Is it better to be clever or stupid? It it better to plan or to try things at random?
The answer seems obvious. Or rather, it used to. Today, let’s take a look at how powerful stupidity can be, and how it can sometimes beat cleverness hands down.
Dumb algorithms vs smart humans
Something scientists from a range of disciplines have been experimenting with in recent years are evolutionary algorithms. And what’s interesting is that most of us still think of evolution in the wrong way.
Let’s be clear – a genetic algorithm does not understand the task you give it. There is interesting work being done in teaching machines to learn, but evolutionary algorithms work precisely because the algorithm is, well, a bit of an idiot.
To explain how it works, let’s ask another question: how did we evolve in the first place? We all know the basics, and yet we tend to think of evolution as the story of a plucky little fish who tried his luck on the land: a saga of intrepid entrepreneurs trying out new approaches to life.
The truth is a little different.
The Original Evolutionary Algorithm
You’ve probably heard the word ‘algorithm’ a lot. It’s the sort of word marketing departments like to use, because it sounds vaguely scientific.
In fact, there’s no formal definition of what an algorithm is. It’s really just a set of instructions that tells a person or a computer what to do.
You use algorithms every day. Our lives are more or less run by them. They tell companies what price to charge us for the food we buy and how much our employers should pay us for the work we do. They crunch through unthinkable amounts of raw data and ensure that this person in this city is going to see an advert for organic deodorant the next time they log-in to Google.
Often, algorithms are very simple. If this happens, then do that. If it doesn’t happen, to something else instead.
We’re very proud of our big, complicated brains, and it’s easy to underestimate the power of simple algorithms. There are few algorithms simpler than evolution, which might explain why so many people struggle to see how it could create something as complex as the human brain.
We could write down the evolution algorithm like this:
When a new organism is created, change a few bits of its genetic material at random.
Throw the universe at it
Only the organisms that survive and reproduce have offspring
Repeat as required
You can see why generations of priests, scientists and newspaper editors have objected to this. It seems ridiculous that those four steps could take a nocturnal shrew and use it to produce, a few hundred million years later, something as complex as, for example, Mark Zuckerberg.
So, how did it happen?
Let’s pretend we were transported back to the early stages of life on earth and given a commission to submit designs for new organisms. What would we do that the algorithm would not?
First, the algorithm has to work with what is already there. It can change a few bits and pieces, but it cannot go back to the drawing board and design a new organism from scratch.
For example, suppose we were in charge of a species of very large, carnivorous reptiles during the last days of the Cretaceous Period. They’re called Velociraptor mongoliensis, and they consider themselves, not without reason, to be perfectly adapted killing machines.
Just as we’re talking to the velociraptors, brainstorming a few potential designs for the next generation,we look up and notice a billion tonne asteroid screaming through atmosphere.
It would be clear to us that we needed to rethink our plans. Quickly.
And we would do that by imagining the future. The dust ejected into the atmosphere would mean that the food chain the velociraptors depended on was about to come apart. Plants would suffer from the lack of sunlight, herbivores would suffer from the lack of plants, and carnivores would suffer from the lack of herbivores.
In fact, it would make most sense to give up on the velociraptor shape entirely and become a worm. That way, the ex-velociraptors could burrow deep into the ground and feed of the decomposing remains of the all the species that were about to go extinct.
And so, by the time the asteroid slammed into the Earth with an explosion around a billion times larger than the blast at Hiroshima, our new generation of former dinosaurs would be safely ensconced as far beneath the surface as possible.
Unfortunately for velociraptors everywhere, evolution does not work like that. It can only change few aspects of an organism at a time, and – and this is the important bit – each change has to work in its own right. It can take a velociraptor as its input and produce a velociraptor with slightly longer toes or different pigmentation. What it can’t do is produce a worm.
In other words, when the asteroid hit, the velociraptors were out of luck.
But the second big difference between us and the algorithm is even bigger. The thing about our design process is that we would have a design process. We would think about what we knew and what might work. We would cross things off the list if they didn’t make sense.
The algorithm has a simpler approach. It tries absolutely everything at random.
This seems like it should fail. And it does.
It fails a lot.
The Corpse Factory
There has been life on earth for just under four billion years ago. Over 99% of the species that ever lived on this planet are now extinct. Not peacefully transitioned into a new species – completely wiped out.
How did they die off? They were hunted to death. They starved to death. They froze to death and they boiled to death. They were vaporized by asteroids and they choked to death on toxic fumes. They died and they died and they died.
An alien intelligence could be forgiven for thinking that the main function of Earth was to be a kind of corpse factory. Evolution doesn’t know or plan, and the result is an awful lot of dead ends.
And yet, it is precisely this total disregard for consequences that makes evolution so powerful.
Because there’s one big problem with knowing things. You can only create a new design based on what you already know. This might not sound like too big a drawback, but consider the following example.
Imagine we were in the imperial court of Emperor Ling during the Han Dynasty in Ancient China. The peasant’s rebellion is escalating, and a group calling themselves the Yellow Turbans is threatening to destabilize the region.
The Emperor gathers his advisers and asks them to design some new super-weapon for his soldiers.
What would they suggest? Bigger swords? Longer spears?
Suppose we piped up and suggested using that black powder the Taoist alchemists were playing around with. We would probably have been laughed out of the room. After all, the enemy are not going to be scared off by powder that goes bang.
And so the imperial forces go on to fight a long, bloody war, which eventually hastens the end of the Han Dynasty. They never pays much attention to gunpowder. They certainly don’t imagine that, combined with their advanced steel manufacturing ability, they have the building blocks of a weapon that wouldn’t just crush the rebels, but completely alter the course of human history.
You see, the problem is that a human designer cares about how his design works out, and so he or she is going to limit their work according to technology they understand well enough to make predictions about. If you don’t know about gunpowder, then you aren’t going to invent the gun.
For the most part, humans have to understand things at least a bit before they use them. We have to push back the boundaries of human understanding first, then work out how to do something useful with whatever we’ve discovered.
In contrast, evolution walks straight past the boundaries of human understanding without even blinking. It doesn’t even guess; it just produces as many new designs as possible.
It’s a bit like firing a million arrows at random into a field. Each individual arrow has a low chance of hitting the target, but there’s a relatively high chance that at least one of them will.
Modern science is very aware of how useful this approach is. Genetic algorithms – algorithms which evolve different solutions to a problem – are used to give us results that humans might never have been able to produce themselves.
For example, in 2006 NASA tried using an evolutionary algorithm to design new antennae shapes.This is a complex task when done by a person, and one that demands particular knowledge. Sometimes, you want an antennae to pick up a new or unusual pattern of radiation, and the existing shapes don’t work very well.
The evolutionary algorithm was set to work, and the result were, well, weird. The best description would be something between a tree and a spider doing its best impression of a paperclip.
Here’s the thing though; they work. It took less time for the algorithm to design the antennae than for a human.
In the words of Gregory S. Hornby and Al Globus, the scientists in charge of the project, the evolutionary algorithm created designs ‘which have unusual structures that expert antenna designers would not be likely to produce.’
That’s the things about evolution. It doesn’t care about how many times it gets it wrong, and it doesn’t care how any of its designs work. It doesn’t care full stop.
When you consider how quickly evolutionary algorithms create designs we can’t match and don’t necessarily even understand, you begin to appreciate how difficult it is to for us, with our logical thought process, to try and work out how the brain works. It does its job very well, but it wasn’t designed.
We often talk about ‘miracles of nature’, or the ‘perfect design’ of a particularly beautiful animal. And from our perspective, i.e. as one of the few species that is still in the landing of the living, it does seem that life is miraculously well suited to its environment.
But what you don’t see, of course, is the millions upon millions of designs that didn’t work. The people at NASA don’t bother making the antennae that didn’t pick up anything at all, just as the Earth isn’t full of the ghosts of all the species who turned out to be a dead-end.
Evolution tinkers at random. That’s all it’s ever done.For us, this seems inherently dumb. That’s because we care about the results, so we try to limit the risk by thinking about what we know. Otherwise, you’re performing the design equivalent of buying a lottery ticket and hoping you win.
As far as evolution is concerned, that’s not a problem. After all, there’s an easy way to win the lottery.
Buy all the tickets.