2.1. Types of mimicry

We are all familiar with common cases of mimicry, such as stick insects or inoffensive droneflies copying the red and yellow stripes of the honeybee to avoid being preyed upon (Brower and Brower Reference Brower and Brower1962). A general definition of mimicry, however, is hard to come by. As a first approximation, mimicry usually (perhaps always) involves a model and a mimic. The model is the entity being mimicked, whereas the mimic is the organism that changes so as to develop some sort of resemblance with the model. For this resemblance to be classified as an instance of mimicry, however, it has to serve certain purposes. Let us consider some illustrative examples.

A first paradigmatic category is “Batesian mimicry,” in which a harmless organism deters predation by imitating a harmful model. A classic example is provided by nonvenomous snakes that have evolved the characteristic tri-color ringed patterns of the venomous coral snake (Greene and McDiarmid Reference Greene and McDiarmid1981). Other related phenomena involve rewardless flowers imitating the color or shape of rewarding flowers to attract pollinators (Dafni Reference Dafni1984)—reproductive mimicry—and the Photuris female firefly imitating the light pattern of Photinus female fireflies to lure males and devour them (Lloyd Reference Lloyd1975)—aggressive mimicry. More generally, a key aspect of these kinds of mimicry is that the mimic somehow imitates the model and enjoys a benefit, whether it concerns protection, predation, or reproduction, without paying the cost. ^{Footnote 2}

By contrast, as traditionally understood, in Müllerian mimicry both mimic and model benefit from the resemblance. Monarch and Viceroy butterflies provide the best-known example: Their mutual resemblance facilitates predator learning. As both butterflies taste unpleasant, any predator that eats one of them will quickly learn and come to avoid both species (by failing to distinguish them). Müllerian mimicry is often presented as involving only two species, but it often includes a larger number of species that form “mimicry rings” (Mallet and Gilbert Reference Mallet and Gilbert1995).

Batesian, Müllerian, reproductive and aggressive mimicry represent some paradigmatic categories, but many other phenomena are sometimes classified as mimetic: background matching, countershading, dazzling, and so forth (Merilaita et al. Reference Merilaita, Scott-Samuel and Cuthill2017; Ruxton et al. Reference Ruxton, Allen, Sherratt and Speed2018). To simplify our discussion, we will focus on only two of these. On the one hand, background matching ^{Footnote 3} involves a set of strategies that avoid detection by resembling the background against which the organism is observed. On the other hand, masquerade consists in resembling an inanimate object, typically to avoid being preyed on. Peppered moths (Biston betularia) and flat-tailed geckos (Uroplatus fimbriatus) are well-known examples of the former case and stick insects (order Phasmatodea) and the common sea dragon (Phyllopteryx taeniolatus) (which resemble branches and seaweed, respectively) of the latter.

A crucial distinction between these different phenomena is that they tend to involve different evolutionary dynamics. Batesian and aggressive mimicry, for example, typically exemplify negative frequency-dependent selection: Mimics benefit from resembling the model without paying a cost, whereas the existence of parasitic mimics has some cost for the models (their mimicked features risk becoming less efficient, e.g., in deterring predators). If mimics are frequent enough, models might thus tend to change their appearance, while the mimics’ interest is to keep this similarity. By contrast, Müllerian mimicry involves positive frequency-dependent selection, as both species usually benefit from sharing an advertisement display (Roy and Widmer Reference Roy and Widmer1999). Other phenomena like background matching and masquerade typically involve resemblance to nonliving objects, so neither of these frequency-dependence evolutionary models apply.

Still, not all strategies for avoiding attack and damage fall under the category of mimicry. For instance, the purpose of disruptive coloration is to avoid or minimize damage by presenting some color or shape that hides or hinders the correct perception of the organism’s shape or edges. Similarly, it is controversial whether organisms that avoid detection by being transparent (e.g., the brush-footed butterfly Greta oto) should qualify as mimics.

2.2. The definition of mimicry

Developing a satisfactory concept of mimicry is a complex enterprise. As suggested in the preceding text, most definitions are controversial for various reasons, including the fact that theorists disagree as to whether masquerade or camouflage should be included. The goal of this article, however, is not to provide a general definition of mimicry that is useful for biologists across the board. Rather, we need a concept of mimicry that may play a certain theoretical role, namely, help illuminate the nature of deception.

Which analysis of mimicry can best serve our theoretical purposes? Batesian, Müllerian, aggressive, and reproductive mimicry are central cases of deception, which the analysis should certainly include. However, because we aim for a broad definition of mimicry that enables us to draw fruitful connections with deception, and because background matching and masquerade include paradigmatic instances of deception such as geckos, stick insects, and the common sea dragon, we consider these should be included as well. As a first approximation our analysis of mimicry should encompass all these phenomena.

Having settled on this minimal set of desiderata, we can draw several lessons from some accounts of mimicry found in the biological literature. Consider, for instance, one of the classic definitions:

This analysis faces various difficulties. First, it assumes mimicry involves signals, which is not necessarily so. For example, orchids in the genus Ophrys display a resemblance with female bees that attract males that attempt to copulate with them and end up spreading the plant’s pollen (Schiestl Reference Schiestl2005; Anders Nilsson Reference Anders Nilsson1992). Consequently, mimicry need not involve signals, unless one assumes an extremely liberal notion of signal, according to which almost anything counts as such (and which would confuse cues with signals). Likewise, some organisms mimic nonliving entities (e.g., rocks, sand, twigs).

Secondly, requiring the identification of mimic and model implies a relatively sophisticated capacity on the part of the receiver. Bolas spiders attract moths by mimicking female pheromones, but whether something like identification takes place in the moth’s brain is unclear. Moreover, the focus on identification could unduly exclude possibly deceptive phenomena, for instance background matching, which involve a “failure of awareness” rather than “mistaken identification” (Font Reference Font2019; the terms are taken from Vane-Wright Reference Vane-Wright1980).

More recent accounts are also problematic for these or other reasons. Consider, for instance, Dalziell and Welbergen’s proposal (Reference Dalziell and Welbergen2016, 612; emphasis added):

Here, mimicry is analysed in terms of perceiving similarity, which is a relatively sophisticated capacity: to perceive a similarity between A and B implies more than merely perceiving A and B. It requires representing that A and B are similar in certain respects and might even require the possession of some concepts. We do not think pollinators that are misled by rewardless mimics, or moths by pheromones released by bolas spiders, need to possess the capacity to distinguish between similar and different objects. Similarity can lead a receiver to error without being represented as such by the receiver. ^{Footnote 4}

Still, mimicry does affect some of the receiver’s cognitive states (broadly understood), as it works by altering the receiver’s capacity for discriminating, identifying, or recognizing organisms (Ruxton et al. Reference Ruxton, Allen, Sherratt and Speed2018). Orchids mislead pollinators such as bees and stick insects mislead predators. Thus we would at least say that some minimal form of perceptual, cognitive, or representational capacity is required on the part of the receiver.

Let us now present a working definition of mimicry that will allow for an account of deception that does not necessitate signaling or some cognitively demanding processing involving conceptual representations. As we have seen, mimicry involves a partial copy of a model (typically, an organism or an object) that need not be understood in terms of signaling. But if mimicry is not a matter of communicating a particular content, what function does it perform? How does an organism benefit from resembling another entity? As seen in the preceding text, it seems hard to deny that the function of mimicry is to influence the receiver in certain ways. More precisely, its purpose is to produce a certain representational state in the receiver ^{Footnote 5} ; this is why current definitions use expressions like “discriminating,” “identifying,” or “recognizing.” But which representational state is supposed to elicit mimicry? We suggest that resemblance is useful because it tends to produce in the receiver the same representational state that the model would have generated. Accordingly, we suggest this is the function of mimicry.

Putting all this together, we may define mimicry as follows:

We think this analysis has all the properties a satisfactory definition should include ^{Footnote 6} and can accommodate the cases mentioned already:

Certainly, among all the suggested definitions of mimicry, this is a relatively broad one because it includes background matching and masquerade as well as Batesian and Müllerian mimicry. Nonetheless, it is not too broad: It still excludes some mechanisms for avoiding predators, such as disruptive coloration, which do not mimic any model.

The definition relies on the notion of a partial copy. This is a purposely vague concept; whereas by definition mimics need to resemble the model in various ways, this copy will always be partial in an objective sense: An identical copy of a model—which has all and only the model’s properties, including its DNA, and so forth—would just be another instance of the model (e.g. another bee, rather than a mimic of a bee). The mimic is always an imperfect copy. Of course, how much it needs to resemble the model and which aspects are worth copying will depend on each case (Ruxton et al. 2004, 159–61). This is influenced, for instance, by limited receiver discrimination, evolutionary lags, or cost-benefit constraints (Kikuchi and Pfennig Reference Kikuchi and Pfennig2013): It may be enough that the predator hesitates, identify the target a bit more slowly, or is deceived because it cursorily checks its environment—at high speed and/or with little attention to detail. For instance, the tail of the spider-tailed viper (Pseudocerastes urarachnoides) is only broadly similar to a spider: It would not bear close scrutiny but suffices to lure birds. It may also stem from the simultaneous mimicking of multiple models, or from cases of exaggerated or “supernormal” traits, such as for the Ophrys orchid (Dalziell and Wenbergen Reference Dalziell and Welbergen2016, box 2). Relatedly, note that the interesting sense of a copy is observer-relative: What matters is that the copy appears similar to the model relative to the observer’s perceptual abilities. As Font (Reference Font2019, 13) suggests, mimicry is “in the eye of the beholder.”

Finally, some cases, such as transparency, are not clearly included or excluded (Gomez et al. Reference Gomez, Pinna, Pairraire, Arias, Barbut, Pomerantz, de Marcillac, Berthier, Patel, Andraud and Elias2021; Barnett et al. Reference Barnett, Michalis, Anderson and Cuthill2020). Transparency does not produce a copy of the background; but the transparent producer does create the same representational state as its surroundings would, by virtue of a partial resemblance (often, only part of the body is transparent). The issue hinges on the weight one puts on the distinction between being misinterpreted and not being seen (as hiding is not a case of mimicry), which we need not decide here.

Having provided a preliminary definition of mimicry, our next question is: What distinguishes deceptive mimicry from other forms of mimicry?

3.1. A mimicry-based definition

What is the relationship between mimicry and deception? The original proposal we would like to explore is that deception is a specific kind of mimicry. This may sound surprising and even counterintuitive, as mimicry is often considered as a kind of deception rather than the reverse, although seldom for conceptual reasons. However, even theorists of mimicry do not agree on this point. For instance, whereas Font (Reference Font2019 11) argues that “deception (understood as perceptual error) is a hallmark of mimetic relationships,” Jamie (Reference Jamie2017) emphasizes the existence of “non-deceptive mimicry,” or “rewarding mimicry,” which benefits the models as well as the mimics (and so includes some cases of Müllerian mimicry). Moreover, insofar as mimicry, which may target signals as well as cues, seems more widespread than signal-based deception, a unified analysis that includes the latter within the former rather than the reverse has better chances of success.

If deception is a form of mimicry, what makes it distinctive from other forms? We suggest that deception is supposed to cause a representational state with false content. As a result, we can modify the previous description of mimicry as follows:

For example:

First, as stated in the definition, deceptive mimicry only necessitates a partial copy of the model. As we argued, mimicry is always imperfect. The definition also covers cases in which deceptive mimicry has a double function, for example both hiding from prey and escaping predators (Aaronson Reference Aaronson Richard1983).Footnote ⁷

Secondly, note that this definition of deception does not require deceptive traits to be signals (for more on that, see the text that follows). Nevertheless, because signals are a specific kind of trait, this approach can easily include them. ^{Footnote 8} Consider the following example:

Interestingly, this approach not only correctly predicts which traits should be classified as deceptive but it also rightly excludes some mimics as not engaging in deception. For instance, Müllerian mimicry is typically not an instance of deception because its function is not to produce a false representational content, but more something like hiding the truth: Given the similarity between monarch and viceroy butterflies, an organism that perceives one of them is not in position to distinguish whether it is in front of a monarch or a viceroy. This is not revealing the whole truth, rather than producing a false representation and that seems to be the right result.

Background matching is a more difficult case, but the explanation for this requires some stage setting. First of all, note that in DEC the expression “causing a state with false representational content” can be understood in various ways (Chisholm and Feehan Reference Chisholm and Feehan1977). For instance, one could interpret it as causing the receiver to acquire a false representational state or as causing it to maintain a false representational state. Some favor acquisition, for instance when choosing to define mimicry on the basis of behavior alteration, and thus exclude masquerade and camouflage (Font Reference Font2019). However, one might suggest that causing an organism not to acquire a true representational state should also be included in the definition. Although a full discussion of these options would be complex and beyond the scope of this article, we think the perspective offered here suggests that the first two options (acquiring or maintaining state) should be ruled in, whereas the latter (not acquiring a state) should be ruled out. In short, if hiding truth sufficed for deception, then probably all cases of mimicry would be classified as deceptive, which would blur some distinctions that we consider worth keeping. Therefore, our tentative suggestion is that deception should be interpreted as involving producing or maintaining a false representational state, whereas causing an organism not to acquire a true representation should be excluded. ^{Footnote 9}

If this reasoning is on the right track, then whether background matching should be classified as deception depends on whether it causes the receiver to produce or maintain a false representational state or not. This might happen in some cases, but not in others. For instance, in some situations the receiver might possess a mental state that we can approximately capture with the expression “there is no prey around,” in which case background matching would be maintaining a false representational state. In other scenarios, however, the receiver might lack any contentful mental state about prey, so background matching only causes it to fail to acquire a true representational state, which we have argued does not suffice for deception. Thus, some cases of background matching probably count as deceptive, whereas others do not. Not much seems to hinge on this issue.

So far we have discussed cases of mimicry and deception that involve relatively stable features (shapes, colors) and recurring signals (fireflies’ color patterns). Yet, because one of the main goals of this article is to illuminate the nature of deception in general, our approach should also encompass one-shot deceptive signals. Fortunately, we think they can also be accommodated by DEC:

Note that utterances can also have functions. The vast literature analysing the concept of “function” has established that functions can be grounded in different processes (Millikan Reference Millikan1984; Neander Reference Neander1991). In orchids and fireflies, functions are probably grounded in natural selection, whereas in the case of human utterances, they might derive from human intentions. Hence, we think our account of deception in terms of mimicry can account for human and nonhuman forms of deception.

In short, we think that according to this definition of mimicry, deception can be regarded as an instance of mimicry. The relationship between some of the phenomena discussed so far is illustrated by Figure 1.

Figure 1. Examples of mimicry classified by two axes: signal/no signal and deceptive/nondeceptive.

3.2. The functional theory of deception

Despite its original inclusion of deception within mimicry rather than the reverse, our approach is coherent with several functional approaches found in the literature. It is in line with most definitions of mimicry, which are also functional (e.g., Dalziell and Wenbergen Reference Dalziell and Welbergen2016). Furthermore, it has strong affinities with Artiga and Paternotte’s (Reference Artiga and Paternotte2018) proposal of a functional definition of deception. Indeed, we will argue that the differences between DEC and their functional theory of deception are minimal. Nonetheless, we think our perspective makes some important contributions. First, we arrive at a similar conclusion from a substantively different point of departure: from an original analysis of mimicry and the claim that all deception is an instance of mimicry. Second, in addition to its other virtues, highlighted in the previous sections, the perspective offered here is also slightly superior, for two reasons: It shows how their account can deal with a recent objection raised by some philosophers and enhances its universality. The goal of this section is to describe and develop these aspects.

According to Artiga and Paternotte’s proposal (Reference Artiga and Paternotte2018):

One reason for the convergence between FT1 and our account is that both avoid characterizing deception in terms of fitness benefits or costs—in particular costs that incur to the receiver (Christy Reference Christy1997). As Font (Reference Font2019) rightly states: “Mimetic signals are deceptive in the mechanistic sense because they elicit receiver errors, i.e. misperceptions or categorical mistakes, not because they are costly.” A functional definition allows a deceptive trait to be favored by natural selection without having to specify in virtue of what fitness combinations render it advantageous.

Now, a central aspect of this original analysis remains unexplained: How should we understand the receiver’s “misinformative state”? What is it for a receiver to possess a misinformative state in the relevant sense? Although Artiga and Paternotte’s (Reference Artiga and Paternotte2018) avoid commitment to any specific interpretation, Birch (Reference Birch2019) and Fallis and Lewis (Reference Fallis and Lewis2021) have recently argued that some straightforward ways of understanding this notion do not work. On the one hand, if “misinformative state” is interpreted as false belief, then organisms such as insects or reptiles, insofar as they do not have beliefs (understood as mental states), cannot be deceived. That would be an unwelcome result because a central motivation for the functional theory of deception is to accommodate human and nonhuman deception. As an alternative, one could interpret “misinformative state” as a state that increases the probability of a nonactual state or decreases the probability of an actual state (along the lines of Skyrms Reference Skyrms2010). However, on that understanding, many cases of hiding truth would be wrongly classified as deceptive (Birch Reference Birch2019). For instance, the appearance of the monarch butterfly increases the probability of a nonactual state (a viceroy). More generally, nonfully informative signals would be classified as deceptive, insofar as they increase the probability of a nonactual state. That result puts some pressure on Artiga and Paternotte: Unless they can find some interpretation of “misinformative state” that yields the right results, their account is in jeopardy.

However, we propose that considering deception as an instance of mimicry provides an answer to this worry: “misinformation state” in FT1 should be understood as a state with false representational content. Note that we are using “representational content,” very broadly, to refer to a state that represents or is about certain aspect of the world and that has truth/accuracy conditions. We certainly don’t presuppose that having representational states requires possessing any conceptual or sophisticated cognitive ability. If one has qualms with the notion of “representation,” we would be happy to use some other expression like “intentional state,” “intentional icon” or the like.

In any case, it is here that our account of deception as mimicry can be understood as a modification of Artiga and Paternotte’s proposal (Reference Artiga and Paternotte2018). The analysis of deception we offered already was the following:

By relegating the identity of the model to the background, we can derive a definition of deceptive state that looks very similar to Artiga and Paternotte’s (Reference Artiga and Paternotte2018) proposal:

Apart from the specification that “misinformative state” now refers to “false representational state,” there are only two differences between FT2 and Artiga and Paternotte’s FT1. First, FT2 definition requires deception to piggyback on a model. Note, however, that we are assuming a very liberal notion of model: A model need not be a living being or be present in the environment of the actors. Indeed, a model need not be an actual organism, object, or state of affairs because this role can also be fulfilled by a noninstantiated type. Although classical theories of mimicry predicted that this phenomenon only exists in the presence of an actual model (reviewed in Ruxton et al. Reference Ruxton, Allen, Sherratt and Speed2018), more recent research puts this assumption into question. There are a number of mechanisms that can explain the maintenance of mimicry in allopatry, such as gene flow, range contraction/expansion, and different forms of selection (Pfennig and Mullen Reference Pfennig and Mullen2010). Some mimics even lack a specific model. For instance, the Australian orchid genus Diuris engages in “guild mimicry”: They evolved a color and shape similar to a guild of yellow and brown pea plants (Faboideae) to deceive pollinators. The caterpillar of the moth Hemeroplanes triptolemus mimics a snake to deter predators, but it doesn’t resemble any specific species of snake. Likewise, in crypsis the heterogeneity of background type might lead to a mimetic compromise between them (Quicke Reference Quicke2017, 31). All these cases can be thought of as involving a copy of a noninstantiated type (which, of course, bears interesting relations with instantiated types).

This point is important because it enables us to generalise the notion of mimicry to other domains. In particular, some human forms of deception also mimic noninstantiated types—for example, a new string of words that no one has uttered before. ^{Footnote 10} If I lie to you by uttering the sentence “Bob moved from Bristol to Canberra to meet Cailin,” this is deceptive in virtue of being a copy of the sentence an honest person would utter, but that has the function of causing a false representational state.Footnote ¹¹ If this sentence has never been uttered before, it makes my utterance a copy of an uninstantiated type (which, again, bears certain interesting relations with instantiated types). Our previous reasoning suggests that this is not an ad hoc move to account for human deception; mimicking noninstantiated types is a common phenomenon in the natural world. Thus, this point highlights the universality of such a functional account, which, as Artiga and Paternotte (Reference Artiga and Paternotte2018) noted, covers both nonhuman and human deception.

Secondly, in contrast to FT1, FT2 does not include failing to acquire a piece of information as being sufficient for deception, for the reasons already outlined: Hiding truth should be distinguished from deceiving. In any case, Artiga and Paternotte were not holding that failing to acquire a piece of information suffices for deception; rather, they suggested that some people might consider it sufficient. Thus, we do not think this is a real disagreement between FT1 and FT2.

Why then does interpreting “misinformative state” as false representational state help to assuage Birch (Reference Birch2019) and Fallis and Lewis (Reference Fallis and Lewis2021) worries? It is because the previous counterexamples do not apply any more. The color pattern of a monarch butterfly is not supposed to produce a false representational state, but at most to cause a failure to acquire a true representational state. More generally, not fully informative signals typically aim at hiding truth, rather than causing a false representational content. Thus, this response grants critics of the functional theory that the various interpretations of “misinformative state” provided are inadequate but replies that there is a more promising alternative that yields the correct results, as well as being motivated by an independent theory of mimicry. Thus, we cannot agree with Skyrms (Reference Skyrms2010, 76) when he writes:

On the contrary, as we have argued, it is important that a definition of deception depends on imputed propositional content that is false, rather than being information based. In section 5 we will argue that one way of understanding this concept of representational content resorts to a teleological theory. But another way to supplement our account with an explicit definition of false representational content would be to use the game-theoretic literature on signaling and deception. Several recent works of this kind have suggested possible definitions of the propositional content of a signal; we further discuss these in section 4. For now, it suffices to note that for a definition of propositional content to fit our account, it would only need to entail that in cases of deception, states involve a false propositional content. That is, it should not hold that a deceptive state has true content because it is understood disjunctively, for example, as “I am either a sexual partner or a predator” (see section 4 for more details).

Overall, regarding deception as mimicry not only provides an original and potentially illuminating perspective on deception in general but also adequately refines and extends Artiga and Paternotte’s proposal.Footnote ¹² Furthermore, crucially, it addresses a recent objection suggested by Birch, and Fallis and Lewis. Therefore “misinformative state” should be interpreted as “false representational state.”