Is a biological computer just another implementation of a conventional computer? Starting from Kurt Gödel, we explore why biology may represent a fundamentally different form of computation.
Biocomputing, Gödel and the Limits of Machines
What is biocomputing?
At first sight, the answer seems simple: computation performed by biological systems rather than silicon processors. But this definition may miss the most interesting question.
Is a biological computer simply another implementation of a conventional computer—or does biology represent a fundamentally different form of computation?
Kurt Gödel provides an interesting starting point.
In his 1951 Gibbs lecture, discussing the consequences of his incompleteness theorems, Gödel formulated a remarkable alternative: either the human mind surpasses the power of any finite machine in mathematical reasoning, or there are mathematical problems that are absolutely unsolvable by the human mind.
Gödel therefore did not prove that humans are “better computers.” His argument points toward a deeper question: Are the capabilities of biological intelligence completely describable by a finite formal computational system?
This question becomes particularly interesting when we compare conventional computing with wet biocomputing based on living neuronal networks.
Von Neumann computing
A conventional digital computer is based on a precisely defined architecture.
At its interface to software, computation is described through an Instruction Set Architecture (ISA). Instructions operate on data according to explicitly defined rules. The physical implementation can change dramatically while the logical computation remains essentially the same.
A calculation performed today on one processor can, in principle, be reproduced tomorrow on another processor.
The physical timescale is primarily a question of performance.
The logical result is not supposed to depend on whether an addition takes one nanosecond or one microsecond.
Time affects speed, but normally not the meaning of the computation.
Biological computing is different
A neuronal network cannot easily be separated from its physical dynamics.
Information is represented by electrical activity, synaptic states, network connectivity, biochemical processes and their interaction across multiple timescales.
A spike occurring after 10 milliseconds is not necessarily equivalent to the same spike occurring after 100 milliseconds.
Timing itself carries information.
Synapses change.
Neurons adapt.
Networks reorganize.
Previous activity modifies future computation.
In other words:
In biological computing, time is not merely the clock of computation. Time is part of the computation.
There is another fundamental difference.
Digital computing largely separates processing, memory and communication.
Biological neuronal networks integrate them.
A synapse simultaneously participates in communication, information storage and computation. The network architecture itself contains information. And computation continuously modifies that architecture.
There may therefore be no biological equivalent of an ISA in the conventional sense.
700 million years of development
Neuronal computation was not designed from first principles.
Nervous systems emerged through roughly hundreds of millions of years of evolution.
Evolution did not optimize neurons for matrix multiplication, floating-point precision or deterministic execution of instructions.
It selected biological systems for survival in uncertain and continuously changing environments.
That produced a very different computational system: distributed, adaptive, noisy, massively parallel and deeply embedded in the physical world.
Noise is not necessarily an error.
Variability is not necessarily a defect.
Plasticity is not an additional software function.
They are properties of the computational substrate itself.
Data processing without separating data from the processor
Perhaps this is the most important distinction.
In conventional computing, we usually think of data as something that is provided to a processor.
A biological neuronal network does not make this distinction so clearly.
Its previous inputs have already changed its synapses, excitability and network state.
The history of the system becomes part of the system.
Biocomputing does not simply process data. It physically integrates information into the computing substrate.
This could have profound consequences for learning systems.
Instead of repeatedly moving information between processor and memory and recalculating enormous numbers of parameters, biological networks continuously adapt their physical connectivity and dynamics.
What should we expect from biocomputing?
Probably not a biological replacement for conventional computers.
Silicon computers are extraordinarily good at precise arithmetic, deterministic algorithms, databases and reproducible high-speed computation. There is little reason to grow neurons to calculate an Excel spreadsheet.
The interesting applications may lie somewhere else.
Biocomputing could become particularly relevant where computation requires continuous adaptation, learning from limited information, temporal pattern recognition, operation in uncertain environments and integration of many weak or ambiguous signals.
And perhaps the biggest mistake would be to judge biological computers exclusively by benchmarks developed for silicon computers.
A neuron is a very poor transistor.
But a brain is not a badly designed CPU.
It represents a fundamentally different architecture that emerged under completely different constraints.
This brings us back to Gödel.
The question is not whether neurons can execute the same algorithms as silicon.
The more interesting question is:
Are biological neuronal networks simply another physical implementation of computation as we already understand it—or do they reveal a broader concept of computation?
That question remains open.
And it may ultimately be more important for biocomputing than simply asking how many operations per second a neuronal network can perform.
Join the discussion and follow the series on LinkedIn: