The origin of consciousness

A new method for determining how and when consciousness evolved

The origin of consciousness was a world-defining event, comparable only with the origin of life itself. But the moment that consciousness emerged is buried deep in the evolutionary record and is hard to identify.

In this article, Simona Ginsburg and Eva Jablonka put forward their new theory about how and when consciousness evolved. They identify a unique marker of minimal consciousness that they believe drove the Cambrian explosion of biodiversity and answers the age-old question of which organisms are conscious.


MuZero is an algorithm with a superhuman ability to learn: it has learned to play 57 different Atari video games as well as Chess, Go and Shogi, and defeated the greatest human masters in every one of them. Yet, this amazing algorithm and the computer in which it is implemented are as conscious as your washing machine. Its “intelligence”, manifest in its learning ability, has nothing to do with consciousness – the ability to feel, perceive and think in the deeply subjective sense that we cherish. If you were told that you would become deprived of all subjective perceptions and feelings, you would be devastated and consider such a life to be meaningless. Intelligence – having the ability to learn and solve complex problems like MuZero does –  and consciousness – being the subject of experience – seem to be unrelated.

But are intelligence and consciousness really unrelated? Most people have the strong intuition that clever animals like chimpanzees, dolphins, elephants and dogs are conscious, whereas they are less sure about animals like sea anemones, worms and slugs that show only very simple forms of learning.

web1Why consciousness cannot have evolved SUGGESTED READING Consciousness Cannot Have Evolved By Bernardo Kastrup In the 19th century George John Romanes, an ardent follower of Charles Darwin, articulated this intuition. He interpreted animal psychology within the Darwinian evolutionary framework and defined mind (which he and others of his time used as a synonym for consciousness) in terms of the ability to make learning-based choices:

The criterion of mind, therefore, which I propose, […] is as follows: – Does the organism learn to make new adjustments, or to modify old ones, in accordance with the results of its own individual experience?” (Romanes 1883).

Romanes thought that most animals were conscious, although the consciousness of the animals with the simplest learning capacity was correspondingly simple. Mind, and the intelligence that marked it, evolved gradually from very simple and fuzzy to very complex. He assumed that the more similar animal intelligence is to human intelligence the more developed its consciousness.

However, as the MuZero example shows, intelligence and consciousness can be decoupled. Do we therefore have to conclude that there is no deep relationship between them and forget about learning as a key to consciousness? We believe that this would amount to throwing out the baby with the bathwater. Although learning is not sufficient for consciousness, we have reached the conclusion that the evolution of learning drove the evolution of consciousness and that the cognitive architecture of complex learning in living organisms constitutes basic consciousness.


The evolution of learning drove the evolution of consciousness and the cognitive architecture of complex learning in living organisms constitutes basic consciousness.


How can we study the evolutionary transition to consciousness?

The possibility that the evolution of learning led to the emergence of consciousness was not initially on our research radar, though our approach to the study of consciousness, has been evolutionary. Not only is evolutionary theory the most powerful theory in biology, but an approach that focuses on the evolutionary origins of consciousness has many specific advantages. For example, identifying the earliest forms of consciousness may offer an answer to the “who is conscious” question, tell us when, and under what ecological conditions consciousness first appeared, what its functions might have been, whether it evolved once or more than once, how it evolved into new, richer forms and what future forms it may take.

Our quest for the evolutionary origins of consciousness was inspired by the work of the Hungarian theoretical chemist Tibor Gánti, who was interested in the nature and origins of life. We used his methodology for investigating the nature of life to study consciousness. Gánti started by characterizing minimal life: he compiled a list of properties that most scientists would agree are sufficient to characterize a very simple (minimal) living being. These included: maintenance of a boundary, metabolism, stability, information storage, regulation of the internal milieu, growth, reproduction, and irreversible disintegration. According to his reasoning, if someone found an entity manifesting all these capacities on Mars, they would be very excited and agree that it is, very probably, a living entity.

Gánti identified a single diagnostic capacity that, when present, requires that all the eight capacities he listed (which suffice for inferring that a system is living) are in place. The diagnostic capacity that marks the presence of a living system was, he proposed, unlimited heredity. This is the capacity to form lineages that vary in open-ended ways from the initial system, so the number of possible different variants is vast. If we find on Mars (or anywhere else in the universe) an entity with the capacity for unlimited heredity, we should be able to re-construct or reverse-engineer on the basis of this capacity the simplest system with all the properties that characterize a living system. On planet Earth it would be something like a proto-cell.

Origins of consciousness 1 

Figure 1: Reconstructing a minimally living system (a protocell) from an unlimited inheritance system (a DNA-like molecule) 

We generalized Gánti’s idea, and called a diagnostic capacity that marks the completion of an evolutionary transition to a new mode of being (such as the transition of life, to consciousness or to rational reflection) an evolutionary transition marker.  A single, tractable diagnostic marker would make it possible to identify the simplest evolved living or conscious system, reconstruct the processes and structures that underlie it, and figure out how they interact. If the evolution of the marker can be followed, we can discover when and how the mode of being that it marks originated.

Origins of consciousness 2 

Figure 2: A transition marker is a capacity that requires that all the attributes that characterize a mode of being, such as life or consciousness, are in place.


Characterizing minimal consciousness and its evolutionary transition marker  

We applied Gánti’s methodology to the study of minimal consciousness. We started by extracting from the literature a list of characteristics that most consciousness researchers would agree are jointly sufficient for the simplest conceivable system to be deemed conscious. These partially overlapping characteristics are:

• Perceiving a composite object or action as an integrated whole with parts that can be discerned and discriminated. For example, a peahen can discriminate among peacocks that differ subtly in their color patterns.  

• Bringing together information from different cognitive systems (sensory, motor, memory, and value systems) so that comparisons, discriminations, generalizations and evaluations that inform decision-making can occur: perceiving the python under your bed as long, large, brown, frightening and requiring immediate retreat, is an example.

• Holding on to incoming information long enough for it to be integrated and evaluated, so the present can be said to have duration, being continuous with the immediate past and immediate future.  

• Flexibly evaluating perceptions and actions as rewarding or punishing. Values can be changed and differently prioritized when conditions change (for example, sources of danger can become sources of pleasure) and new goals are pursued.

• Selectively excluding or amplifying signals resulting from the effects of changes in the body or in the world according to evaluations based on present and past experience, such as the formation of search images for salient percepts and perceptual blindness to (uninformative) noise.

• Mapping signals from the world and body, and their relations. The maps or representations are constantly updated. The spatial maps constructed by birds and bats are an example.

•  Representations must be implemented in a body that is not just reactive but has spontaneous intrinsic activity leading to variations that can be harnessed by selection processes. Exploratory activity enables controlled learning about both the world and one’s own activities.

•  Having a stable perspective from which the system can construct models of the world and body and respond to changes in them. Such a system is able to distinguish between the effects of a stimulus that is the result of its own activities and the effects of an identical stimulus that is independent of its own activities. (Think of the effect of being tickled by another person and tickling oneself).

Wherever in the universe an entity that has all these capacities is found, scientists would take the possibility that it is conscious very seriously.


Wherever in the universe an entity that has all these capacities is found, scientists would take the possibility that it is conscious very seriously.

The next step we took was search for an evolutionary transition marker that requires that all the characteristics we listed are in place. We looked at genes, proteins, anatomical brain regions and neurophysiological processes, but none of the many possibilities we examined entailed all the characteristic of consciousness.  After a year of searching we found a promising marker: a capacity for open-ended associative learning, which we called unlimited associative learning.

Unlimited associative learning (UAL) refers to an organism’s ability to:

(i)  Discriminate among differently organized, novel, multi-featured patterns of sensory stimuli and select among new compound patterns of action. For example, between more and less promising sources of food or mates.

(ii)  Learn about a predictive, composite neutral stimulus or action even when there is a time gap between the presentation of the composite stimulus or action and its reinforcement. For example, learning that a sound-pattern predicts danger, even if danger appears few seconds after the sound subsided.   

(iii)  Alter the value attributed to patterns of sensory stimuli and motor actions when conditions change. For example, learning that a shelter has become a trap when a new predator appeared.

(iv)  Use previously learned stimuli and actions as a basis for future learning. For example, birds learn that fire predicts tasty escaping insects and that smoke predict fire (and hence escaping tasty insects).

Unlimited associative learning (UAL) requires that all of the characteristics of minimal consciousness in the list we compiled are in place: it requires that the system has an ability to integrate spatial and temporal information; map relations between objects and actions from a stable point of view; discriminate between different patterns of percepts and different actions; generalize and flexibly transfer what it learned from one domain to another; direct, shift and sustain attention and assign changing values to different models of the body and the world flexibly and rapidly. UAL is an enormously rich and generative learning capacity: the number of associations that can be formed within and between different sensory stimuli during the individual’s lifetime is vast, as is the number of patterns of action that can be linked to them.

Experimental evidence (at present mostly based on human studies) supports our claim that UAL and consciousness are intimately related. UAL can be manifest only when organisms are conscious of the composite stimuli presented to them. UAL tasks, such as complex decision-making and discrimination among new patterns cannot be learned when the relevant stimuli are presented subliminally (unconsciously), while simpler learning tasks can be learned when the stimuli are presented unconsciously. 

The consciousness explosion: ancient origins and dramatic effects

If one accepts UAL as an evolutionary marker of minimal consciousness, one can begin to trace its evolutionary origins, the ecological context in which it evolved, and its evolutionary effects. Unlike consciousness, UAL is a tractable cognitive capacity, which has observable behavioral manifestations and which requires supporting brain structures. Both UAL and the brain anatomy supporting it can be studied in living present day organisms, while in well-preserved fossils the brain structures supporting UAL can be identified. What, then, does comparative psychology, anatomy and paleontology tell us about the distribution of UAL and its evolution?   

Our survey of the vast (yet very patchy) learning literature of the last 100 years revealed no evidence of UAL in most animal groups, including medusa, flat worms and slugs. It has, so far, been found only in three groups: most of the vertebrates (fish, amphibians, reptiles, birds and mammals), some of the arthropods (e.g., crabs, bees, crickets, cockroaches) and some mollusks (the cephalopod – squid, cuttlefish and octopus).. We discovered that although the brains of these animals are anatomically very different, they have similar functional units that generate models of the world, the body, and prospective actions, a memory system that can store composite representations, and an integrating and flexible system that evaluates and updates them. This cognitive architecture gives us a clue to the function of consciousness: it enables the organism to make context-dependent decisions that are based on its subjectively-experienced perceptions and motivations. In the words of William James, who laid the foundations of consciousness studies, consciousness is “a fighter for ends, of which many, but for its presence, would not be ends at all” (James, 1890, James’s emphasis).

Once we know which brain structures can support UAL we could begin to find out when it evolved. The fossil record told us that in arthropods and vertebrates brain structures that could support UAL and consciousness first appeared during the Cambrian era, a geologically short period beginning 542 million years ago (MYA) and ending 485 MYA. This era is aptly called the Cambrian explosion, because it was during this period that almost all currently existing animal phyla originated and diversified. The cephalopods mollusks appeared in the fossil record 250 million years later, so UAL and consciousness seem to have originated more than once and the first origins of UAL are very ancient indeed.


Animal evolution ever since has been guided and driven by the perceptions, motivations, aversions, appetites and choices of learning, conscious animals.

Unlimited associative learning was an adaptive strategy that dramatically expanded the ability of animals to learn to exploit new environmental resources during their own lifetime. So was it one of the engines that drove the Cambrian explosion? We believe that it was. The abilities of associatively-learning animals made them more effective predators, more discriminating mates, and more evasive prey. They exerted enormous selection pressure on interacting species, leading to co-evolutionary arms races – to the evolution of adaptations and counter-adaptations. This, we believe, fueled the rapid, explosive, adaptive Cambrian diversification. Animal evolution ever since has been guided and driven by the perceptions, motivations, aversions, appetites and choices of learning, conscious animals. The intricate, sensory, motor and social patterns such as the seductive adornments of male peacocks, the haunting songs of nightingales and whales, the elaborate patterns of warning, attracting and camouflaging of insects and the rich colors and smells of flowers would not exist was it not for the ability of learning, conscious animals to discriminate among mates, cooperators, competitors and prey and predator species. Life would be much impoverished without consciousness.

Origins of consciousness 3

Figure 3: Would the beautiful patterns on the body of the male fish have evolved were it not for the female’s ability to discriminate and choose among subtly different male adornments?

Once in place, consciousness and learning evolved further. Imaginative, planning animals, like great apes, elephants and crows, which are not only conscious of the actual present reality but also of the virtual realities of the past and the future –  appeared. And in the human lineage, a powerful capacity for symbolic communication and representation evolved, leading to further reality-expansions through the creation of artefacts and learning algorithms like MuZero.       





Acknowledgement: We thank Marion Lamb for her critical and constructive comments.

Further Reading

Readers who want to delve deeper into our ideas can read The Evolution of the Sensitive Soul: Learning and the Origins of Consciousness by Simona Ginsburg and Eva Jablonka MIT Press, 2019. The citation by Romanes is from: Romanes, G.J. (1883). Mental Evolution in Animals, with a Posthumous Essay on Instinct by Charles Darwin. London: Kegan Paul, Trench & Co, pp. 20−21. The citation by James is from: James, W. (1890). The Principles of Psychology. New York: Dover Publications, vol 1, p. 141.

Pictures 1 and 2 are taken from an art-science book Picturing the Mind: Consciousness through the Lens of Evolution (due in February 2022, MIT Press; Simona Ginsburg and Eva Jablonka (Texts); Anna Zeligowski: Art), with permission of MIT press. All three pictures were drawn by Anna Zeligowski. 

Latest Releases
Join the conversation

James Cross 21 August 2021

The unanswered question still is what is it about consciousness that makes it a requirement for UAL.

I might suggest it likely that it might be that consciousness is most economical way from an energy standpoint for living organisms to perform the computations involved in learning.