2024. Here, I investigate whether the mind needs causal representations. I make the case that not all causal models are equal, but that basic causal models are central to an organism’s ability to survive.
“Causality is a constraint common to all ecological niches.” — Macphail (1987, 645)
Does the mind need causal representations? In this essay, I will argue that the minds we are familiar with do indeed need causal representations. I will ground my discussion in an account of evolution and agency, and argue that the minds of organisms, such as ourselves, need different types of causal representations precisely because living agents face—and have faced since the origin of life—the need to survive and reproduce. However, I do not want to claim that any possible mind needs causal representations.
David Marr defined a representation as “a formal system for making explicit certain entities or types of information” (1982, 20). I will suggest that causal representations can take multiple forms. These are distinguished by the types of information they make explicit and for what (or for whom) these representations are made available.
Life as model-maker
Why begin with life? Causal representations are generally not associated with life as such, but with reasoning agents in complex environments. However, I believe that the causal representations of animal and human minds are part of a larger set of representations that originate with life’s most basic forms. In Anticipatory Systems (2012), the theoretical biologist Robert Rosen shows the pervasiveness of anticipatory behaviour in organisms. Living beings do not passively react to their environments, but actively attempt to anticipate future states of the environment and/or themselves.
For example, deciduous trees, which shed their foliage at the end of a growing season, need some mechanism by which to anticipate seasonal changes. There are different ways to do this, reflecting Marr’s (1982) point that computational problems can be solved using different algorithms embodied in different physical substrates. Deciduous trees regulate seasonal growth by mechanisms such as sensitivity to photoperiod (day length), temperature, and moisture (Basler & Körner 2014). Bacteria that live in environments of unevenly distributed nutrients need a way to get to the nutrients. Without a cognitive map of their environment and their position in it, bacteria such as E. coli have simpler ways to reach their food. They swim not towards a food location, but in the direction of increasing food, i.e. by following a concentration gradient (Mitchell 2023).
In these cases, the behaviour is geared toward future states of the environment. The means by which this anticipation is accomplished are models, not necessarily complicated ones.[1] An organism engaging in model-based anticipation does not need to have consciousness or understanding. As Rosen points out, in such simple cases, the models are “wired-in” (2012, 7), what we might call hard-wired. While these models do not represent—make explicit—the causal mechanisms and relations that connect temperature and leaf shedding or nutrient gradient and nutrient location, these models must track such causal relations in the organism’s environment.
Whether we call such models causal or not is debatable. They do not represent causal information to an agent, but track causal information as part of a control system. Within these systems, there are no causal interpretations. Examined from the outside, however, we observe that these systems track information that is causally relevant to the functioning of these systems. If a deciduous tree could not tell winter from spring, it would die, as would an unfortunate E. coli with a faulty gradient detector.
The reason I include these models—let’s tentatively call them causality-tracking control models—is that they share some fundamental similarities with the more complex causal models we will examine next. A complex causal model, for example, may function in such a way that when an agent is presented with perceptual input A, it expects outcome B, as a consequence of which it considers doing C. The simple models we have just discussed skip the intermediate stage; given perceptual input A, do B. Both involve if-then conditionals and both are constrained by the need to correctly anticipate species-relevant factors.
Why not stick with basic causality-tracking models? Why did organisms evolve fancy model-enhancing wetware if it is possible to go without it? It has to do with the basic models’ lack of flexibility. A species will go extinct if it encounters a new environment where its old models are no longer helpful. David S. Wilson (2011, pp. 51-53) beautifully captured this situation by saying that such a species is “dancing with ghosts”, and “will continue dancing with ghosts until they go extinct or until natural selection teaches them new dance steps.”
Agents need causal models
In their textbook on artificial intelligence, Russell and Norvig (2020) distinguish between different types of environments along several dimensions, such as degree of observability, determinism, and dynamism. The structure of the environment along these dimensions, they argue, “determines, to a large extent, the appropriate agent design” (p. 43). This approach to artificial intelligence is equally applicable to organic intelligence, and it mirrors the question evolutionary psychologists and biologists ask all the time: what sort of environment made this evolved trait appropriate?
If basic hard-wired models work in some environments, what environments favour the emergence of more complex models and behaviour? Peter Godfrey-Smith’s (2016) account of the transition from the Ediacaran period (635-542 million years ago) to the following Cambrian period offers one possible answer. The Ediacaran environment was a largely peaceful world, where species consumed nutrients by filter feeding or roaming around on the sea floor, with little evidence of mutual interaction. This contrasts sharply with the Cambrian world of animals with eyes, antennae, claws, and shells—tools of predation and protection. This “arms race”, Godfrey-Smith argues, led to a new evolutionary environment: “In the Cambrian, each animal becomes an important part of the environment of others. … From this point on, the mind evolved in response to other minds” (p. 36, emphasis in original).
Similarly, Michael Tomasello (2022, p. 21) argues that it is the uncertainty created by competitive or collaborative interaction between and within species that has driven the evolution of flexible, agentic behaviour. Tomasello’s model of agency is one where the organism not only reacts, but actively directs its attention and actions in the pursuit of a goal. The goal-oriented action is dynamically controlled by feedback, enabling flexibility and adjustment to changing conditions. Tomasello (2022) argues that the agent needs three basic components: first, to represent a goal, second, to perceive the environment relevant to the goal, and third, the means to compare its perception and goal so that it can act towards the goal.
This tells us something about the structure of goal-oriented action, but it does not tell us about how goals are constructed by agents nor the flexibility with which goals are constructed and pursued. This is where we come back to causal representations. What sort of representations enable flexible goal pursuit? To behave flexibly, the organism must be able to pursue a goal by multiple means across changing environments. This requires that the organism represents something about what is relevant to its goal-pursuit that is sufficiently general for it to apply in different circumstances (Tomasello 2014, p. 12). In other words, the organism must have representations of relevant invariants (for a discussion of invariants, see Sloman 2005, ch. 2). It needs a model of the world around it.
Different organisms construct different models (Stevens 2021; von Uexküll 2010), and the specificity of their evolved goals and motivations also vary greatly, from fixed action-patterns to strategic decision-making (de Waal 2016; Rosati 2017). However, in the same way that we humans learn a schematic understanding of how the world hangs together—where things are, what characterizes things, how one event leads to another—other organisms also learn such schema, even though they differ significantly from ours. A functional approach to causation, such as the one developed by Woodward (2021), emphasizes that thinking in causal terms is useful for various goals and purposes. As such, there are several areas of overlap between human causal models and the causal models of other organisms. They both facilitate planning, inference, and manipulation.
Planning involves a series of steps to achieve a goal (Miller et al. 1960). A precondition for effective planning is that the steps of the plan form a causally coherent whole. As Lagnado (2022, p. 19) points out, our mental models must be causal in the sense that our manipulation of them reflects how the world would change if manipulated. Nonhuman primates, for example, can form causally coherent plans in the context of tool use (Hermann et al. 2008). There is also evidence of logical inference in chimpanzees (Tomasello 2014), and tool manipulation in a variety of species such as elephants, crows (de Waal 2016), and octopuses (Godfrey-Smith 2016).
To summarize, organisms living in unpredictable and changing environments benefit from learning models of their environment which can form the basis of agentic and flexible behaviour. The chief advantage of such models are that they enable the organism to adapt—by acting on and updating their models—in real time, whereas hardwired models can only be updated over evolutionary time (Rosen 2012). This is the essence of Kenneth Craik’s (1952) idea of the organism carrying “a ‘small-scale model’ of external reality and of its own possible actions within its head,” which it can use to simulate and evaluate action. Thus, in addition to the causality-tracking control models of the previous section, we have now added causal models, which represent species-relevant causal relations to agents.
Formal causal models
The evolutionary history we share with our primate cousins means that there is considerable overlap in how we and other primates parse and process the world. However, humans do stand out in some respects. As Boyd, Richerson, and Henrich (2013, p. 119) point out:
Humans have a larger geographical and ecological range than any other terrestrial vertebrate. About 60,000 years ago, humans emerged from Africa and rapidly spread across the globe. By about 10,000 years ago, human foragers occupied every terrestrial habitat except Antarctica and a number of remote islands, like Hawaii, Iceland, and Madagascar. To accomplish this unparalleled expansion, humans had to adapt rapidly to a vast range of different environments: hot dry deserts, warm but unproductive forests, and frigid arctic tundra.
What has enabled this expansion? Technology may be a good answer, but that only pushes the question down the line; what has enabled such varied and effective use of technology? There are two complementary accounts that may help answer this question. These are human cognitive and cultural evolution. The hominid brain more than tripled in size over the past few million years (Schoenemann 2006). The timing and causes of what is sometimes called “behavioural modernity,” characterized by abstract thinking, symbolic behaviour, technological innovation, among other things, is a hotly debated topic (Mcbrearty & Brooks 2000; Henshilwood & Marean 2003). The key takeaway is that humans at some point acquired traits for symbolic communication and creative tool use. Given the sheer variety of tools made by humans, it is likely that some new forms of causal representations may have underwritten this productivity. Whether these differ in quality or degree from those of our primate cousins is an open question (see Povinelli 2003).
Alongside this cognitive evolution, there is a process of cultural evolution based on language. Dubbed the “Symbolic Species” by Terrence Deacon (1997), humans have the ability to generate and accumulate symbolic knowledge across generations. Conventionalized language, by its arbitrariness, has also led to abstractness, which enables reference to complex phenomena by a simple sign or vocalization (Tomasello 2014). Language allows humans to flexibly generate and communicate new models. It is crucial to note, however, that these models are by no means necessarily causal models. Indeed, Yuval Noah Harari (2015, p. 27) argues that the “ability to speak about fictions is the most unique feature of Sapiens language.”
While language has been used to mythologize, it has also been used as a tool for systematic investigation. Causality has been among the many topics investigated by philosophers, thinkers, and scientists. The analysis of causality through the creative and abstract power of language has yielded what I call formal causal models. These models seek to describe explicitly causal relations, such as Aristotle’s fourfold account of causality. Later, graphical and mathematical language has been added to enrich our understanding of causality (Pearl & Mackenzie 2018).
These formal causal models are different from the causal models discussed earlier. The formal casual models have not emerged from the pressures of natural evolution, but through cultural evolution. To say that Aristotle needed his formal causal scheme seems misleading. Formal causal models may be viewed as an elaborate outcome of our motivation to seek information. Along these lines, Sharot and Sunstein (2020) suggest that information is sought in part for its “cognitive value”, by which they mean that information is valued if it can enhance people’s understanding of the world around them. Curiosity, however, is something we share with many other organisms (Kidd & Hayden 2015).
Necessary representations?
Having examined causality-tracking models, causal models, and finally, formal causal models, which of these representations does the mind need? In this essay, I have used an evolutionary criterion of need. The only minds with which we are familiar are organic minds sculpted by natural selection, and so the most plausible need for causal representations comes from the demands of survival and reproduction. As these demands vary in different environments, so the causal models of organisms vary. At the most basic level, there are control models, such as reflexes, that track relevant processes in the organisms environment. At the more complex level, there are causal models such as the one employed by octopuses to assemble and disassemble coconut shells as portable shelters (Godfrey-Smith 2016, p. 64). Rosen’s diagram of the “modelling relation” helps to illustrate this function of causal models:
Figure 1. Rosen’s modelling relation (Rosen 2012, p. li)
In this diagram, the natural system is the organism’s environment, while the formal system is the modelling system—the mind—of the organism. By building causal representations, the mind is able to make sense of the world, guide appropriate action, and thereby survive in changing environments. What about the formal causal models, does the mind need those? The formal causal models of philosophers and scientists are noteworthy achievements, but human lives and minds would most likely be able to go on without them.
Discussing whether or not the mind needs causal representations implicitly presupposes that causal representations exist and that they form a coherent type of mental representations. While I do think it is possible to investigate causal representations, I think it is important to note how intertwined such representations are with the other parts of a mind’s conceptual schema. For example, de Waal (2016) presents the case of an elephant that moved, even from great distances, a sturdy box so that it could use it as a ramp on which to stand to reach otherwise unavailable food.
One might view this as the application—not necessarily the conscious use—of the following causal reasoning: if I move object X to location Y, it will help me reach Z. For this plan to be effective, the elephant already needs to know a great deal about the world, such as the properties and affordances of objects, how its own actions affect the environment, and so on. It may not be possible to separate its causal representations from those that represent objects, motor programs, and sensory-motor integration. A better way to look at this may be that causal relations, how objects, actions, events affect one another, are built into the conceptual structure itself (Sloman 2005, ch. 9). In other words, it is not the case that a human or an elephant has concepts, which they then happen to combine in causally coherent ways, but that these concepts are inherently designed to be combined in causally coherent ways.
This, I believe, is the best account of the role of causal representations in our minds. Causal representations are not separate concepts that are sometimes applied, but they concern the ways in which our concepts are linked together. This helps explain why we find it much easier to create a causal explanatory model on the fly than reflecting on the relationship between our causal model and the world (Lagnado 2022). The latter requires turning our attention to the process of constructing causal coherence that we take for granted and do unconsciously (Michotte 2017; Kahneman 2011).
Conclusion
To encapsulate the core argument presented here, one might say that the mind as an evolved organ needs causal representations the same way the heart needs a cardiac muscle. In a sense, this argument mirrors what Godfrey-Smith (1998, p. 3) has called the “environmental complexity thesis”, which states that the function of cognition, of the mind, is “to enable the agent to deal with environmental complexity.”
While I believe this ecological perspective is necessary to understand our and other organic minds, the prospect of inorganic minds also exists. An artificial mind would not be subject to the same constraints as an evolved mind. Unconstrained by the need to track causal relations in its environment, it becomes challenging to speculate on the kind of coherence such an artificial mind might construct. Human thought and planning are constantly engaged in prospection and simulation (Seligman et al. 2013). Given our human goals, we often think in terms of if-then conditional operations. If one abstracts away the evolved goals that give purpose and direction to these operations, do levels of coherence still remain? Does an agent still exist with a will to perform these operations? One possibility is that artificial minds may be created with agency and goals different from those of evolved organisms, and that these minds for that reason conceptualize the world and causality in novel ways. With the emergence of increasingly capable AI systems, the topic of the relationship between an artificial agent’s goals and its conceptual structure appears a promising area of research.
[1] The terms “model” and “representation” are often used interchangeably (e.g., Schenck & Wilson 2020; Hughes 1997). In this essay, they are also used interchangeably.
References
Barsalou, L. W. (2005). Continuity of the conceptual system across species. Trends in Cognitive Sciences, 9, 309– 311.
Basler, D., & Körner, C. (2014). Photoperiod and temperature responses of bud swelling and bud burst in four temperate forest tree species. Tree Physiology, 34(4), 377-388.
Boyd, R., Richerson, P. J., & Henrich, J.. (2013). “The Cultural Evolution of Technology: Facts and Theories”, in Cultural Evolution: Society, Technology, Language, and Religion. Peter J. Richerson, Morten H. Christiansen (eds.). MIT Press.
Chomsky, N. (1965). Aspects of the Theory of Syntax. Cambridge, MA: MIT Press
Craik, K. J. W. (1952). The Nature of Explanation. Cambridge: Cambridge University Press.
De Waal, F. (2016). Are we smart enough to know how smart animals are?. WW Norton & Company.
Deacon, T. W. (1997). The Symbolic Species: The co-evolution of language and the brain (No. 202). WW Norton & Company.
Godfrey-Smith, P. (1998). Complexity and the Function of Mind in Nature. Cambridge University Press.
Godfrey-Smith, P. (2016). Other minds: The octopus, the sea, and the deep origins of consciousness. Farrar, Straus and Giroux.
Harari, Y. N. (2015). Sapiens: A Brief History of Humankind. Vintage.
Henshilwood, C. S., & Marean, C. W. (2003). The origin of modern human behavior: critique of the models and their test implications. Current anthropology, 44(5), 627-651.
Herrmann, E., V. Wobber, and J. Call. 2008. Great apes’ (Pan troglodytes, Pan paniscus, Gorilla gorilla, Pongo pygmaeus) understanding of tool functional properties after limited experience. Journal of Comparative Psychology, 122, 220– 230.
Hughes, R. I. G. (1997). Models and Representation. Philosophy of Science, 64, S325–S336. http://www.jstor.org/stable/188414.
Kahneman, D. (2011). Thinking, Fast and Slow. Penguin Books.
Kidd, C., & Hayden, B. Y. (2015). The psychology and neuroscience of curiosity. Neuron, 88(3), 449-460.
Lagnado, D. A. (2022). Explaining the Evidence: How the Mind Investigates the World. Cambridge University Press.
Macphail, E. M. (1987). The comparative psychology of intelligence. Behavioral and Brain Sciences, 10(4), 645-656.
Marr, D. (1982). Vision: A Computational Investigation into Human Representation and Processing of Visual Information. W. H. Freeman and Company.
McBrearty, S., & Brooks, A. S. (2000). The revolution that wasn’t: a new interpretation of the origin of modern human behavior. Journal of human evolution, 39(5), 453-563.
Michotte, A. (2017). The Perception of Causality. Routledge.
Miller, G. A., Galanter, E., & Pribram, K. H. (1960). Plans and the Structure of Behaviour. Holt, Rinehart and Winston.
Mitchell, K. J. (2023). Free Agents: How Evolution Gave us Free Will. Princeton University Press.
Pearl, J., & Mackenzie, D. (2018). The Book of Why: The New Science of Cause And Effect. Basic Books.
Povinelli, D. (2003). Folk physics for apes: The chimpanzee’s theory of how the world works. Oxford University Press.
Rosati, A. G. (2017). The evolution of primate executive function: From response control to strategic decision-making. In J. Kaas & L. Krubitzer (Eds.), Evolution of Nervous Systems (Vol. 3, pp. 423–437). Amsterdam: Elsevier.
Rosen, R. (2012). Anticipatory Systems: Philosophical, Mathematical, and Methodological Foundations. Springer.
Russell, S. J., & Norvig, P. (2020). Artificial Intelligence: A Modern Approach. Pearson.
Schenck, D. A. & Wilson, P. R. (2020). ‘Models and representations’, Information Modeling: The EXPRESS Way. Oxford Academic.
Schoenemann, P. T. (2006). Evolution of the size and functional areas of the human brain. Annu. Rev. Anthropol., 35, 379-406
Seligman, M. E., Railton, P., Baumeister, R. F., & Sripada, C. (2013). Navigating into the future or driven by the past. Perspectives on psychological science, 8(2), 119-141.
Sharot, T., & Sunstein, C. R. (2020). How people decide what they want to know. Nature Human Behaviour, 4(1), 14-19.
Sloman, S. (2005). Causal models: How people think about the world and its alternatives. Oxford University Press.
Stevens, M. (2021). Secret Worlds: The Extraordinary Senses of Animals. Oxford University Press.
Tomasello, M. (2014). A natural history of human thinking. Harvard University Press.
Tomasello, M. (2022). The Evolution of Agency: Behavioral Organization from Lizards to Humans. MIT Press.
von Uexküll, J. (2010). A Foray into the Worlds of Animals and Humans: with A Theory of Meaning. University of Minnesota Press.
Wilson, D. S. (2007). Evolution For Everyone: How Darwin’s Theory Can Change the Way We Think About Our Lives. Delta.
Woodward, J. (2021). Causation with a Human Face: Normative Theory and Descriptive Psychology. OUP.
Magnus F. Gjerde 