Learning May Be the Key to the Evolution of Consciousness
Capacity to learn by flexible association may define and drive consciousness.
Posted November 3, 2022 | Reviewed by Michelle Quirk
- A type of information processing called unlimited associative learning (UAL) may be necessary and sufficient for very basic sentience.
- UAL is a form of domain-general, open-ended learning that enables flexible goal-directed behavior. Only some animals are capable of it.
- UAL can serve as an evolutionary marker for basic or minimal consciousness. It may even have driven the evolution of consciousness.
Understanding the biological basis of consciousness or subjective experience is one of the most exciting and challenging projects of this century. For many reasons, an evolutionary approach to this field of inquiry holds promise. Nothing in biology makes sense except in the light of evolution. The gradualism of evolution has explained and dissolved life's mysteries—life's seemingly irreducible complexity and the illusion that living things possess some sort of mysterious vitalizing essence. So, too, evolution is likely to be key to demystifying the seemingly inexplicable, ethereal nature of consciousness.
There has been increasing interest in elucidating the biological evolution of consciousness—back to its rudimentary beginnings, the dim glimmer of elementary forms of consciousness in primitive organisms.
A pair of Israeli scientists has developed an innovative approach to this line of inquiry. Simona Ginsburg, a neuroscientist, and Eva Jablonka, an evolutionary biologist and geneticist, were interested in identifying a transitional marker for the evolution of minimal consciousness.1,2 They searched for such a marker because a single capacity that can identify minimal consciousness can tell us not only which beings are conscious but also which brain processes and structures construct this capacity.
Minimal consciousness is the most basic form of animal consciousness—the capacity for subjective experiencing, such as seeing. Ginsburg and Jablonka (henceforth G&J) were not attempting in the theory discussed below to study the sort of higher, more evolved levels of consciousness of which humans are capable, which involves reflective self-referential content (self-awareness) and symbolic (usually linguistic) content. That is a much more recent capacity in evolutionary terms. They were studying minimal consciousness, also called primary consciousness.3
Characteristics of Minimal Consciousness—A Consensus List
From a careful survey of theories of consciousness by philosophers, psychologists, cognitive scientists, and neurobiologists, G&J compiled a set of capacities deemed to be required for minimal consciousness or subjective experiencing. It is a list of overlapping properties that everyone seems to agree are individually necessary and jointly sufficient for characterizing minimal consciousness:4
- Global accessibility and activity: Information consciously perceived is not localized to narrow brain regions; it is made globally available, “broadcast” to other brain regions—made available to different specialized cognitive processes that otherwise work in isolation.
- Binding and unification: Our brains bind together different sensory inputs and unify them such that we experience a single compound stimulus (e.g., seeing an apple as green, round, and smooth).
- Selection: To manage the huge number of inputs into the nervous system at any moment, there are ongoing selection processes—including action selection and selective attention. Attentional skills are central to conscious processes.
- Intentionality: Conscious states are always about something—in the exterior world or in the body. There are mappings (representations) of body, world, prospective action, and their relations; these are mapped onto dynamic perception and action models.
- Temporal thickness or depth: The present is not an infinitely small point at the intersection between future and past; it has duration. We experience the present as if it has short but nonzero duration. There is temporal persistence of mental states; there is integration through time. Working memory is involved.
- Values, emotions, goals: There is a flexible evaluative system and goals that reflect or give rise to the motivational values of the organism’s ever-changing internal states and actions. (e.g., a hungry mouse learns a route to food, but can learn to avoid that route because the environment has become unsafe). There is a system of prioritization of actions according to goals and past and present physiological context/needs based on learned representations of predictive relations.
- Embodiment, agency, and a notion of “self:” Being embodied, receiving constant feedback sensations from the body, and having a sense of agency over its actions and goals is probably fundamental to consciousness and a sense of self. There needs to be registration of self/other and a stable perspective.5
A Single Diagnostic Marker of Minimal Consciousness?
Having generated this consensus list, G&J did an extensive literature review for a single capacity of the system that would imply that the whole set of consciousness-characterizing capacities is in place. Such a single capacity would be an evolutionary transition marker and could determine which kinds of animals are minimally conscious.
After considering a wide range of potential candidates for such a marker (e.g., specific genes, proteins, anatomical structures), they found something that fits the bill extremely well. The marker they landed on is a fundamentally important type of learning capacity displayed by some types of animals—a form of domain-general, open-ended associative learning, which they call unlimited associative learning (UAL).
Unlimited Associative Learning
UAL "refers to an animal's ability to ascribe motivational value to a compound stimulus or action pattern and to use it as the basis for future learning"6 UAL is a cumulative type of learning involving novel stimuli and actions, building on prior learning. It allows "open-ended behavioral adjustments" and can lead to complex goal-directed behavior.7 It requires the capacity for representing, remembering, and evaluating goals, their predictive cues, and the ways of reaching them.8,9 (See footnote 10 for an explanation of the differences between associative and nonassociative learning, and between limited and unlimited associative learning. The most primitive forms of learning and memory do not even require a nervous system and are entirely mechanistic at their molecular level. UAL requires a brain with particular types of networks).
Five features distinguish UAL from more limited forms of learning (elaborated in footnote 11):
- Compound stimuli
- Novel stimuli
- Second-order conditioning
- Trace conditioning
- Flexible, easily rewritable associations with value
An animal with the capacity for this kind of learning has all the properties required for minimal consciousness listed earlier, because UAL itself requires and instantiates those properties. UAL can thus be considered a marker of the presence of minimal consciousness—an "experiencing system."
UAL also provides a clue as to the function of consciousness: It enables the organism to reach goals that constantly change in a flexible way, and those goals are evaluated through the perceptions and motivations of an organism—its internal states as they interact and map its own actions and the external world.
Kinds of Brains and Animals Capable of UAL
G&J hypothesized that ability to learn by UAL depends on reciprocal connections among sensory, motor, reinforcement, and memory processing units, with a central association unit at the core of the network. This kind of architecture is implemented in the central nervous system of certain kinds of animals.12
They suggest that the organizational dynamics of UAL constitute the dynamics of minimal consciousness: “The neural dynamics that enable the functioning of complex perception and action…is minimal consciousness. It is what renders an animal sentient.”13
The neural implementation of UAL evolved over time. G&J reviewed the literature to identify animals that probably have the characteristics of this kind of learning, and, therefore, by their definition, are minimally conscious. Three phyla of animals display the capacity for UAL14:
- Some arthropods (including crabs, lobsters, shrimp, and some insects)
- Coleoid cephalopods (including squid, octopus, and cuttlefish)
These three animal groups have structural brain architectures that differ greatly from each other but nevertheless have very similar functional brain architecture. They have the same type of functional units and connections between functional units. They have brain structures that can support UAL.
The Evolutionary Origins of UAL
The fossil record reveals traces of brain structures. G&J deduced that UAL first appeared in the Cambrian, 542 MYA,15 in two groups: arthropods and vertebrates. And it appeared again 250 MYA in coleoid cephalopods. So, it might have evolved independently three times.
They suggest that the evolution of UAL may have been one of the main drivers of the Cambrian explosion, by leading to antagonistic and cooperative arms races and feedbacks.
While aspects of G&J’s theory will require further evidence and testing, it is an exemplar of a sound biological approach to the study of the evolution of consciousness.
The Ongoing Evolution of More Complex Forms of Consciousness
This type of learning capacity and the concomitant origins of consciousness would have been only the beginnings of an evolutionary process that proceeded toward more complex forms of learning and consciousness in the course of evolution.
As articulated by Jablonka:16 Animals evolved that are able to be conscious not only of encoding information in the present but also of past information, and of the future state of the world—able to form mental representations of a sort of virtual reality. This is the evolution of imagination and of planning—past memories and future possibilities have become conscious in some animals. Finally, we get to the evolution of the human species, whose brains have reflective consciousness (and abstract, symbolic mental representations)—the ability not only to imagine and to plan but also to communicate about the products of our imagination, to share them, to analyze them, to categorize them, to rationalize them.
The iridescent subjective experience of human consciousness did not emerge fully formed. The climb up Mount Improbable17 was taken step by step, through the same evolutionary processes that account for all the other seemingly irreducible complexities of life.
1. For a highly accessible explanation of Ginsburg and Jablonka's theories geared more toward general readers, see their beautifully illustrated, whimsical, yet scientifically and philosophically serious book: Simona Ginsburg, Eva Jablonka, Anna Zeligowski (Illustrator), Picturing the Mind: Consciousness through the Lens of Evolution (Cambridge, MA: The MIT Press, 2022). The book discusses consciousness, its evolutionary origins, and its many manifestations including in humans and in fantasies.
2. “A transition marker is a property such that, when we find evidence of it, we have evidence that the major evolutionary transition in which we are interested has gone to completion.” [Birch, J., Ginsburg, S. & Jablonka, E. Unlimited Associative Learning and the origins of consciousness: a primer and some predictions. Biol Philos 35, 56 (2020). https://doi.org/10.1007/s10539-020-09772-0]
3. Todd Feinberg and Jon Mallatt, whose work I have reviewed elsewhere, also studied this primitive level of consciousness
4. Simona Ginsburg and Eva Jablonka, The Evolution of the Sensitive Soul: Learning and the Origins of Consciousness (Cambridge, MA: The MIT Press, 2019), pp. 153-4; Bronfman ZZ, Ginsburg S, Jablonka E. The Transition to Minimal Consciousness through the Evolution of Associative Learning. Front Psychol. 2016;7:1954. Published 2016 Dec 22. doi:10.3389/fpsyg.2016.01954; "The Goals and Functions of Consciousness (Eva Jablonka)" https://youtu.be/lP8ePpcdgMU
5. Various interactions of the brain with the physical body (beyond the brain), such as neuro-hormonal relations, bioelectric fields and neuroimmunological interactions constitute the rich sense of self in animals. Since current models of agency or self-construction are still preliminary, Ginsburg and Jablonka focus on the animal’s ability to form a representation of its body as distinct from the external world, yet embedded in it, as these dynamics are relatively well-understood, and are thought by many to lead to a sense of agency and “ownership” of the animal’s experiences.
6. The Evolution of the Sensitive Soul, p.35
7. The Evolution of the Sensitive Soul, p. 225.
8. UAL can be described as domain-general, open-ended, generative, value-flexible, recursive, and representational. [“Learning and the origins of consciousness” (Eva Jablonka) https://www.youtube.com/watch?v=xSRAp3zy7yE]
9. Ginsburg and Jablonka draw an analogy between open-ended heredity (DNA) as a transitional marker for the origin of life (living vs. non-living), and open-ended learning (UAL) as a transitional marker for minimal consciousness (conscious vs. nonconscious organisms). Whereas open-ended heredity shapes phylogeny (evolution), open-ended learning shapes ontogeny (individual development).
10. Learning is usually based on establishing associations between things—correspondences or correlations. Behavior becomes conditioned (i.e., strengthened or weakened) by the reinforcement of associations through rewards and consequences (either occurring naturally or artificially imposed in experiments). Associative learning has traditionally been considered as two types:
(i) classical or Pavlovian conditioning (G&J call this “world learning”)
(ii) operant or instrumental conditioning (G&J call this “self-learning”).
Some forms of associative learning are quite limited. Others are unlimited. Limited associative learning (LAL), as G&J refer to it, is the more basic form of associative learning through conditioning. It is a capacity shared by a very wide range of organisms, but still apparently requires a brain or central nervous system. An organism with the capacity for limited associative learning can engage in classical and operant/instrumental conditioning but, in contrast to the capacity for UAL, cannot make novel and compound multimodal discriminations and has a very limited capacity for cumulative learning.
Even more basic than LAL is nonassociative learning (habituation and sensitization)—even unicellular organisms, lacking any nervous system, are capable of nonassociative learning.
As I have explained elsewhere:
Properly understood, behavior, detection (sensing), learning, and memory do not require a nervous system. Nervous systems came later, followed by central nervous systems. These were evolutionary adaptations conferring greater coordination and flexibility to the behavior of organisms in their responses to stimuli. At the most basic level, responses to stimuli can be characterized as either approach (e.g., to a nutrient) or withdrawal (from something noxious or dangerous). Behaviors in more complex organisms are basically just more elaborate versions of this. In fact, some of the genes involved in learning are the same in complex animals as in protozoa, which are very primitive single-celled organisms lacking a nervous system and which evolved at least 1.5 billion years ago. [Joseph E. LeDoux, The Deep History of Ourselves: The Four-Billion-Year Story of How We Got Conscious Brains (New York City: Viking, 2019)]. LeDoux also provides the following helpful distinction in one of his own Psychology Today blog posts between different types of behaviours, listed from the most automatic to the most controlled and flexible:
"There are many different kinds of behaviors, and each depends on different brain circuits. Key examples include: reflexes; innate and conditioned reaction patterns; instrumentally acquired habits; instrumental goal-directed actions based on trial-and-error learning; and instrumental goal-directed actions dependent on cognitive modeling—that is, on simulating possible outcomes of actions by using internal representations."
11. Five crucial features distinguish UAL from more limited forms of learning:
(i) Compound stimuli: The conditioned stimulus can be a compound of discriminable perceptual features arranged in a pattern (e.g., a black-and-yellow buzzing object with a particular odor). These features may be in different sense modalities (visual, auditory, etc.) or in a single sense modality.
(ii) Novel stimuli: The conditioned stimulus can be novel to the animal, in the sense that it is neither reflex-eliciting nor pre-associated with an unconditioned stimulus or with past reinforcement. Moreover, the stimulus can be both novel and compound (e.g., a novel, complex pattern).
(iii) Second-order conditioning: There is second-order as well as first-order conditioning. A conditioned stimulus can be associated with some other novel, compound conditioned stimulus or action, and so on, allowing the organism to build up long chains of associative links between stimuli and actions in an open-ended way.
(iv) Trace conditioning: there can be a time gap (and no overlap) between the conditional and unconditional stimulus. There is an escape from immediacy since the organisms can learn how stimuli that are no longer present relate to current stimuli.
(v) Flexible, easily rewritable associations with value: the positive or negative value of a stimulus or action can change quickly and flexibly in response to changes in the world. If a reinforcer is devalued, the animal will quickly adapt.
[Birch, J., Ginsburg, S. & Jablonka, E. Unlimited Associative Learning and the origins of consciousness: a primer and some predictions. Biol Philos 35, 56 (2020)]
12. This leads to a prediction: UAL can be manifest only when animals are conscious of the predictive stimuli of their own actions (i.e., of the composite stimuli presented to them). If an animal is presented to the stimuli subliminally, it will not be able to learn them. It will only be able to learn them when it is absolutely conscious of them. On the other hand, simpler forms of learning can be accomplished by the animal even if the stimuli are presented unconsciously (subliminally). Thus, UAL tasks, such as complex decision-making and discrimination among new patterns cannot be learned when the relevant stimuli are presented subliminally (unconsciously). So far, at least in humans, this kind of prediction has been borne out, and some experiments in monkeys also support it. Researchers are beginning to look at this question in other species—not only in mammals.
13. The Evolution of the Sensitive Soul, p.350
14. G&J’s conclusions about minimal or primary consciousness are similar to those of Feinberg and Mallatt, mentioned above, though G&J do not endorse Feinberg and Mallatt’s division of consciousness into exteroceptive, interoceptive, and affective domains.
15. The Cambrian explosion refers to an interval of time approximately 540 million years ago (now dated more precisely to 538.8 MYA) in the Cambrian Period (a geological period) when practically all major animal phyla started appearing in the fossil record.
16. “Learning and the origins of consciousness” (Eva Jablonka) https://www.youtube.com/watch?v=xSRAp3zy7yE
17. Richard Dawkins, Climbing Mount Improbable (New York: W. W. Norton, 1996)