5 minute read

Over tens of thousands of years of evolution, why did humans develop consciousness? What survival advantage does this phenomenon provide?

This post is a reflection on Simona Ginsburg and Eva Jablonka’s “How Did Consciousness Evolve? An Illustrated Guide,” published by MIT Press [1].

Our existence is a product of evolution. While those who believe in creationism often reject evolutionary theory, arguing about the past is futile, as it cannot be falsified (Karl Popper). Evolution, however, is a process that can be tested in the lab, offering strong explanatory and predictive power.

The theory of evolution began with Jean-Baptiste Lamarck and Charles Darwin. Initially, it was about the inheritance of traits and how natural selection favored traits suited for survival. In the 20th century, John Maynard Smith introduced the idea of DNA-based evolutionary diversity, emphasizing genes as the primary drivers. However, in the 21st century, the importance of an organism’s active role in shaping its ecological niche has come to be recognized, diminishing the gene-centric view.


When significant transitions occur in evolutionary history, the lens of evolution offers valuable explanations. For example, why did aquatic organisms move to land, and some even take to the skies, establishing their ecological niches? Similarly, we can ask why single cells formed multicellular clusters or how non-communicative organisms developed symbolic language.

Philosopher Daniel Dennett, a proponent of evolutionary theory, also explores this in his “generate-and-test-tower” framework, categorizing the stages of life’s evolution as follows. The first is creatures selected by the survival game, the second by learning behaviors, the third by imagination and ideas, and the fourth and fifth by culturally selected tools and memes.

  1. Darwinian Creatures: Formed and fixed by random mutations and environmental pressures.
  2. Skinnerian Creatures: Capable of learning by experimenting with behaviors in response to external stimuli.
  3. Popperian Creatures: Able to imagine various scenarios and pre-select the optimal action.
  4. Gregorian Creatures: Able to use tools to create internal environments from the external world.
  5. Scientific Creatures: Capable of communication using social and linguistic tools and employing the scientific method.

From an evolutionary standpoint, how and why did simple carbon compounds become sentient beings capable of experiencing emotions and sensations? Can evolutionary theory help us understand this?


Approaching the origins of life, Source: [2]


Similar questions have been asked about the origins of life. Researchers created a list of abilities that define minimal life and developed theoretical models for the simplest possible life forms. Evolutionary transition markers were used to pinpoint when life emerged [2].

Similarly, the authors propose Unlimited Associative Learning (UAL) as an evolutionary marker for consciousness. This idea traces back to 20th-century learning research. In the 19th century, biologists suggested learnability through behavioral adaptation as a criterion for consciousness, laying the groundwork for 20th-century behaviorism.

Behaviorism in the 20th century defined learning in three ways: (1) sensory stimuli that change the internal state of the system, (2) memories that trigger positive/negative reinforcement, and (3) exposure to the same stimuli again. While associative learning was criticized for being too artificial due to its reliance on controlled experiments, the idea of evaluating learning ability to determine consciousness from the 19th century seems to have revived this behavioral perspective. Moreover, when observing organisms that appear conscious, their vast associative learning abilities could serve as evolutionary markers [2].

Approaching the origins of conscious organisms, Source: [2]


Psychologists, philosophers, and neuroscientists agree on the following characteristics of conscious beings:

  1. Binding/Unification
  2. Global Accessibility and Broadcast
  3. Selective Attention and Active Exclusion
  4. Intentionality
  5. Integration through Time
  6. Flexible Evaluation System and Goals
  7. Agency and Embodiment
  8. A Sense of Self

These traits allow organisms with UAL to continuously learn from the world and their experiences in an unlimited way throughout their lives. Here, “unlimited” refers to four specific characteristics: (1) complex patterns of stimuli and behavior, (2) second-order learning, (3) trace conditioning, and (4) quick and easy re-association of values.


From the perspective that the evolutionary origin of consciousness is UAL, the authors argue that the origin of pain can be explained by overlearning. Fish, for example, have a logic that makes them avoid threats in their environment, even if those threats are not actually dangerous. Similarly, our mental disorders—like anxiety, paranoia, and neurosis—operate on this principle. The authors suggest that animals that are smarter and suffer more are better suited for survival than dull-witted, pain-free creatures. The benefits provided by UAL might outweigh the costs associated with pain. Could it be that emotions are side effects necessary for survival? From this perspective, the Cambrian Explosion, a period of intense survival competition, could be the origin of pain.

Moreover, UAL is not only applicable to Skinnerian Creatures, which have basic learning abilities, but also to Popperian Creatures, which can imagine various scenarios. The Western Scrub Jay, for instance, not only determines the order in which to eat food based on expiration dates but also hides food in optimal locations and returns to consume it within the expiration date. If another creature sees where the food is hidden, the jay will re-hide it elsewhere! They are even known to steal food by secretly observing other jays hiding their caches.

The Western Scrub Jay

The key point of this post is that UAL can be considered an evolutionary marker for consciousness. The authors suggest using UAL as a primary marker to understand why and when consciousness emerged, proposing the Cambrian period as the birth of consciousness.

However, a few questions arise:

While organisms with consciousness have UAL capabilities, can we say that UAL alone implies consciousness?

The authors consider this question and emphasize that UAL is an evolutionary marker, not a developmental one. Therefore, the answer to my question is likely NO! While it provides a useful theoretical foundation for understanding access consciousness, I am somewhat disappointed as someone curious about the mechanisms of phenomenal consciousness and the possibility of implementing artificial consciousness.

How does UAL relate to Yann LeCun’s paradigm of Self-Supervised Learning?

Recently, Yann LeCun argued that most intelligent animals learn through what was traditionally considered Unsupervised Learning, which should be revised to Self-Supervised Learning, structured by cognitive processes. Given that Self-Supervised Learning involves unlimited learning of relationships in incomplete spatiotemporal data, it seems highly relevant to UAL.

Yann LeCun’s Self-Supervised Learning Paradigm

Similarly, Dreamer V3, which achieved SOTA in Model-Based RL, was able to create and use tools within the Minecraft game [4]. Could we consider this Agent a Gregorian Creature?

The following video shows Dreamer V3 sequentially gathering resources and crafting tools to mine diamonds. If UAL is the ultimate marker, could we consider it a Gregorian Creature? Does this model interact with the world through conscious experiences (Qualia)?

References

[1] The MIT Press Reader. (2022, June 29). How Did Consciousness Evolve? An Illustrated Guide. https://thereader.mitpress.mit.edu/how-did-consciousness-evolve-an-illustrated-guide/

[2] Ilpo. (n.d.). Consciousness – An Introduction – Jolyon’s Website. https://www.jolyon.co.uk/illustrations/consciousness-an-introduction/

[3] Self-supervised Learning: The Dark Matter of Intelligence (2021), Yann LeCun. https://ai.facebook.com/blog/self-supervised-learning-the-dark-matter-of-intelligence/

[4] Mastering Diverse Domains through World Models. (n.d.). https://danijar.com/project/dreamerv3/

Comments