In this section, we show how an organism can have mental misrepresentations. We focus on functional theories of representations that include misrepresentation understood in terms of dysfunction. Such accounts differ from traditional accounts of mental representations, e.g., simple causal theories, in which there is no place for dysfunction and misrepresentations are simply not functional at all (for an elucidation why such causal theories have problem with misrepresentation, see Bielecka 2016). Furthermore, the capacity of a cognitive system for misrepresentation can help account for (limited and evolutionarily merely satisficing) rationality of biological agents. First of all, no rational biological agent is faultless. And only cognitive agents capable of misrepresentation actually also represent. Second, cognitive agents may commit mistakes that would render them irrational unless we could explain their behavior as caused by their misrepresentation. For example, someone may sit on the bus stop even if no bus is coming. If we know that the bus no longer stops there, sitting there in order to get somewhere is irrational; however, it may be explained as rationally justified (and caused) by someone’s misrepresentation of the bus schedule. Hence, someone at the bus stop may simply be in error about the bus schedule; she is not faulty of trying to get transported to the city center by just sitting on the bench. This shows that it’s not global irrationality that is the main problem with the person on the bus stop; it’s the local misrepresentation. Preserving the assumption of her rationality without some form of misrepresentation would be difficult if not completely implausible. We take the possibility of system-detectable error or misrepresentation for the system to be a necessary condition for the satisfactory account of mental representation in biological agents.
First, we will follow Mark Bickhard to show how misrepresentations are possible as dysfunctional representations. In order to do that, first, we introduce Bickhard’s account of mental representation and his account of function. Then, we show that his account of function needs an etiological dimension in order to better explain why representational mechanism can be functional and further how mental disorders can be partly dysfunctional. That’s why in our account of representation, we apply a hybrid account of function, conjoining Bickhard”s and Millikan’s accounts of function.
Bickhard understands mental representations in terms of his interactivist model. According to his model, all organisms are interactive systems as they are self-maintenant but also they are autonomous in the sense of self-reconstruction: organisms (every organism is an interactive system) are able to “recruit and even manipulate (themselves in) their environment so as to (contribute to) maintain(ing)” (Bickhard 2009, p. 555). A representational mechanism appears at a second level of interactive systems, in a system that not only interacts with environment but also can differentiate organizational and processing properties of the first level. Only those organisms that have representational mechanisms can be fully autonomous in a sense of manipulating themselves in their environment so they are in a constant process of constructing and reconstructing their representations and confronting them with environment. Although Bickhard has an account of what a representational mechanism is, his view is not entirely satisfactory because it lacks further explanation why representational mechanisms could be adaptive.
In Bickhard’s account, function is accounted for in biological and dynamic (thermodynamic) terms:
Far from equilibrium processes require maintenance in order to be stable, and such maintenance is functional relative to the stability of that system—it serves a function insofar as it contributes to that stability (Bickhard 2009, p. 555).
In this view, a feature X is functional only if X helps a system to self-maintain itself as a system far from its thermodynamic equilibrium. Basic representations (emergent from the basic structure of the biological organism) are functional because they play a current role in its self-maintenance while indicating organism’s future possible actions (a similar account of intentionality is defended by Deacon 2012). The representational function helps an organism to stay far from its thermodynamic equilibrium; thanks to such a function, the organism can anticipate its action results, as the representational function is related to anticipations of action results, and the organism can survive or even learn something. These anticipations of future possible actions have satisfaction conditions. An organism or its subsystem can detect whether these are satisfied by confronting anticipations with results of actions caused by them. Furthermore, the idea that organisms themselves recognize the adequacy of their representations is enshrined in Bickhard’s notion of system-detectable error. So, only an organism that can detect its own errors can really anticipate its future actions. In this sense, error detectability makes representations necessary in a stability-maintenance process.
Let us illustrate this idea with a deliberately simplistic example of a frog’s mistakenly snapping a bumblebee. A frog that anticipates snapping a fly but snaps a bumblebee recognizes it as a misrepresentation while realizes that it snapped a bumblebee (as long as snapping a bumblebee has different consequences than snapping a fly; for example, it may be sour for the frog; the details here are fictional and are meant here for illustrative purposes only).
Furthermore, a frog can use its own error in a process of learning that in particular conditions, the representation of an object X as a fly, is a misrepresentation, because it leads to wrong actions. However, Bickhard’s approach seems to be somewhat unsatisfactory in situations in which an organism’s action fails, even though its representation is correct. For example, the frog’s representation may be correct but during snapping, the insect is eaten by a bird that flies nearby. Even if Bickhard insists that the contents of basic representations is implicit and the satisfaction conditions of actions are identical to satisfaction conditions of representations, which would mean that he could reply that the frog implicitly represented the situation that included non-intervention of birds in snapping, such an insistence is not biologically plausible. This would imply that implicit representations are unboundedly precise in representing all possible conditions of accuracy, which is what Bickhard obviously and plausibly denies. He could only reply that the frog will, in such situations, detect misrepresentation even if there is none. However, the frog’s needs are better served if it ignores such instances and detects no misrepresentation, and continues attempts of snapping if another fly appears. How would this be possible then?
To make such instances possible, in our account the criterion of correctness is not the success of action but the internal consistency of representation. Two bits of (structural) information should be compared by a system to check their consistency. If there is a problem, it may mean that one of them is in error. For example, if one is more reliable for a system, it may be inferred that this one is correct, and another will be rejected. Deacon 2012 seems to hint at a similar solution to this problem:
The consistency (redundancy) and inconsistency (non-redundancy) of the evidence is not itself a guarantee that a given interpretation is accurate. Faced with the problem of comparing alternative interpretations of the same events, one is often forced to analyze other features of the source of the information to determine if there are systematic biases that might be introducing spurious or intentionally skewed levels of redundancy. Creating the false appearance of independent sources of information is, for example, a major tool employed in propaganda and confidence schemes (Deacon 2012, p. 406).
Note however that even if Deacon uses the notion of interpretation in the above citation, the account defended here uses the notion of structural, and not semantic information, so the notion of interpretation is not required. So, if a frog has auditory information of a fly and visual information of a bumblebee and assuming that a visual information is more reliable, only then a frog has a misrepresentation of bumblebee as a fly (here also the details are meant for illustrative purposes only).
To sum up, according to Bickhard’s account of representation, the representational content is constructed by an organism in a constant process of constructing and reconstructing its representations and confronting them with its environment. An organism, or its subsystems (at least sometimes) recognizes its own errors and can correct them. However, Bickhard’s account has some problems concerning failed actions as consequences of accurate representations, and this is why we propose to supplant his account with internal consistency checking as offering a more basic mechanism for error detection. After all, recognizing that an anticipation deviates from reality also arguably requires consistency checking (between the anticipation and the detected result of the action).
In addition, the etiological dimension can make a mechanism responsible for a given type of mental representation if the mechanism appeared at least once in a history of an organism and played its role that had helped an organism to survive. According to Millikan’s (1984) biological account of function, representations are functional only if they are products of adaptive mechanisms. To have a function:
an item must also come from a lineage that has survived due to a correlation between traits that distinguish it and the effects that are “functions” of these traits, keeping in mind that a correlation is defined by contrasting positive with negative instances. Intuitively, these traits have been selected for reproduction over actual competitors. Because the correlation must be a result of a causal effect of the trait, the trait will not merely have been “selected” but will have been “selected for” (Sober 1984). Thus a thing's proper functions are akin, intuitively, to what it does by design, or on purpose, rather than accidentally (Millikan 1984, p. 8).
So an organism has a representational function F if a certain property P was selected for F. It means that such a property is positively correlated with realization of that function and it contributes to further replication of P because of the effect that P was naturally selected. Sober distinguishes two different concepts, selection for and selection of (Sober 1984, 1993). According to him, selection for a property P means that having that P causes success in survival and reproduction. Selection for is contrasted with selection of; it says why some properties increased in frequency while the former one describes causes, the latter–effects of selection:
to say that there is selection for one trait (Fast) and against another (Slow) is to make a claim about how those traits causally contribute to the organism's survival and reproductive success (...) One trait may be fitter than another because it confers an advantage or because it is correlated with other traits that do so. (Sober 1993, p. 83)
A trait of having a heart by an organism is adaptive for pumping blood in a population when their members have hearts as a result of earlier selection for having a heart and also because having a heart contributed to their fitness because the heart pumped blood. So, having a heart was selected for in a sense it has a role in biological adaptation, because it helped an organism to survive while pumping blood, but the heart-beat was selected of in the sense that it is a latter-effect of selection, not causally important for a heart to function properly. Applying Sober’s terminology to representational mechanisms, while a representational mechanism was selected for in the sense that it has a role in biological adaptation, vehicular property are selected of in a sense it is an effect of selection.
According to Bickhard and Christensen, what makes the etiological account of function different from the dynamical one is that only the latter emphasizes serving a function, which they find more important to understand as it is actually much more important in the system dynamics:
In certain respects we are simply focusing on a different issue: etiological theory takes as its task the problem of explaining what it is for a part of a system to have a function, whereas we focus on what it is to be an adaptive system, and on the nature of serving a function relative to such a system (Christensen and Bickhard 2002, p. 4)
Moreover, Bickhard and Christensen notice some problems with the etiological account of function and argue that etiological function is epiphenomenal. Let us introduce an iconic thought experiment used in a discussion on causal relevance of mental representations (Davidson 1987; Christensen and Bickhard 2002; Bickhard 2009; Krohs 2007). Imagine a lion that springs into existence and is atom-by-atom identical to a real lion.Footnote 2
According to Millikan, only the real one has functions because it has evolutionary history, while the science-fiction lion doesn’t. Bickhard and Christensen find her claim counterintuitive:
Function, in this view, is dynamically—causally—epiphenomenal. It makes no difference to the causal or dynamic properties of an organism whether or not its organs have functions. Etiological models thus fail to naturalize function. Etiological history explains the etiology of something, but it does not constitute any of the causal or dynamic properties of that something. Etiology cannot constitute the dynamics of what it is the etiology of (Bickhard 2009, p. 558)
So history doesn’t constitute any causally-relevant property – it is causally epiphenomenal. Although such a consequence seems to be intuitive in an artificial example of a science-fiction lion, it doesn’t in biological ones. In our view, a representational mechanism seems to be still functional even though it doesn’t work properly because it produces delusions and they fail to be corrected because the system is no longer capable of detecting error in some of its representations. But this mechanism normally has a feature that allows it to correct error; this feature has been selected for detecting error. However, this feature currently has no effect, or is causally epiphenomenal in the system, and does not serve its function anymore.
We propose, in essence, to assume a hybrid account of function (Davies 2001) that includes both etiological and dynamical dimensions. The notion of etiological function is used to account for the why-questions concerning the biological structures, while the dynamical function is best understood as answering the how-questions that pertain to the current dynamics. Therefore, both can be used in a more complex, multidimensional account of biological functionality, which has its roots in Tinbergen's (1963) account of explanations of behavior (Miłkowski 2016).
In this account, cognitive errors themselves are currently dysfunctional (as in dynamical function) but the representational mechanism is adaptive, or etiologically functional. Hence, this mechanism has also a function to detect its own errors, even if it is not currently performed or served. Surely Bickhard doesn’t claim that error should be always detected so he would admit that an error detection mechanism is dysfunctional, but it isn’t possible for him to claim that it is an error detection mechanism if its damage is persistent in an organism. If such a damage is inborn, there is no evidence that such a mechanism has a role to detect errors because it has never played such a role in an organism so it didn’t contribute even once to being far from dynamical equilibrium. In contrast, according to Millikan’s account of proper function, this mechanism can be classified as having a proper function that is now not served. That’s why supplying Bickhard’s account of function with etiological elements helps to save the basic principle of his account.