From Biology to Computation
In the year 2000, Toshiyuki Nakagaki placed Physarum polycephalum in a maze with food at both ends. The slime mold explored every corridor, then retracted from dead ends, leaving behind only the shortest path connecting the two food sources. That simple experiment triggered a revolution in bio-inspired computing.
What made this result so striking was the implication: an organism without neurons, without any central processing unit, was solving a problem that normally requires graph search algorithms. The slime mold did not need to "think." The solution emerged from the physics of its own body, specifically the way cytoplasm flows through its network of veins.
To understand the biological basis of this behavior, see our pages on slime mold intelligence and how slime mold moves.
How Slime Mold Computes
The computational power of Physarum comes from a process called cytoplasmic streaming, also known as shuttle flow. The plasmodium is a network of tubes filled with cytoplasm. Rhythmic contractions push cytoplasm back and forth, and the flow patterns carry information about the environment.
Here is how the computation works in practice:
- Exploration: the plasmodium extends pseudopods in all directions, covering available space
- Feedback: tubes that carry more flow (because they connect food sources via shorter paths) grow thicker, while underused tubes shrink
- Pruning: tubes that carry almost no flow eventually disappear
- Convergence: what remains is an optimized network connecting all food sources
This process is a physical implementation of a positive feedback loop: success reinforces itself, failure eliminates itself. It is strikingly similar to how ant colonies optimize foraging paths through pheromone trails, or how neural networks strengthen useful connections during learning.
The Physarum Solver Algorithm
In 2007, Atsushi Tero, Ryo Kobayashi, and Toshiyuki Nakagaki published a mathematical model that captures the essence of Physarum network optimization. This model, known as the Physarum Solver, translates the biological process into equations that a computer can run.
How the Physarum Solver works
The algorithm models the slime mold network as a graph where:
- Each edge has a conductance value (analogous to tube thickness)
- Flow through each edge follows Kirchhoff's laws (like electrical circuits)
- Conductance increases for edges that carry high flow and decreases for edges with low flow
- The system is iterated until it reaches a stable state
The mathematical update rule is:
| Variable | Meaning | Update Rule |
|---|---|---|
| Dij | Conductance of edge between nodes i and j | Increases proportionally to flow magnitude |
| Qij | Flow through edge ij | Calculated from pressure differences and conductance |
| pi | Pressure at node i | Determined by flow conservation (Kirchhoff's current law) |
Over multiple iterations, the conductance of edges on the shortest path grows, while other edges shrink to zero. The result is the shortest path (or near-shortest path) between source and sink nodes.
Proven mathematical properties
Remarkably, the Physarum Solver has been mathematically proven to converge to the shortest path in a graph. A 2013 paper by Bonifaci, Mehlhorn, and Varma provided a rigorous proof that the algorithm finds the optimal solution for the shortest path problem. This was a significant result because it showed that the biological system is not just approximately good but provably optimal for this class of problems.
Andrew Adamatzky's Unconventional Computing
No discussion of slime mold computing is complete without Andrew Adamatzky, a professor at the University of the West of England who has spent over two decades exploring what Physarum can compute.
Logic gates from slime mold
Adamatzky demonstrated that Physarum plasmodia can function as basic logic gates. By carefully positioning food sources and observing which tubes the slime mold builds, he showed implementations of AND, OR, and NOT gates. In principle, these gates could be combined to perform any computation that a silicon chip can, though at enormously slower speeds.
Country road network experiments
Starting around 2010, Adamatzky ran a series of experiments placing oat flakes on maps of different countries, each flake representing a major city. The resulting slime mold networks were compared with actual motorway and highway systems. Countries tested include:
- United Kingdom
- Spain and Portugal
- Canada
- Australia
- Brazil
- Africa (continent-wide)
- Mexico
The results were published in his book Slime Mould in Arts and Architecture (2017) and numerous journal papers. In many cases, the slime mold networks matched engineered highways with remarkable accuracy, and occasionally suggested more efficient alternatives.
Slime mold music and art
Adamatzky has also explored creative applications. By mapping slime mold electrical oscillations to sound frequencies, his team created "slime mold music." The growth patterns of Physarum have been used to generate visual art and architectural designs, blurring the line between biological process and creative expression.
Comparison: Slime Mold vs Traditional Algorithms
| Feature | Physarum Solver | Dijkstra's Algorithm | Ant Colony Optimization |
|---|---|---|---|
| Type | Bio-inspired, continuous | Exact, discrete | Bio-inspired, stochastic |
| Guarantees optimal solution? | Yes (for shortest path) | Yes | No (approximate) |
| Handles dynamic changes? | Yes (naturally adaptive) | Must restart | Adapts over iterations |
| Parallelism | Inherently parallel | Sequential | Parallel (multiple agents) |
| Multi-objective optimization | Yes (balances cost, efficiency, resilience) | Single objective | Can be adapted |
| Speed | Slower convergence | Very fast | Moderate |
| Robustness to noise | High | N/A (deterministic) | Moderate |
The Physarum Solver does not replace traditional algorithms for most practical purposes. Its value lies in situations where the problem is dynamic (changing over time), multi-objective (requiring trade-offs between competing goals), or so large that exact solutions are computationally infeasible.
Applications in Artificial Intelligence
Optimization problems
Researchers have applied Physarum-inspired algorithms to a range of optimization challenges:
- Traveling salesman problem (TSP): finding the shortest route visiting a set of cities. Slime mold approaches produce competitive approximate solutions.
- Steiner tree problem: finding the shortest network connecting a set of points (allowing extra junction points). Physarum naturally solves a close approximation.
- Network design: telecommunications, power grids, and water distribution systems
- Supply chain logistics: optimizing routes and warehouse placement
Machine learning connections
The positive feedback mechanism in Physarum has parallels with reinforcement learning. Both systems strengthen successful pathways and weaken unsuccessful ones based on feedback from the environment. Some researchers are exploring hybrid approaches that combine Physarum-style network optimization with neural network architectures.
Adaptive routing
In computer networking, slime mold-inspired protocols can dynamically reroute data around congested or failed nodes. Unlike static routing tables, these protocols continuously adapt to changing network conditions, similar to how the plasmodium reroutes around obstacles.
Living Computers: Physarum as Hardware
Beyond software algorithms, researchers have explored using living Physarum as an actual computing substrate.
The concept
In these experiments, the slime mold itself is the computer. Inputs are provided as food placement or light stimuli. The organism processes these inputs through its growth and network optimization. Outputs are read from the resulting network topology or electrical signals.
Advantages and limitations
- Advantages: massively parallel computation, self-repair, minimal energy consumption, biodegradable
- Limitations: extremely slow (hours to days for a single computation), difficult to control precisely, sensitive to environmental conditions
Living Physarum computers are not going to replace silicon chips. Their value is conceptual: they demonstrate that computation does not require electronics, transistors, or digital logic. This has implications for understanding computation in biological systems more broadly, from brain networks to immune systems.
Hybrid bio-electronic systems
More practical are hybrid systems where Physarum interfaces with electronic components. For example, robots controlled by slime mold (where Physarum growth on electrodes determines motor outputs) and systems where slime mold electrical oscillations are amplified and processed by conventional electronics.
For more on the robotic applications, see our page on slime mold science applications.
The Slime Mold Algorithm (SMA) in Metaheuristics
In 2020, Li, Chen, Wang, Deng, and others published the Slime Mould Algorithm (SMA), a metaheuristic optimization algorithm inspired by the oscillation patterns of Physarum. Unlike the Physarum Solver, which directly models the organism's tube dynamics, SMA uses the slime mold's adaptive behavior as a metaphor for searching solution spaces.
SMA has been applied to:
- Engineering design optimization
- Feature selection in machine learning
- Image segmentation
- Parameter tuning for other algorithms
Published benchmarks show SMA performing competitively with other metaheuristics like Particle Swarm Optimization and Genetic Algorithms, particularly on multi-modal optimization problems where many local optima exist.
Future Directions
The intersection of slime mold biology and AI continues to expand. Current research frontiers include:
- Neuromorphic computing: using Physarum-like networks as models for brain-inspired computer architectures
- Embodied intelligence: studying how the slime mold's "thinking" is inseparable from its physical body, informing theories of embodied cognition in robotics
- Multi-agent systems: translating the decentralized control of Physarum into algorithms for autonomous vehicle coordination and drone swarms
- Quantum-classical hybrid: exploring whether Physarum-inspired algorithms can be adapted to run on quantum computers for even faster optimization
What started as a biology experiment with a maze and some oat flakes has grown into a genuine subfield of computer science. The slime mold has taught us that intelligence and computation can emerge from the simplest physical systems, a lesson that continues to reshape how we think about artificial intelligence.
To learn about other scientific applications, visit our overview of slime mold in science. For the biological foundations, see slime mold memory and single-cell biology.