Substrate Dependency: Does It Matter What a Computer Is Made Of? In this second post, we explore what may be one of the most fundamental characteristics of biocomputing: substrate dependency.
1. Functional Computing vs. Substrate-Dependent Computing
One of the fundamental principles of conventional digital computing is the separation of software and hardware.
A program can be executed on different processors, provided they support the required computational operations. The physical properties of the hardware are largely abstracted away from the logical computation.
Biological computing is different.
In living neuronal networks, computation depends directly on the physical properties of neurons, synapses and their molecular environment.
Synaptic strengths, membrane potentials, ion-channel dynamics, biochemical processes and network connectivity determine how information is processed.
The physical state of the biological substrate is part of the computation itself.
This has an important consequence: biological computation cannot simply be transferred to another physical substrate in the same way that software can be copied to another computer.
2. Computing Is More Than Processing Information
A conventional computer processes information according to defined computational rules.
A living neuronal network continuously changes its own physical state while processing information.
Learning modifies synaptic connectivity, neuronal excitability and molecular processes. These changes influence future computations.
The computational history is therefore embedded in the physical structure and state of the network.
In substrate-dependent computing, the processor, its memory and its computational history are inseparably connected.
3. Embodiment and Artificial General Intelligence
The relationship between intelligence and the physical world has been discussed extensively in the context of Artificial General Intelligence (AGI).
Ragnar Fjelland (2020), building on arguments by Hubert Dreyfus, questioned whether AGI could be achieved through conventional computing without the physical and experiential relationship between an intelligent system and its environment.
The Importance of Embodiment
The central argument concerns embodiment: intelligence may require more than abstract information processing. It may depend on physical interaction, sensory experience and the ability to act within the real world.
Substrate dependency adds another dimension to this discussion.
A biological neuronal network does not merely receive information about the physical world. It is itself a physical system whose computational behavior emerges from molecular interactions, electrical activity and continuously changing biological states.
The complexity of the physical world is not only represented as information. It participates directly in biological computation.
Whether this distinction is essential for AGI remains an open question.
4. Mortal Computing – When Knowledge Cannot Simply Be Copied
In 2022, Geoffrey Hinton introduced the concept of Mortal Computation.
Conventional digital computing allows software and learned parameters to survive independently of a particular hardware implementation.
Hinton proposed a different approach in which computation exploits the specific physical properties of the hardware itself.
In such systems, learned knowledge becomes closely connected to the individual computational substrate.
When the hardware is destroyed, the learned computational state may be lost with it.
This is what makes the computation mortal.
Living Neuronal Networks as an Example
Living neuronal networks provide a biological example of this principle.
Their learned states are distributed across synaptic connections, cellular properties and molecular processes. They cannot simply be copied and installed in another neuronal network.
This does not mean that knowledge cannot be transferred through learning or communication. It means that transferring the complete physical computational state is fundamentally different from copying digital software.
5. Can Functional Computing Completely Reproduce Biological Computing?
This brings us back to the question raised in our first post.
Functional computing can construct increasingly sophisticated models of neuronal networks.
Such models can reproduce electrical activity, synaptic plasticity, network dynamics and other properties of biological computation.
But a model necessarily involves abstraction from the physical system it describes.
The Fundamental Question
Can a functional model capture every physical property that is relevant to biological computation?
If the answer is yes, substrate-independent computing could, in principle, reproduce the computational functions of biological networks.
If the answer is no, biological substrate dependency represents a fundamental distinction that cannot be eliminated through increasingly sophisticated simulations.
We do not yet know the answer.
6. What Does This Mean for Organics.AI?
With Organics.AI, we are developing a computing architecture based on living neuronal networks.
Rather than simulating every biological process in silicon, we aim to use the computational properties of the biological substrate directly.
This approach may provide opportunities for:
- Continuous learning
- Adaptation
- Temporal information processing
- Energy-efficient computation
Challenges of Biological Computing
However, substrate dependency also introduces significant challenges, including:
- Biological variability
- Limited reproducibility
- Long-term stability
- The transfer of learned information between networks
Biocomputing is not simply software running on biological hardware. The biological substrate itself participates in the computation.
Understanding this principle is essential for developing biological computing architectures.
7. What's Next in the Organics.AI Series?
In our next post, we will discuss one of its most important consequences:
Continuous Learning – What Happens When a Computer Changes While It Computes?
Stay tuned for Part 3 of our Organics.AI series.