Totally Unexpected Solution

In his book, Black Box Thinking, Matthew Syed cites a very interesting problem solving incident at Unilever. The nozzle they were using to make detergents wasn’t working well at all. So they turned to the in-house experts in the fields of maths, fluid dynamics and high pressure systems. They couldn’t fix the problem.

When maths-physics expertise couldn’t do the job, Unilever turned to their experts in (hold your breath) biology! The biologists took 10 copies of the nozzle; made random tweaks to each; and tested them. They then took the one that did the best and repeated the process. After 45 “generations”, they had an outstanding nozzle! The pic below shows how the nozzle “evolved”:

The biology technique described above is what they call a “genetic algorithm”. Why that name? Nick Bostrom explained why in his book titled Superintelligence: because it mimics natural selection (mutation, inheritance and selection)!

So does this mean trying random tweaks is the way to solve problems? What are the drawbacks of such genetic algorithms? Let Bostrom list them: You may get nowhere with this approach (after all, it’s just random mutations); it’s expensive (lots of trial and even more errors); at times, defining what’s an improvement is complex (is the improvement worth the rise in cost?); and here’s the key constraint: this technique will always get stuck at a local maxima. Huh? Local what? Relax: it means that every change must be an improvement; so if there’s a path to a vastly better solution that involves degradations along the way, sorry, “natural selection” will kill it off.

Equally, it doesn’t mean that knowing the theory behind something is useless. After all, only after the theory of relativity was published did we know of the power inside the atom (denoted by that famous equation: E = mc²), which in turn led to the development of nuclear weapons. That new weapon could never have “evolved” by just trying to improve on the destructive capabilities of the weapons of the day iteratively.

Unfortunately, we don’t always know which technique works for which problem. What we can do is to be open to possibilities, however counterintuitive they may seem.