Top-Down Causation (Davies-Walker Framework)
Top-down causation is the mechanism Paul Davies and Sara Imari Walker identified as the hallmark of living systems. In non-living chemistry, causation runs strictly bottom-up: the behavior of the whole is determined by the behavior of its molecular parts. In living systems, the informational architecture of the whole—the genome, the regulatory gene network, the cell's overall organization—exercises causal power over the parts. The genome directs which proteins are synthesized. The regulatory network determines which genes are expressed. The organism's needs shape which metabolic pathways are activated. This bi-directional causal structure is what makes life genuinely novel in the cosmic story: not the chemistry, which can be found in meteorites and interstellar clouds, but the informational architecture, which cannot. Davies and Walker's 2013 paper 'The Algorithmic Origins of Life' formalized this insight and connected it directly to the question of artificial intelligence: current AI systems are predominantly bottom-up, processing data through fixed architectures without the capacity for the informational whole to reshape its own processing rules.
In the AI Story
The concept emerged from Davies and Walker's collaboration at Arizona State University's Beyond Center for Fundamental Concepts in Science, where they brought together physics, biology, and computer science to investigate the transition from non-living chemistry to living systems. Traditional biology explained life through chemistry—specific molecules, specific reactions, specific pathways. Davies and Walker argued this was insufficient. Many non-living systems contain the same molecules. What distinguishes life is not the components but their organization—and specifically, the capacity of that organization to act as a causal agent, directing the components' behavior rather than merely emerging from it.
The paradigmatic example is the genetic code. DNA is a molecule, subject to the same physical laws as any other molecule. But DNA also encodes information—instructions for protein synthesis—and this information exercises downward causal power. The sequence of nucleotides determines which amino acids are assembled into proteins, which proteins determine cellular structure and function, which structure and function determine the organism's capacity to survive and replicate. The information flows down from the code to the chemistry, and this downward flow is what makes the system alive rather than merely complex. Remove the informational architecture and you have the chemistry of life without life itself—amino acids, nucleotides, lipids, all present but not organized into a self-maintaining, self-replicating system.
Davies and Walker's framework has direct implications for artificial intelligence. Current AI architectures—neural networks trained through backpropagation—are predominantly bottom-up. Data flows in at the input layer, transforms through hidden layers according to fixed mathematical operations, and produces output at the final layer. The architecture is fixed. The weights adjust during training, but the topology of the network, the nature of the operations, the flow of information from input to output—all of this is determined at the design stage and does not change in response to what the network learns. There is no equivalent of the genome reshaping cellular machinery, no regulatory network deciding which parts of the architecture to activate. The system processes information but does not use information to reshape its own information-processing structure. This absence may explain why AI excels at pattern recognition and recombination but struggles with the kind of genuinely open-ended creativity that biological systems exhibit.
Origin
The concept has deep roots in systems biology and complexity theory—Stuart Kauffman's autocatalytic sets, Humberto Maturana and Francisco Varela's autopoiesis—but Davies and Walker gave it a precise information-theoretic formulation. Their 2013 paper argued that life is an 'algorithmic phenomenon'—that what makes a system alive is not the stuff it is made of but the informational operations it performs. The paper connected this insight to questions about the origin of life, the possibility of non-carbon-based biology, and the criteria by which artificial systems might be classified as living. The framework has been extended by subsequent researchers into questions about consciousness, agency, and the minimal conditions for top-down informational causation.
Key Ideas
Bi-directional information flow. Living systems are characterized by information flowing both up (from parts to whole) and down (from whole to parts), creating a causal loop that non-living chemistry lacks.
Algorithm, not chemistry. What distinguishes life is not its molecular composition but its informational architecture—the algorithms encoded in DNA and executed by cellular machinery.
Limitation of current AI. Neural networks exhibit bottom-up causation—outputs determined by inputs and weights—but lack the top-down reshaping of their own processing rules that characterizes living intelligence.
Open-ended evolution requires top-down. Genuinely unbounded creativity may depend on the capacity of a system to use information to modify its own information-processing architecture, a capacity current AI has not demonstrated.
Appears in the Orange Pill Cycle
Further reading
- Paul Davies and Sara Imari Walker, 'The Algorithmic Origins of Life,' Journal of the Royal Society Interface 10:79 (2013)
- Stuart Kauffman, The Origins of Order (Oxford University Press, 1993)
- Humberto Maturana and Francisco Varela, Autopoiesis and Cognition (Reidel, 1980)
- Sara Imari Walker, 'Top-Down Causation and the Rise of Information,' Information 5:3 (2014)