We are spending hundreds of billions of dollars building artificial intelligence - vast data centres, thousands of GPUs, nuclear power ambitions, and a lot electricity - while somewhere in a damp rotting autumn leaves of my garden a yellow blob is quietly solving optimisation problems. No brain, no neurons, no GPU, and no OpenAI subscription.
In 2010, Japanese researchers led by Atsushi Tero placed oat flakes on a surface, representing Tokyo and the surrounding cities, and introduced a slime mould called Physarum polycephalum. The organism went looking for lunch. What emerged was a transport network remarkably similar to Tokyo’s railway system, balancing efficiency, construction cost and resilience. The Japanese had spent generations developing sophisticated transport infrastructure. A blob of slime produced a comparable network while searching for breakfast. The extraordinary part is that the slime mould was essentially one enormous cell: not a brain containing billions of neurons, but one cell, albeit a rather unusual one with many nuclei.
Slime also learns. Experiments have demonstrated that slime mould can habituate to harmless unpleasant conditions, and more astonishingly that when two slime moulds fuse, an experienced individual can transfer its learned response to an inexperienced one - learning, memory transfer, adaptation, and no neurons required. meanwhile, Fungi, quite different organisms, have their own Wood Wide Web, the mycorrhizal network. Nature has been experimenting with intelligence for rather longer than Silicon Valley.
Those of us who write software, particularly in finance, already know something about borrowing computational ideas from nature: flocking, swarming, simulated annealing. Simulated annealing comes from metallurgy: heat a material, allow its atoms to rearrange, then gradually cool it towards a low-energy configuration. We turned a physical phenomenon into an optimisation algorithm. Related principles even appear in quantum annealing, although the underlying physics differs.
Slime mould does something similar with its own body. Its networks strengthen where resources flow effectively and weaken where they don’t - exploration, feedback, adaptation, optimisation - and researchers have already developed Physarum-inspired algorithms for shortest paths and network optimisation. Could similar approaches help with financial transaction networks, liquidity flows, risk dependencies or computational workloads? Not every problem needs another billion-parameter model. Sometimes the answer might be a better algorithm inspired by a very small organism.
one biological cell verus 640 billion transistors: An unfair comparison? But worth considering. A large AI inference system might use eight GPUs containing, collectively, something like 640 billion transistors. Our slime mould uses one biological cell. Transistors aren’t cells, and slime mould cannot write Zig code (not yet) or explain a derivatives portfolio; these are very different forms of computation. One still needs an industrial infrastructure to produce its answers. The other needs a few oat flakes. But what if we take a page out of the EV world, and we measured the performance of intelligence as useful adaptation per unit of energy, the rankings might look rather different - 143 microwatts verus 8,000 watts? - and that is just for inference, let’s not discuss the MEGA WATT COST of training the LLM.
We tend to associate intelligence with thinking, reasoning and planning. But our slime mould does something rather different. It explores, adapts and finds solutions without anything resembling a brain.
Maybe intelligence isn’t just about how well you think. Maybe it’s also about how well you adapt. And if that’s true, we may have been underestimating a great many creatures. Including slime.
Our illustrious leader, President Trump, has officially proclaimed that we are living in the age of Superintelligence (SI), perhaps he has a point, and perhaps we’ve simply been looking in the wrong place.
This article is completely inspired by Samanth Subramanian’s article in the Guardian’s The Long Read, “‘This is dangerous’: slime moulds and the bitter debate over the nature of intelligence”