Search
  • Cameron Gordon

What can we learn from slime mould?

A slime mould is an amorphous blob of unthinking biological material, capable of extraordinary acts of creative planning and navigation. In studies, scientists have found that paths created by slime moulds can mimic the complexity of mass transit systems (e.g. the Tokyo subway system), efficiently routing a similar map of hub connections that in the real world evolved as the result of decades of intelligent traffic engineering. Others have shown it navigate across mazes and even act as basic logic gates.

How does this work?

Consider the following gif of a slime mould searching for food.

Initially it searches diffusely in all directions, shooting out tendrils like gooey forks of lightning, until it chances upon a food source. The slime then strengthens this connection and begins extracting nutrients, before continuing the search. An unthinking mass of creeping protoplasm is able to efficiently explore and exploit an uncertain, sparse reward.


We can note two things immediately here: first, the way the slime searches indeed looks a lot like a lighting strike*; second, it is similar to a fast and efficient algorithm known as a Rapidly Exploring Random Tree (RRT) or the directed version RRT*.

The goal behind the RRT algorithm is to rapidly explore an unfamiliar environment. It consists of a series of randomly branching forks, which covers the domain of the environment with a sparse net of exploring arms. Each point on the domain is a potential configuration (e.g. an x and y coordinate), with the branch representing a potential link between the two (which may be blocked by an object in the way). If any arm happens upon the goal - bingo, it's found the tasty reward. An efficient complexity can arise from simple rules.


What can a business or individual learn from the slime's strategy?

Similar to the slime businesses face an uncertain environment with sparse rewards. The difference is that instead of travelling across a 2D plane (slime mould on a petri dish), businesses need to search across a space of potential strategies, across dimensions including product types, international markets, customers, costs, weights, or partners. Here branches represent conceptual links between different strategy configurations (which again may be constrained). By examining the result of small random changes to a potential strategy, the RRT algorithm can efficiently search across a wide range of potential strategies.


The RRT algorithm is similarly considered efficient for high dimensional problems. The similar appearance to lightning is not coincidental: the physical properties that lead lightning to fractally fork from sky to ground as it meets a point to discharge is likely efficient in terms of energy. For a business this energy relates to the time and energy spent searching for a new profitable enterprise or viable strategy.


The ideas that this leads to are well known in innovation literature:

  • Start where you know - chancing upon a new profitable strategy near to your original position avoids the cost of searching.

  • Avoid early assumptions - the benefit of the RRT algorithm is that it can efficiently map the space of potential solutions, but it requires searching widely and diffusely (it's a form of blue-sky thinking).

  • Don't get locked in - avoid putting too much weight or get bogged down on any particular line of thinking, as this isn't useful until a viable solution has been found.

  • Use what you know - credible information restricts the space you need to search (this is the idea behind RRT* which combines RRT with a heuristic to aim more forcefully towards potential solutions).

  • Exploit early - once a profitable strategy has been found, begin exploiting it (get the minimum viable product to market).

  • Continue innovating - keep exploring for more optimal or alternative solutions (note in the gif that the mould misses the nearest source of food initially, but eventually finds it due to continued exploration).

This is, of course, a vast oversimplification - businesses are not as simple as a slime hunting food across a jungle floor, and the math doesn't translate neatly to a half-dozen aphorisms. But the point is that the same unthinking process can be an efficient way to conduct a complex search for a reward. In the artificial intelligence community, the RRT algorithm has been successfully applied to motion planning problems (e.g. autonomous vehicles, or robotic control). Conceptually it can be applied by businesses too.


And wouldn't it be fitting if the slimy world of business was able to take some inspiration from actual slime?


* More generally, the process is linked to diffusion-limited aggregation and the development of Brownian Trees.



2,970 views

©2019 by Cameron Gordon.