Mammals thus function as intentional agents who selectively attend both to relevant situations in the environment and to the constituents of their own goal-directed actions so that they can make the best behavioral decisions. We may thus say that mammals are instrumentally rational not only in the minimal sense that they intelligently pursue their goals (the economists’ minimalist definition), but also in the further experiential sense that they know what they are doing.
A number of philosophers have addressed the issue of animal rationality, using a variety of different definitions and criteria (e.g., see Okrent, 2007, and the various papers in the edited volumes by Hurley & Nudds, 2006, and Andrews & Beck, 2018). None of them believes that any nonhuman animal is rational in the human sense – requiring adherence to socially constituted rational norms and the ability to provide explicit reasons for one’s actions – but many scholars attribute to one or another species at least some capacity for rationality involving either actions or decision-making. The proposal I defend here is that chimpanzees and other great apes (I remain agnostic about other nonhuman primates) are reflectively rational in their agentive decision-making – in a way that other mammals are not—and this leads to some novel ways of experiencing the world. Specifically, as chimpanzee agents are deciding what to do: (i) they attend to underlying causal and intentional relations in the world, organized into logical paradigms of implication, which brings rational coherence to how things work; and (ii) they rationally reflect on their own process of decision-making (via a second-order tier of executive monitoring and control, a.k.a., metacognition) that enables them to diagnose problems in their first-order executive decision-making and to intervene in them. These two features are connected in the sense that reflecting on one’s own process of decision-making makes available the concepts necessary for attributing causal and intentional relations to entities and events the external world.
Understanding the logic of causes and intentions
Organisms that understand causality do not just understand what is happening, but also, to some extent, why it is happening, which creates the agentive possibility of manipulating the cause in order to produce the effect (Woodward, 2003). Thus, in experiments involving a completely novel problem, chimpanzees are able to choose a tool that is causally appropriate, and, moreover, to take control of the causal process and make new tools that will work in the new context.
But beyond exploiting tool properties as enabling causes, chimpanzees can also understand causal forces that operate totally independent of their own actions. For example, in one study chimpanzees knew without training that among a series of opaque bottles, those that were heavy must contain the tasty liquid they were seeking whereas those that were light must not (Hanus & Call, 2008). And in a study in which they could not act on the objects at all, they inferred that when one end of a balance beam tilted down it meant that the opaque cup on its end contained a banana (whereas the cup on the other end did not; Hanus & Call, 2011). These studies indicate that chimpanzees understand that heavy things exert a downward causal force. Importantly, in some instances, chimpanzees seem to assume that even when there are no obvious causal forces at work, there must be some somewhere, and so they attempt to discover them. Thus, when chimpanzees were rewarded for taking overturned blocks and setting them upright, and then one of the blocks would not stand upright, some of the chimpanzees picked up the recalcitrant block and visually inspected it underneath, seemingly trying to discover the cause of the problem (Povinelli & Dunphy-Lelii, 2001).
Great apes’ causal understanding generates creative inferences organized into logical paradigms. For instance, in the experiments on tool choice, apes infer such things as “if a tool with property A is used, then B must happen”. Then, actually using the tool completes the inference: (i) if A is used, then B happens; (ii) A is used; (iii) therefore B should happen. In other experiments, apes can make backward-facing inferences from effect to cause, in this case using exclusion based on a simple kind of negation (what logicians call contraries). Thus, Call (2004) showed a chimpanzee a piece of food, which was then hidden in one of two cups. Then, in the key condition, the experimenter shook the empty cup. The chimpanzee observed only silence. To locate the food the chimpanzee had to infer backward in the causal chain to why that might be, namely, that there was no food inside the cup. The chain of inferences was thus something like: (i) the shaking cup is silent; (ii) if the food were inside the shaking cup, then it would make noise; (iii) therefore, the shaking cup is empty (and so the food is in the other cup). Following Bermudez’s (2003) analysis, these inferences and paradigms involve the two key elements of logical thinking: the if-then conditional and negation. Both occur in only “proto” form: the if-then conditional is proto because it only concerns causal (not formal) relations, and the negation is proto because it only concerns contraries such as presence-absence, noise-silence, etc.
Causality operates differently in the animate world. To understand an agent’s actions, one must understand that its behavior is generated by the goals toward which it is aiming and the perception/knowledge it has about how to achieve those goals in the situation. Knowledge of an agent’s goals and perceptions in a situation then enables prediction of its behavior. Thus, for example, when a subordinate chimpanzee is competing for food with a dominant, it can take into account whether or not that dominant sees a potentially contested piece of food (because of judiciously placed barriers). And it even can tell if the dominant has seen the food in that location in the immediate past and so knows it is there (even though at the moment it cannot see it; Hare et al., 2000, 2001). In general, in such experiments, apes understand that a competitor will go for an object only if: (i) it wants or has a goal/desire for that object (i.e., it would not compete for a rock); and (ii) it perceives or knows that that object is in a certain location. Chimpanzees thus understand how competitors work as agents – that is, in terms of their goals and perceptions—and can use this understanding in novel contexts to predict their behavior. In addition, at least some chimpanzees seem to understand even more about an agent’s decision-making process. In particular, human-raised chimpanzees do not imitate a human performing a strange action, such as turning on a light with his foot, when he has no other choice since his hands are otherwise occupied: he is not freely choosing to use his foot and so I should not imitate him since I have a free choice and so can use my hand, as normal. But chimpanzees do imitate a human when he has freely chosen that same strange action in the absence of constraints (since he and they are similarly unconstrained). This process has been called “rational imitation” because the social learner is comparing its own process of situation-sensitive decision-making to that of another agent (Buttelmann et al., 2007).
As in the physical/causal domain, in the social/intentional domain apes’ inferences are logically structured. In the food competition experiments the competitors inferred of one another in this situation: if he has the goal of getting the food, and he perceives its location (and so knows how to get it), then he will go for it. But if he does not have the goal, or does not perceive a way to achieve it, then he will not pursue the goal. And from the other direction: if my competitor is engaged in a particular activity, then she must have had both a relevant goal and a relevant perception. These inferences constitute a kind of logically structured paradigm analogous to the human practical syllogism. Moreover, in the rational imitation study, chimpanzees made a backward-facing exclusion (counterfactual) inference based on proto-negation. Specifically, when they saw a human operate a device with his foot when his hands were externally constrained, they inferred from his behavior backward to his decision-making: (i) he is not using his hands; (ii) normally, if he had a free choice, he would be using his hands; (iii) therefore he must not have a free choice (so I can ignore his action choice). As in the case of logically structured causal inferences in the physical domain, then, these logically structured inferences about others’ intentions and actions in the social domain employ the two most basic elements of human logical thinking: proto-conditional (if-then) causal inferences and proto-negation based on contraries.
Chimpanzees thus seem to understand the underlying causal and intentional structure of their physical and social worlds – why things happen as they do—in ways that other mammals do not. And they see these causal and intentional relations as logically interrelated; their physical and social worlds make rational sense.
Reflective planning and decision-making
Like all mammals, great apes plan their actions. But, in addition, they can plan for a future goal that they do not at the moment actually have. Thus, when they are sent out of a room in which they have previously chosen and used a tool successfully, they will take with them the tool that they can anticipate they will need in the future, assuming that the problem situation recurs (the study is actually with chimpanzees’ sister species bonobos; Mulcahy & Call, 2006). Planning for a future imagined goal in this way would seem to require some new executive, that is reflective, cognitive skills. Such skills would also seem to be required when chimpanzees are able to perceive and resolve a goal conflict by comparing how the means to achieve different simultaneously present goals are incompatible but could be made compatible (Herrmann et al., 2015).
Chimpanzees’ ability to reflect on their executive functioning is on full display in experiments on their decision-making. Thus, in one study chimpanzees either did or did not witness a human hiding food inside of one of several tubes. When they witnessed the hiding process, they chose a tube immediately. But when they did not witness the hiding process, they went to some trouble to look into the tubes to discover where the food was located before choosing. The apes knew when they did not know, or at least when they were uncertain, but beyond simply opting out they diagnosed that they were missing a specific piece of information and then determined how to acquire it (Call & Carpenter, 2001; Call, 2010; Bohn et al., 2017). This would seem to require reflecting not just on one’s behavioral functioning but on one’s process of decision-making. Attempting to causally diagnose problematic decisions before they are behaviorally executed fulfills a standard criterion for rationality: self-critical reflection on one’s own decision-making. Such behavior also indicates that they are employing a kind of “computational rationality” (Gershman et al., 2015) in the sense that they must decide if the potentially available information is worth the effort needed to gather it.
In a variation on this theme, chimpanzees seem to reflect on their decision after they have made it to see if they have made a mistake. In an experiment apes were given the opportunity to visually locate the best food at location X. They did this, indicating their belief by choosing that location (though not receiving the food as a result). Then, they were exposed to new information that called their initial belief into question: the new information suggested that the best food might be in location Y. The apes had the possibility at this point to seek further information (or not) that could either confirm or disconfirm their initial belief. Many apes then actively sought more information to resolve the discrepancy between their original belief and the new information, by looking into location X from another angle to double-check their initial judgment (so as to make the best decision; O’Madagain et al., submitted). The apes in these cases were self-monitoring and controlling their executive decision-making after they had made an initial decision; they were reflecting on the decision in the light of newly obtained information and discerning the need to possibly revise that decision.
Planning for future goals, resolving goal conflicts before making a final decision, and diagnosing problems and intervening in executive decision-making, all reflect chimpanzees’ ability to executively monitor and to some degree control their own executive functioning. This suggests that not only do they have an executive tier of functioning in the manner of all mammals, but, on top of this they operate with a second-order reflective tier of executive supervision and control, relying on what have been called metacognitive skills. Most mammals are self-regulating their intentional actions executively, but in the absence of a second-order reflective tier of functioning they cannot monitor and control the executive decision-making processes themselves. In contrast, chimpanzees monitor and control not only their goal-directed actions, but also the cognitive processes involved in their own executive functioning. Mammals attend to what they are doing, but chimpanzees, in addition, attend to what they are thinking (via processes of metacognition; see Carruthers, 2005, for a higher-order theory of consciousness and Shea & Frith, 2019, for a “global workspace” model of metacognition that are somewhat similar to this account).
Second-order executive (reflective) functioning
Importantly, the reflective tier of agentive organization was crucially important in the evolution of chimpanzees’ unique cognitive skills for understanding logically interconnected causal and intentional relations in the external world. Specifically, chimpanzees’ understanding of causality and intentionality resulted from an attribution to external events of some of their own decision-making processes that they were now able to consciously access from their new second-order reflective tier of functioning, which also provided the common workspace and representational format necessary for comparing and aligning internal (1st person) and external (3rd person) events in the attribution process. The way this worked was similar but slightly different for intentionality and causality.
Beginning with the “easier” case, chimpanzees understand others as intentional agents acting and making decisions toward goals as guided by perceptions. The proposal is that this understanding originates evolutionarily with self-experience, a variant of so-called simulation theory (Gordon, in press). The point is a conceptual one. If a Martian came down to earth and informed us that without any obvious organs it could still “see” things, how could we understand this except through our own experience of seeing. If the Martian said it could “grue” things based on some kind of radiation unknown to humans, how could we understand this without any firsthand experience? There is always an element of “theory” as well, of course, because the other agent’s particular experience is different from ours in its specifics: perhaps it has perceptual access to something that we do not. But the equating of self and other experience is a necessary prerequisite.
Several studies demonstrate chimpanzees’ ability to understand the experience of others in terms of their own experience. In one study, a chimpanzee experienced a situation in which it could see through a screen lid on a box to detect what was inside (Karg et al., 2015). The box was then reoriented, so that now from the chimpanzee’s new side-viewing angle, the screen lid was opaque. A competitor then approached the box and looked straight into it, from the angle that the chimpanzee subject had used originally. When the two of them now competed for the food inside the box, the chimpanzee subject knew that the competitor could see through the lid to the food inside even though she herself could not see the food at the moment. The only way the subject herself could know this was from her own previous experience of having looked directly through the lid into the box from the original viewing angle, which she was now attributing to the competitor. (See studies by Kano et al., 2019, and Schmelz et al., 2013, for further evidence along these lines.)
The process of attributing mental states to other agents based on one’s own mental states is at least relatively straightforward because there is a clear similarity between the actions of self and other: all individuals of the same species, including the self, are highly similar in their bodies and behavior (and so the individual is able to use so-called “inverse planning” to predict the other’s actions; Baker et al., 2009). But the generalization from self to other is not nearly so straightforward when considering attributions to physical events involving inanimate objects and physical causality. Unlike animate agents, physical objects only move when they are “forced” or caused to move by an animate agent (or else by some mysterious action at a distance like gravity, which Isaac Newton himself considered an occult force). David Hume (1739/40) thus wondered about the basis of human causal understanding. When one billiard ball strikes another and knocks it across the table, we experience only a spatial-temporal contiguity: a moving ball contacts a stationary ball and it then moves, seemingly as a result. But what justifies an inference that there is a causal “force” involved?
Recall the argument that rats do not just associate their act with its result, but they understand that their act caused the result (Dickinson, 2001). But there is a huge gap between the experience of such internally generated causality and external causality among inanimate objects. Piaget’s (1952) idea is that what bridges the gap is the use of tools, at which chimpanzees are expert. To use a tool flexibly and reliably there must be an integration of the movement of the tool, as caused by the agent, and the properties of the tool in relation to the substrate. Therefore, the cause of successful tool use is both the organism’s action and the properties of the tool in relation to the substrate, across the organism/environment divide, as it were. But in the process of tool use, the causal properties of the tool are only participating passively as enabling causes. It is still a further step to understand objects as exerting a causal force on their own, independent of the self’s own actions. For this, it may be that the ape needs to somehow see physical objects as operating in the manner of intentional agents, that is, on analogy with the causal relations that hold between an agent’s action and its effect in the environment. Perhaps apes are making some such animistic attribution to physical events, and this is the basis for their understanding of causal forces (just as humans may do; Collingwood, 1940, p. 322, suggests: “Causal propositions … are descriptions of relations between natural events in anthropomorphic terms.”).
Evidence for this proposal comes from the fact that chimpanzees structure their causal understanding into paradigms of logical inferences, as described above. If they know that event X causes event Y, then they know that if X happened then Y did also, and also that if Y did not happen then X did not either. Such logically structured inferential paradigms constitute evidence for a self-based hypothesis for the origins of causal attribution because they almost certainly derive from the causal logic of the agent’s own actions. Thus, the kind of causal understanding of one’s own action that rats possess yields such inferences as: if I act, there will be a result; if I do not act, there will not be a result; if there is no result, then I did not act causally effectively; if only one of two ways can cause a result, and the first one is not causally effective, then the other one will be causally effective; and so on. These kinds of inferences are made on the first-order executive tier aimed at one’s own actions and their effects. Then, from the second-order reflective tier, chimpanzees (as tool-using manipulators of the external environment) attribute these internal causal inferences about self-action to external events that seem self-generated (e.g., objects that spontaneously fall or are blocked). Channeling Piaget (1974), then, we may say that an ape’s inferences about the causes of its own actions are implications, whereas its attempts to explain external events (e.g., so as to predict them) are explications, with both requiring a reflective understanding. They both reflect on the same “logic of action”, just differently.
Beyond the intentional agency and instrumental rationality of mammals in general, then, chimpanzees are rational agents who can reflect on their thinking and decision-making metacognitively, using a second-order executive tier of functioning, which also empowers them to experience logically organized causal and intentional relations in the external world. And so, if we humans wish to imagine ourselves as chimpanzees, we might engage in an act of tool use or tool making, for example, employing our causal understanding in preparing and using a stick to pry off the bark off a tree. Or perhaps we might engage in an act of predicting what another person will do when we are competing with them in a concrete situation or just observing them from afar as they go about concrete goal-directed activities. Invoking the continuity assumption, I see no reason to think that our experience would be substantively different from theirs in these situations, assuming, that is, that we can ignore those aspects of our human experience that are not available to chimpanzees. And so let me now be more explicit in specifying exactly what must be ignored.