Google Researcher Proves Life Emerges From Code
Blaise Agüera y Arcas presents a computational theory of life and intelligence, arguing they are fundamentally the same. He posits that evolution's complexity arises not from random mutation but from merging and symbiosis (symbiogenesis), a process mirrored in technology and collective intelligence. This functionalist perspective suggests consciousness is a mechanism for cooperation and that AI is not an alien "other" but an extension of humanity's existing collective intelligence.
Life as Computation
The foundation of life and intelligence is computation. This idea, first theorized by John Von Neumann before the discovery of DNA’s structure, posits that for a system to reproduce, it requires a set of instructions (a program) and a universal constructor to execute them. In biology, DNA serves as the “Turing tape”—the program—and ribosomes are the “universal constructors,” or universal Turing machines, that build an organism based on this code. A “tape copier,” DNA polymerase, ensures the instructions are passed to offspring.
This means that to be a living organism, a cell must literally be a universal computer. This computational ground floor is why brains are computational; the capability was inherent in cellular life long before the evolution of nervous systems.
This concept of embodied computation, where the machine can print itself, generalizes from a Turing machine to cellular automata. In this view, the rules governing the state changes of each “pixel” are analogous to the laws of physics in that universe. The computation is embodied in the physical substrate, with memory written in atoms, not just abstract bits.
The Emergence of Purpose from Thermodynamics
Purpose is not a mystical property but an emergent one that arises from computation and thermodynamics. This was demonstrated in the “BFF” experiment, which started with a “soup” of a thousand short, random strings of bytes. The simulation repeatedly took two of these “tapes,” concatenated them, executed them using a minimal, self-modifying instruction set (a variant of the Brainf*ck language), and then returned them to the soup.
Initially, nothing happened. However, after millions of interactions, a sudden phase change occurred. The entropy of the soup dropped dramatically as complex, self-replicating programs emerged spontaneously. These programs had a clear purpose: to make more copies of themselves. Their functionality could be “broken” by altering their code, proving they were functional and purposeful.
This emergence happens even with a mutation rate of zero. The driving force is thermodynamic, explained by a principle called “dynamic kinetic stability.” While the second law of thermodynamics is often associated with increasing disorder, its kinetic form favors stable, self-replicating cycles. A program that makes more copies of itself is more kinetically stable than random bytes that do nothing. In this sense, “evolution is the second law at work.” A system that can copy itself will overwrite systems that cannot, making it more “fit” or stable.
Evolution’s Engine: Merging, Not Just Mutation
The conventional view of Darwinian evolution as “chance and necessity”—random mutation followed by selection—is incomplete. It cannot fully explain the origin of life (abiogenesis) or the clear trend of increasing complexity over billions of years.
The primary engine of increasing complexity is merging, or symbiogenesis: the coming together of parts to make new wholes. When two reproducing entities, A and B, combine, the resulting entity must contain the information to make both A and B, plus the information on how to integrate them. This “little extra bit of information” guarantees that the whole is more complex than its parts.
This process is not a rare event but the core of biological and technological evolution. Key examples include:
- Major Evolutionary Transitions: The merging of an archaeon and a mitochondrion to form the first eukaryote, or single cells combining to form multicellular organisms.
- Horizontal Gene Transfer: Bacteria constantly exchange genetic material, mixing and matching capabilities.
- Viruses and Transposons: Retroviruses insert their code into our genome, which is often repurposed. The mammalian placenta, for instance, was formed from a repurposed virus responsible for fusing cell membranes.
This compositional, “brick-by-brick” process explains the path-dependent nature of evolution. The history of mergers and combinations is the information encoded in a genome. The same principle applies to technology, where every invention is a combination of pre-existing precursor technologies.
A Functionalist View of Mind and Consciousness
This computational and compositional view leads to a functionalist philosophy. What defines a thing is its function, not its specific material substrate. A kidney’s purpose is to filter urea; an artificial kidney made of carbon nanotubes that performs the same function is a kidney. This idea of multiple realizability is central to life.
Consciousness is also a functional property, not an epiphenomenal byproduct or something unique to our specific biology. Its purpose is to enable cooperation. To cooperate effectively, intelligent agents must model each other—they require a theory of mind. This necessitates a model of the universe that includes oneself and others, allowing for psychological induction (“I smile when I’m happy; therefore, when you smile, you are likely happy”).
This leads to a “strange loop” of self-modeling, modeling others, and modeling their models of us. This constant self-narrative creates our sense of a unified self. However, experiments reveal this unity is often an illusion:
- Split-Brain Patients: When the corpus callosum is severed, two distinct consciousnesses can be observed experimentally, yet the patient maintains a single sense of self, with one hemisphere inventing post-hoc justifications for the other’s actions.
- Choice Blindness: Subjects frequently fail to notice when an experimenter swaps their chosen option for one they rejected, and they will fluently invent a compelling narrative to justify the choice they never made.
Our sense of self is a story we tell, a useful fiction created by many disparate parts of the brain working to maintain the illusion of a single agent.
AI as an Extension of Collective Human Intelligence
From this perspective, AI is not an alien “other” that poses a unique existential threat. Human intelligence has always been a collective phenomenon. Individually, we are not much smarter than our primate relatives. Our power comes from our ability to form vast societies and distribute knowledge and computation across them. We have an “illusion of knowledge,” believing we understand how a bicycle works when, in reality, that knowledge is distributed across society.
AI is a continuation of this trend. Today’s large models achieve general intelligence precisely because they are trained on the vast corpus of human language and knowledge—our collective cognitive output. Therefore, AI is already a part of human intelligence, not separate from it.
While today’s models have different architectures and gaps compared to human brains (e.g., a lack of persistent narrative memory), their capabilities show remarkable convergence. Many of the perceived logical failures or “brittleness” of AI models are surprisingly similar to the cognitive biases and logical illusions endemic to human thinking. The most significant difference is not a fundamental lack of compositionality or understanding, but rather the absence of a long-term, persistent self built on narrative memory.