What if intelligence is not a property that suddenly appears with the human brain, but the result of a much older history, linked to the ability of systems to preserve information, transform themselves, anticipate regularities and influence their own future?
This is one of the underlying questions running through What Is Intelligence?, by Blaise Agüera y Arcas. Published by The MIT Press, the book begins with a familiar question but quickly shifts its centre of gravity: rather than simply asking which systems we should consider intelligent — humans, animals, machines or language models — it attempts to reconstruct the conditions that make intelligence possible in the first place.
This shift in perspective requires us to go much further back than we might expect. Studying brains, neurons or nervous systems is not enough. Agüera y Arcas takes the question back to the origins of life and, further still, to processes capable of preserving structure, interacting, changing and generating continuity.
The question is no longer simply “what is intelligent?” It becomes something deeper: how can intelligence emerge from systems that, initially, do not appear to possess it?
Perhaps we have started the story too late
Intelligence is usually described in terms of the brain: a system receives information from its environment, processes it, constructs some form of internal representation, makes decisions and acts. From an evolutionary perspective, however, this account begins too late.
An organism without a brain still faces a fundamental problem: maintaining its organisation in a changing environment. To do so, it must detect changes, regulate internal processes, respond differently to different stimuli, exploit certain conditions and avoid others.
These capacities do not necessarily imply cognition in the usual sense of the term, but they introduce a central question: to what extent does the survival of a system require some form of anticipation?
In Agüera y Arcas’s account, prediction occupies a particularly important place. As he develops throughout the book, intelligence can be understood in relation to a system’s capacity to model regularities, anticipate future states and act in relation to those possible futures.
Figure 1. A possible continuity between simple interacting systems, self-organising structures and systems capable of modelling regularities and anticipating future states.
Before intelligence, continuity
If we want to understand how increasingly complex forms of organisation, regulation and, eventually, intelligence can emerge, we need to go back to the origins of life. There we find molecules interacting, structures forming and dissolving, chemical processes modifying other chemical processes and, over time, systems capable of generating new structures similar to themselves.
At this point, it becomes very easy to introduce teleological language — that is, to attribute a purpose to the process —: to say, for example, that a structure “tries to survive” or that a system “wants to persist”. But such formulations project intentionality onto processes that, initially, can be described simply in terms of their dynamics.
A self-replicating structure does not need to represent itself, anticipate its own disappearance or pursue any goal. It is enough for its dynamics to produce a particular consequence: after the process, there are more instances resembling that structure than there were before.
From this, a fundamental asymmetry appears. A pattern that replicates can increase in frequency; if replication also allows heritable variation, some variants may propagate more effectively than others. At that point, an evolutionary dynamic begins to emerge.
Replication is not life. Replication is a property of a process; evolution is a population-level dynamic. Life remains a more complex category.
Life involves organisation, metabolism, autonomy, regulation, interaction with the environment and evolutionary capacity, among other dimensions. It is therefore important not to conflate replication, evolution and life: the fact that a system can copy itself or participate in an evolutionary dynamic does not, by itself, mean that we should consider it alive.
Association, interdependence and evolution
One of the most important ideas in Agüera y Arcas’s account is symbiogenesis.
Evolution is often explained through a narrative centred on competition, selection and differential survival. This perspective is fundamental, but incomplete. A significant part of biological complexity has also emerged through processes of association, cooperation and integration between systems that were initially independent.
Endosymbiotic theory provides a paradigmatic example. The classic 1967 paper by Lynn Sagan, later known as Lynn Margulis, proposed that organelles such as mitochondria originated from ancient prokaryotic organisms that became incorporated into other cells.
What were initially distinct systems eventually became part of a new evolutionary entity.
How do new levels of organisation emerge?
One possible answer is that they do not always emerge because a single entity internally develops all the necessary capacities. Sometimes they appear when different systems establish such deep interdependence that selection begins to operate on the whole.
At that point, the question of the individual becomes more difficult. Where does one entity end and another begin? What is the unit that persists? What is the unit upon which selection acts?
Intelligence may not be confined to the individual either
We tend to locate intelligence inside the brain, yet much of what we are able to know, remember or calculate depends on systems outside our bodies: language, writing, libraries, scientific institutions, mathematical tools, computers, communication networks and collective memory.
No individual could reconstruct alone most of the knowledge they use in everyday life. A significant part of human cognition therefore depends on structures that are socially and technologically distributed.
This connects with the extended mind hypothesis proposed by Andy Clark and David Chalmers, according to which certain external elements can function as genuine components of cognitive processes.
How far does a cognitive system extend?
This perspective also connects with a central idea in the book: intelligence may be compositional. Intelligent systems can themselves be composed of other systems whose individual capacities are more limited, but whose interaction gives rise to new properties.
When does a tool become part of the thinking system?
When we write a piece of information in a notebook, we externalise part of our memory. When we use a calculator, we delegate an operation. When we consult a library, we incorporate into our cognitive process knowledge produced by other people and accumulated over time.
When we use a computer, a search engine or a database, we expand our capacity to retrieve and process information. And when we use AI to reorganise information, explore hypotheses or generate alternatives, we introduce a new layer of processing into the system with which we are working.
And then artificial intelligence appears
It is significant that What Is Intelligence? is not primarily structured as a book about the future of AI. Artificial intelligence functions instead as an epistemological instrument: a technology that forces us to reconsider categories we had treated as settled.
A central part of the training of large language models consists of an apparently modest task: predicting which element is likely to come next in a sequence. Yet when this task is scaled across vast quantities of data and architectures capable of capturing complex regularities, systems emerge that can translate, programme, summarise, connect concepts and participate in problem-solving processes.
Agüera y Arcas himself places the relationship between prediction, evolution and intelligence at the centre of the book’s broader argument.
To what extent does the ability to predict force a system to construct structured representations of the world?
Perhaps there are no clean boundaries
Matter, life, intelligence, consciousness, humans and machines remain useful categories, but they may be more porous than a rigid classification suggests.
Between chemistry and life there may be self-replicating systems; between simple response and cognition, different degrees of modelling and regulation; between the individual and society, language, culture and distributed memory; and between humans and technology, tools that alter what we are able to perceive, remember, calculate and project.
Perhaps we are not looking at a series of absolute leaps. Perhaps we are looking at a set of transitions.
Questions arising from the book
- If replication can emerge without intention, at what point does agency appear?
- If evolutionary dynamics can exist in computational substrates, to what extent can we speak of evolution without yet speaking of life?
- If symbiogenesis can generate new units from previously independent entities, what makes a collection of systems begin to function as a new evolutionary unit?
- If a significant part of human cognition depends on distributed structures, where does the individual end and the cognitive system begin?
- If humans and AI form increasingly interdependent environments for work and decision-making, are we still talking about tools, or are we configuring new cognitive systems?
What AI reveals about intelligence
For decades, we have tried to build intelligent systems, but perhaps one of the most significant outcomes of that effort is something else: in attempting to reproduce intelligence, we have been forced to reconsider what we actually mean by intelligence.
What Is Intelligence? places it within a much broader history of organisation, information, evolution, prediction and cooperation. From this perspective, intelligence does not necessarily appear as a sudden property exclusive to one particular kind of system, but as a possible outcome of processes that accumulate structure, memory, modelling capacity and new forms of interaction.
This shifts the original question once again. Perhaps we should not ask only what intelligence is, but also what conditions make it possible for intelligence to emerge, transform and become integrated into increasingly complex systems.
And if this history began long before us, what makes us think that we are its endpoint?
Two ideas to keep thinking about
Replication does not imply a desire to persist.
Persistence can be a consequence of the dynamics, not an intention of the system.
Where does the individual end?
The question is biological, cognitive and, potentially, technological.
References
Agüera y Arcas, B. (2025). What is intelligence? Lessons from AI about evolution, computing, and minds. The MIT Press. https://mitpress.mit.edu/9780262049955/what-is-intelligence/
Agüera y Arcas, B., Alakuijala, J., Evans, J., Laurie, B., Mordvintsev, A., Niklasson, E., Randazzo, E., & Versari, L. (2024). Computational life: How well-formed, self-replicating programs emerge from simple interaction [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2406.19108
Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19. https://doi.org/10.1093/analys/58.1.7
Knierim, C., Versari, L., Obryk, R., Agüera y Arcas, B., & Saurous, R. A. (2026). BFF: Simple explanations for complex phenomena [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2607.01483
Sagan, L. (1967). On the origin of mitosing cells. Journal of Theoretical Biology, 14(3), 225–274. https://doi.org/10.1016/0022-5193(67)90079-3
