very exciting. However, as we noted earlier, there are uncertainties, suggesting to some that the results are not credible (e.g., Healy & Rowe Reference Healy and Rowe2007). Moreover, the anecdote compilation work relies on operationalizations of innovation that are implicitly the same as those we proposed in our target article. It is therefore all the more important that a more rigorous independent alternative is developed to complement this approach. The method proposed here is likewise subject to possible biases, but because these are different ones, we feel it is worth pursuing. If independent approaches produce comparable results, we can have more confidence in the correlates of innovation. The operationalization we propose, if successful, will also help to generate novel insights into the conditions that bring out innovative tendencies, and thus allow us to produce hypotheses on the conditions that favor the evolution of innovative abilities.

R2.3. The grain problem

In the target article section entitled “Delineating behaviors” (sect. 5.1), we pointed out some of the challenges of distinguishing different tokens of the same behavioral type from behaviors that represent genuinely distinct behavioral types. This challenge – the “grain problem” – was discussed by several of the commentators. Rendell et al., for example, use our example of hammering: Are hammering with a stone tool and hammering with a wooden tool distinct behavioral types, or it is merely flexibility of a single behavioral type (“hammer with whatever seems to be a good tool at the time”)? Rendell et al. are rather pessimistic about the possibility of making this distinction and hold that one would need (difficult to obtain) knowledge about the mechanisms involved. We are a bit more optimistic and feel that although detailed knowledge about the animal's cognitive mechanisms would be invaluable, purely behavioral observations and experiments will also help make these delineations.

Locke raises the interesting issue of “behaviors that previously occurred only in an ‘old’ context that, because of some change, came to be emitted in a ‘new’ context[.]” Would such a behavior be counted as a new behavior or not? Our answer would be that since behavioral novelty is context dependent, it is possible that the behavior is a genuinely new innovation. It just depends on whether the individual is indiscriminately performing the behavior (in which case it would not be considered a new innovation) or whether its performance in the new context precipitated a change in the individual's repertoire (in which case it would be considered a new innovation).

As Russon et al. emphasize, it is not “possible to determine what behavior to enter into the key without knowing what grain the actor uses.” We agree that one needs to create a typology of behaviors before one can use the key. But we would like to point out that this is not a problem unique to us. It is a problem for any operationalization of innovation. Reader and Laland's (Reference Reader, Reader and Laland2003a) population approach fares no better than ours: To claim that a behavior is novel at the population level is to claim that it represents a novel behavioral type (and not mere flexibility of a preexisting behavioral type).

R3.1. Mechanisms and innovation

We intentionally did not attempt to specify the kinds of mechanisms that lead to innovation. We did this not because we think such mechanisms are unimportant. On the contrary, we think the study of the mechanisms that underpin innovative behavior is vitally important. We simply feel that such research is at too early a stage to define innovation in terms of it. We fully agree with Gajdon that the investigation into such mechanisms is necessary to understand innovation. Although there are ways to tell whether a behavior is novel without identifying specific mechanisms, knowledge of the mechanisms would provide a powerful way of deciding whether such behaviors are correctly classified and would be invaluable for cases that are difficult to classify.

Gajdon brings up the interesting case of individuals being more or less innovative in different environments: Keas in an aviary were more successful at completing a novel task than individuals in the wild. This is very important, and seems to complement observations of primates (e.g., Menzel Reference Menzel1966; van Schaik et al. Reference van Schaik, Deaner and Merrill1999). Gardner's examples point to a developmental component to these environmental differences, indicating that the kinds of innovations produced by an individual depend on its accumulated experience. It is exactly to facilitate this kind of exciting work that we need to be clear about what we consider an innovation.

In answering Cachel's question as to whether some behaviors of social insects (such as ants) are innovations, it is useful to recall our second definition of innovation, which is posed in terms of repertoires. If the behavior – “insects linking together to bridge a spatial gap,” to use one of Cachel's examples – is part of the repertoire of the members of the species, then it is not to be considered an innovation. But if some insects modify their repertoire and thereby produce novel behaviors, innovation has occurred. There is no a priori reason to exclude insects or other small-brained taxa, though we feel that it is probable that the frequency and complexity of innovations will increase with increasing intelligence, as also suggested by the comparative studies reviewed earlier. This practical problem of recognizing innovations underscores how little we know about the cognitive dimensions of innovation, which future studies must address.

Reader asks whether “innovative propensities are adaptations rather than side-products or indicators of other phenomena, such as behavioral flexibility.” Knowledge of the mechanisms involved is necessary to answer this question. If all of the abilities and dispositions that make one innovative have been selected for not because of the innovations they produce, but because of other roles they play (neophilia or general flexibility, say), then innovative propensities will be side-products. But if it can be shown that at least some elements of the mechanisms underlying innovative dispositions are the result of selection for the production of innovations, then the innovative disposition can be considered an adaptation.

R3.2. Baldwin, learning, and innovation

Hunt & Gray claim that we presume that “the behavior of primates is generally learned and that of non-primates is generally ‘innate’.” In our article we offer no such claims. We do draw many of our examples from primates, but this is not presented as data confirming the claims (which we do not make) that primate behavior is “generally learned” and that the behavior of non-primates is “generally innate.”

Hunt & Gray also hold that we incorrectly characterize what we label “Baldwinized behaviors” as being not learned. Their claim is based on what they feel is a mistaken view of the Baldwin effect. We feel instead that their view of Baldwin is mistaken. (For a more positive take on our discussion, see McCall's treatment of the Baldwin effect and the myriad ways behavior can affect genetic evolution.) Baldwin (Reference Baldwin1896b) draws the distinction between natural and social heredity:

By “natural heredity” Baldwin means innate and by “social heredity” he means social learning. His point is that social heredity can influence natural heredity, that is, that socially transmitted behaviors can become innate. In the target article, we made no claims about the degree to which this is realized in nature. Instead, our point was merely that if this process occurs, our operationization will not pick these behaviors out as innovations.