In 1981 I was an eager post-doctoral fellow, learning to record place cell’s in Jim Ranck’s lab and beginning to understand John O’Keefe and Lynn Nadel’s “Cognitive Map” theory of the hippocampus. One afternoon, while I had a rat in the maze and watched traces of action potentials sweep by on the oscilloscope, Jim Ranck looked over my should and said …
“This is terrific! Place cells are the gateway to understanding how the brain produces cognition.”1
This was both inspirational and opaque. On the one hand, I understood that our project was new and exciting. On the other hand, I wasn’t sure what the word “cognition” really meant or why hippocampal place cells were the path to understanding cognition. Why the hippocampus and not, for example, the visual cortex?
Looking up conventional definitions of “cognition” was of little help. Most are not definitions; rather, a list of properties or functions. For example, wikipedia says …
“Cognition is a group of mental processes that includes the attention of working memory, producing and comprehending language, learning, reasoning, problem solving, and decision making.”2
Eminently unsatisfying. For decades the issue festered. About a month ago I decided to re-evaluate the question. What follows is a proposed re-definition of “cognition.” I start by separating cognition into two components: pure cognition and embodied cognition. I find this approach helpful in several ways. First, it is a definition, not a list of attributes. Second, it separates the work of the brain into categories that seem helpful. The definitions are followed by a rough-draft of a cognitive theory.
Definition: Cognitive processes are brain-based mental operations capable of operation without feedback from the world3
In other words, cognitive operations are high-level mental operations that do not require immediate sensory input or motor output.
Processes that qualify as “pure cognitive” include imagination, episodic memory, what-if scenarios of imagined behavior, planning, creative artistic processes, day dreaming, dreams, language, math, and hallucinations.
Processes that are not purely cognitive include perception, motor action, and (especially) sensory-motor flow.
In that pure cognitive processes cannot use real-world feedback as a crutch in representation, they are totally dependent on internal representations. Rather than relying on a sensory-motor assistance, a representation of the world in a pure cognitive task must be completely brain-generated. The cost of producing a complete, internal representation results in a large benefit: the internal representation is a model of the world (or self-world relations). Having a virtual world permits playing out imagined action scenarios, without the downsides of real-world timing constraints or failures.
Let’s contrast two navigational problem-solving scenarios. Imagine an animal sitting at a location and attempting to go to a goal location.
- Non-cognitive navigation. The animal can see features of the goal, and begins movement towards the goal. A slight modification of this scenario is if the animal takes a route marked by a series of seen familiar landmarks that ultimately lead to the goal.
- Cognitive Navigation. The animal cannot see the goal, but has a brain-based representation of the layout of the environment. This representation is similar to a map-like overhead view, and does not represent immediate sensory input or anything the animal has ever seen. Using this representation, the animal sits in its current location, imagines the goal location, and computes an optimal route to the goal. Finally, the animal transforms the map representation into its current view of the world and initiates a path towards the goal.
One could argue that the above second example is not pure cognition, that an animal or human may use a combination of internal representation and immediate sensory information to plan an optimal route. To make the cognitive process pure, imagine planning a route with your eyes closed, with or without a map-like representation. Clearly doable. The point is that pure cognitive operations exist and require internal representations.
Pure Cognition Versus Embodied Cognition
The concept of embodied cognition has been laid out by Andy Clarke, Alva Noë and others4. In embodied cognition, the sensory and physical constraints of the world aid an animal (or human) in optimizing behavior. These real-world aids are termed “affordances”.5 Affordances are action aids produced by the environment. They are effective because the environment is organized and predictable. Noë, for example, argues that “consciousness” is not in our heads, it is in the interaction of our brains, ourselves and the world. Note the contrast with “pure cognition” where external support mechanism are not essential; in place of action aids, the brain must substitute a complete representation. My sense is that embodied cognition and pure cognition are complementary operations. A second notion is that during action the brain does not need a complete, internal representation of the environment: the constraints of self-and-environment are part of the representation. JJ Gibson’s cover picture from his 1979 book captures some of the essence. During flight, the bird’s brain need not compute all pieces of sensory data and derive optimal flight instructions. Rather, the birds body and brain interact with simplified patterns, such as optic flow, to maintain smooth flight. Indeed, many flight adjustments are made in bird-wing accommodations to shifting air currents, and are not processed by the brain at all.
Embodied cognition roughly corresponds to “flow”, the almost thoughtless, smooth process of self-world interaction. For animals I think of Gibson’s drawing of a bird’s smooth flight. At the human level I think of a high-level athlete, such as a downhill skier, operating in complete control. Or a highly-trained surgeon smoothly operating on the heart. In the human examples, consciousness is engaged, but only at the highest level. Many critical behavioral elements are highly practiced, efficient, fast and “automatic”. While embodied cognition involves both perception and action, the clearest examples have active body movement.
Rough Draft of a Theory
In contrast, Pure Cognition is characterized by stasis. The body is doing almost nothing. In place of sensory and motor constraints on cognition, frameworks, or representations, must be internally generated. Pure Cognition is thought, absent of immediate sensation or action.
Pure Cognition is useful because it is NOT connected to action. Imagine yourself at a choice-point, a split in the road where there are four possible routes. You have a goal, but you haven’t been at this point on the road before and are unsure which route to take. You could make an arbitrary selection of one of the routes. On the positive side, there is a one-in-four chance that this will work. On the other hand, on the high likelihood of failure, you will return to the choice point and continue making arbitrary choices until you find a working route. Pretty inefficient.
Pure Cognition gives you the option of standing still and thinking about it. If you have the mental capacity to imagine what you will see as you walk along one of the paths, without actually walking on the path, you may, in your mind, see that it does or does not lead to the goal. If it fails, you could continue this process until an optimal path is found. This second mechanism is faster than the first, energy efficient and doesn’t expose you to potential danger. It’s termed vicarious-trial-and-error8. Vicarious (imagined) perception and action is the domain of pure cognitive systems; it is dependent on complete internal representations; although characterized by inaction, it ultimately produces efficient action.
Origins of Pure Cognition
If Pure Cognition has such advantages, do all animals have it? What are the costs? Embodied Cognition is evident in virtually all animals including small-brained creatures. A house fly that jumps and flies quickly to avoid a slap is exhibiting embodied cognition. In contrast, it’s difficult to find examples of Pure Cognition in small brained creatures. Perhaps it requires a large brain. While clearly common in humans, other mammals appear to use pure cognition less frequently. Outside of birds and mammals, clear examples are hard to find. I propose the following, highly simplistic evolutionary scenario.
Pure cognition evolved from embodied cognition. Initially, there are special purpose embodied processes, such as the fly avoiding the slap, fixed and inflexible. These responses, similar to the affordances of a bird wing in flight, do not require a central representation, and may be the product of separate, discrete processors. Later, central representations begin to develop, which permit singular decisions, along with generalized, learned and flexible responses. These early central representations still rely on immediate sensory inputs and behavioral responses. Still later, the central representations reach a capacity where they can operate with, or without sensory input, and with or without, immediate behavioral responses. Our primate ancestors had modest pure cognitive capacity, largely used for cooperative game hunting. As brains evolved and grew, pure cognitive capacity enlarged6. Further, while pure cognition may have originally be restricted to certain domains, perhaps spatial, expanded cognitive capacity released pure cognition from domain specificity. While pure gestural communication may not depend on pure cognition, symbolic language does. If the utterance “dog” represents an unseen dog, producing and understanding the utterance requires pure cognition. The sound is a symbol for the unseen animal. Furthermore, if the sound represents a group of animals, rather than a specific dog (“dog” rather than “Rufus”) an abstraction is being represented.
Cognitive Frameworks. Each cognitive operation, whether embedded or pure, occurs within a framework — or context. An example of a framework is the visual field. An object can be seen in the upper left portion of the visual field if it is real or imagined — as in “imagine an apple in the upper left of your visual field”. Kant proposed that there were two innate frameworks, space and time; termed a priori intuitions. As with cognition, cognitive frameworks can be embedded in the real or abstract. Abstract frameworks, (described briefly in footnote 8) give added power to cognition and are critical to some of the functions listed below. 9
Let’s return to the list of pure cognitive functions, and discuss core functions and specific examples briefly.
- What-if scenarios are the core of problem solving and creativity. Exploring imagining behaviors, in series, for an optimal solution to a problem. I believe this process is the core of all creative acts, and the greatest benefit provided by pure cognition.
- Creativity is produced by the “what-if” scenarios described above. The specifics of creativity derive from the specific associations that produce each imagined behavior. (The nature of associations is a separate topic. The “next association” in a series will be determined, at least in part, by the experience and endowment of the individual.)
- Symbolic representations may be a stretch, but a critical one. The notion: a symbolic representation is laid out in a framework that does not correspond to a real-world framework. For example, a graph is laid out in a virtual space. We have a cognitive capacity for space for behavioral and navigational purposes. Although a graph does not correspond to a behavioral space, the capacity to graph derives emerges from brain structures that use real-world space to construct effective spatial behavior. (perhaps this is why graphs above 3d are ineffective).
- Language is a particular type of symbolic representation. The basic structure of language is a linear, temporal structure, the sentence. The sentence likely derives from “natural” time-lines of behavioral sequences, but can be used for radically enlarged purposes. A symbolic representation requires a virtual structure. Once there is the basic sentence structure, any sorts of objects, unconstrained by real things, can be placed in the structure.
- Mathematical representations (algebraic, graphic) are abstract and symbolic. Algebraic representations are cousins to the sentence, laid out in linear time. Graphical representations are described above.
Specific Functions (built from core functions):
- Imagination. If I “imagine a beach”, no current sensory data nor motor output support my mental image.
- episodic memory The mental image of memory of an event (e.g. first day in school) does not require sensory input or motor output.
- Planning creative artistic projects see “epilog” below.
- Personal Timeline The concept that your life has a beginning (birth), landmarks (episodic memories) and an end (grim reaper).
- Historical Timelines Cosmological, geological, life-on-earth, civilization, biblical, national, etc.
- Day dreaming
Is the Hippocampus a Cognitive Map?
This post started with a quote from Jim Ranck, my post-doctoral advisor. Jim asserted that the study of place cells was the gateway to understanding cognition. The proposition was strong at the time and is stronger today. Although, during normal navigation hippocampal place cell firing reflects the animal’s current environment, the affordances of the current environment are not critical. We now know that place cell firing occurs in the dark, during sleep, during moments at rest, and appears to exhibit the vicarious-trial-and-error processing indicative of pure cognition. Although the process is not finished, I believe Jim was correct. Hippocampal place cell firing is a gateway to understanding the mechanisms of cognition.
I’ll conclude with with an ancient description of pure cognition, as translated by Stephen Mitchell in The Second Book of the Tao:
Ch’ing the master woodcutter carved a bell stand so intricately graceful that all who saw it were astonished. They thought that a god must have made it
The Marquis of Lu asked, “How did your art achieve something of such unearthly beauty?”
“My Lord,” Ch’ing said, ‘I’m just a simple wood-worker—I don’t know anything about art. But here’s what I can tell you. Whenever I begin to carve a bell stand, I concentrate my mind. After three days of meditating, I no longer have any thoughts of praise of blame. After five days, I no longer have any thoughts of success for failure. After seven days I’m not identified with a body. All of my powers are focused on the task; there are no distractions. At that point, I enter the mountain forest. I examine the trees until exactly the right one appears. If I can see a bell stand inside it, the real work is done, and all I have to do is get started. Thus, I harmonize inner and outer. That’s why people think that my work must be superhuman”10
1Jim’s quote is a paraphrase, but close enough. Date is also an estimate.
2Checking for definitions from more reputable people and sources not helpful. For example, William James, Edward Tolman or definitions of cognitive psychology.
3From hereon out, the word “cognition” or the words “pure cognition” will assume this definition.
4Andy Clark Whatever next? Predictive brains, situated agents, and the future of cognitive science
A. Clark and D. Chalmers, “The Extended Mind“ 1998
Alva Noë: Out of our Heads
5Affordances, JJ Gibson, The Ecological Approach to Visual Perception, 1979
6Pure cognition may have been a driving forced for brain enlargement, a consequence (a spandrel), or a combination of the two.
7 O’Keefe and Nadel’s “cognitive map” is derived from Kant, but specialized, a subset of Kant’s spatial a priori. Kant’s space includes visual fields, egocentric space and personal space; O’Keefe and Nadel’s does not; theirs is restricted to “allocentric space”.
8“Vicarious Trial and Error” (VTE) is a term that Muenzinger and Tolman in the 1930s used to describe the rat’s conflict-like behavior before responding to choice. More recently, Redish and colleagues have investigated VTE behavior in parallel with recording hippocampal neurons, and reported that hippocampal neurons fired as if proceeding along imagined paths. Muenzinger KF (1938). (Vicarious trial and error at a point of choice. I. A general survey of its relation to learning efficiency. Journal of Genetic Psychology 53: 75– 86. Redish, A. David (2013-06-19). The Mind within the Brain: How We Make Decisions and How those Decisions Go Wrong (Kindle Locations 8350-8351). Oxford University Press, USA.)
9 Real and Abstract Frameworks. In a manner similar to the cognitive dichotomy described above, I propose that frameworks can be divided into two categories, real and abstract. Initially, the time and space frameworks are “real” in that each can map onto immediate sensory-motor features of the world-and-self. An example of the “real” use of the spatial framework is the current visual field. A real temporal framework is current working memory (the set of events that transpired in the previous few moments). The proposal is that “real” each of the core “real” frameworks can be abstracted for non-real applications. A graph is an abstraction of the spatial framework. A historical timeline, a narrative, or a sentence are abstractions of the temporal framework.
10The Second Book of the Tao, translated by Stephen Mitchell. 2009, Penguin Press, NY.