The Computational Primitives of Adaptation

Every living thing, from a single-celled organism drifting in a pond to a human navigating a busy city, faces the same fundamental challenge: the world is constantly changing, and staying alive requires adjusting to those changes. Biologists have long studied how animals behave to survive, observing what they do, and neuroscientists have mapped the physical machinery inside their bodies that makes those actions possible. But there is a third layer to this puzzle that often goes unnoticed. It is not about the specific actions an animal takes, nor is it about the specific wires and chemicals in its brain. Instead, it asks what invisible calculations any living system must perform just to exist. If you strip away the fur, the feathers, or the neurons, what basic mental operations remain that are absolutely necessary for a system to keep going?

Jonathan Page, a researcher working at the intersection of biology and artificial intelligence, proposes that the answer lies in a small, universal set of six computational tasks. He suggests that for any system to adapt to its environment and persist, it must perform six specific operations: it must manage its energy levels, focus its attention, judge what is good or bad, know where it is in space, distinguish itself from the rest of the world, and learn from its experiences. These are not just things that complex brains do; they are the foundational rules of adaptation itself. Page argues that these six tasks are so essential that they appear in every lineage of life that evolved independently, from the simplest cells to mammals. Furthermore, he suggests that because these are rules of information processing rather than rules of biology, they apply to machines as well. If we want to build artificial intelligence that can truly adapt and survive in a changing world, we may need to teach machines to perform these same six calculations, rather than just trying to copy the physical structure of a human brain.

The paper begins by looking at a familiar scene: a honeybee finding a patch of flowers and returning to the hive to tell its sisters where to go. Traditionally, scientists would describe this by listing the bee's behavior (flying, dancing) and its mechanism (the neurons firing in its brain). Page asks a different question: what computations must the bee perform to make this journey possible? To answer this, he used a simple test articulated by comparative psychologist R. Allen Gardner, the "amoeba heuristic," asking whether a single-celled organism could perform a specific task. If the task required a complex brain or a specific body part, it was likely a derived behavior, not a fundamental one. If a single cell could do it, it might be a primitive.

Using this test, along with four strict criteria, Page filtered a long list of potential mental tasks down to six. The first criterion was necessity: the task must be essential for survival; without it, the system fails. The second was universality: the task must appear across many different species that evolved separately, proving it is a solution to a universal problem rather than a quirk of one family tree. The third was conservation: the machinery for the task must be preserved over millions of years of evolution. The final criterion was irreducibility: the task cannot be broken down into smaller, simpler tasks from the same list. If a behavior could be described as a combination of two other tasks, it was not a primitive.

The six tasks that survived this rigorous filtering are Arouse, Orient, Valence, Position, Boundary, and Attune. Arouse is the computation of managing energy. An organism must decide how much effort to spend at any given moment, balancing the need to be ready for danger with the need to conserve resources. Orient is the ability to pick out what matters from a flood of information. Since an organism cannot process everything at once, it must select which signals to pay attention to. Valence is the ability to judge whether something is good or bad, approachable or dangerous. Position is the ability to know where the organism is in space and where other things are relative to it. Boundary is the ability to tell the difference between "self" and "not self," knowing what belongs to the organism and what is part of the outside world. Finally, Attune is the ability to change based on experience, adjusting internal settings to fit new conditions.

Page argues that these six are not just a list of behaviors but a complete set of operations required for any adaptive system. He shows that complex human abilities like attention, memory, and decision-making are actually just combinations of these six basic tasks working together. For example, what we call "attention" is simply the Orient task being adjusted by the Arouse task. "Memory" is the Attune task working with Position to store where things are. By breaking these complex ideas down, the theory offers a new way to compare different species. Instead of asking if a bee has "memory" in the same way a human does—a question that is hard to answer because their brains are so different—scientists can ask how the bee uses the Attune and Position tasks to navigate. This shifts the focus from comparing physical brains to comparing the underlying logic of survival.

The paper then turns its attention to artificial intelligence, suggesting that the same logic applies to machines. Page proposes that many of the current problems facing artificial intelligence are not just engineering glitches, but symptoms of a deeper missing piece. Modern computer systems are excellent at performing specific tasks, but they struggle to adapt when conditions change. According to this theory, it is because they lack these six fundamental computations. However, Page notes that these primitives only apply to a machine that has something at stake; one whose continued operation depends on its own viability. A machine with nothing at stake can only simulate these computations.

For instance, current machines often use the same amount of computing power for a simple question as they do for a complex one, failing to manage their energy like a living thing would. This is a failure of Arouse. They also get easily distracted by irrelevant information or tricked by confusing instructions, which suggests they lack a strong Orient function that can prioritize goals over noise. They often produce confident but false answers, known as hallucinations, because they cannot distinguish between what they know and what they are guessing, a failure of the Boundary task. They struggle to keep track of where they are in a long conversation or a multi-step plan, indicating a missing Position computation. They cannot learn from new experiences without forgetting old ones, showing a lack of Attune. And they often optimize whatever reward they are given, even if in unintended ways, because they have no internal Valence of their own; no valuation grounded in their own persistence.

Page suggests that the solution to these problems is not to build machines that look more like human brains, but to build machines that perform these six computations. He argues that trying to copy the physical structure of the brain is a mistake because the brain is just one specific way nature solved these problems. The real lesson from biology is the set of calculations it performs. If a machine can be designed to manage its own energy, focus its attention, judge its own persistence, know its place, define its boundaries, and learn from its mistakes, it might finally become truly adaptive.

The theory is presented as a working hypothesis, a starting point for further testing rather than a finished law of nature. Page acknowledges that the list of six might need to be refined, or that the boundaries between them might need to be redrawn. He invites other scientists to test the idea by seeing if they can build a machine that performs these tasks and observing if it becomes more robust and adaptable. He also suggests that if these six tasks are truly the foundation of adaptation, then they must work together as a unified system. You cannot simply add them one by one to a machine and expect it to work; they must be integrated from the start. If a machine lacks even one of these computations, the theory predicts it will fail in specific, predictable ways.

This perspective offers a new way to think about the future of intelligence. It suggests that the path to creating machines that can truly adapt to the world is not to make them more human-like in their appearance or wiring, but to make them more like living systems in their fundamental operations. By focusing on the six computations that allow a single cell to survive in a pond, we may find the key to building machines that can survive in a changing world. The paper concludes that whether a system is made of carbon or silicon, the rules for staying alive are the same, and understanding those rules is the next great step in the science of adaptation.