Adaptive Robots: Yet More Evidence for Evolution

Not only do biological organisms adapt before our eyes, but in recent years researchers have developed robots that adapt via Darwinian selection. As one recent research paper explained:

Darwin suggested that adaptation and complexity could evolve by natural selection acting successively on numerous small, heritable modifications. But is this enough? Here, we describe selected studies of experimental evolution with robots to illustrate how the process of natural selection can lead to the evolution of complex traits such as adaptive behaviours. Just a few hundred generations of selection are sufficient to allow robots to evolve collision-free movement, homing, sophisticated predator versus prey strategies, coadaptation of brains and bodies, cooperation, and even altruism.

Can robots illustrate how complex traits evolved? Yes they certainly can, the paper concludes. According to the authors, such experiments provide “a spectacular demonstration of the power of natural selection.” Is that true?

Joseph Raphson was a seventeenth century English mathematician who is most famous for devising a numerical method that was also devised, independently, by Isaac Newton. Newton derived the method years earlier but it, as with so much of his work, went unpublished for decades. Therefore Raphson’s work was his own and he rightly is credited alongside the great Newton. Of course standing alongside a giant often means one is overlooked. Today the method is usually referred to simply as Newton’s method, sometimes as the Newton-Raphson method, but never as Raphson’s Method.

The Newton-Raphson method is a relatively simple iteration for finding where a curve has a value of zero. This root finding algorithm is a handy way of calculating difficult formulas, such as the square root of a number. The idea is to temporarily replace the complex curve with a simple straight line. The straight line, which approximates the curve but is far easier to work with, takes you closer to the answer and allows you to generate a new, better, straight line approximation. After only a few iterations you have a an answer that is good to a few decimal places.

An important property of the Newton-Raphson method is that it works with no knowledge of the overall curve. It needs to know only the shape of the curve locally, at the current approximate answer.

Today’s numerical methods are more complex and sophisticated, but the Newton-Raphson method remains an excellent pedagogical tool as well as a useful method in many practical applications. It is also a simple example of the connection between numerical methods and evolutionary theory.

The Newton-Raphson method’s ignorance of the overall curve is analogous to evolution’s ignorance of a population’s overall fitness surface. The Newton-Raphson method nicely converges to a mathematical solution just as evolution is proposed to nicely find biological solutions. Neither knows anything more than the local shape of the curve, but this is sufficient to locate the roots.

In fact certain branches of numerical methods have used biological and evolutionary concepts to explore new approaches. Genetic algorithms and neural networks are computational methods that use concepts such as fitness, genes, chromosomes and selection.

Not surprisingly evolutionists have interpreted the success of these numerical methods as yet more compelling evidence for evolution. Do not these computational algorithms demonstrate the underlying feasibility of Darwin’s idea?

Of course not. Such fallacious logic has it backwards. Clearly Darwin’s idea is mathematically tractable. That is, if fitness landscapes are relatively smooth and reasonably shaped, and if an initial population just happens to appear, and if biological variation just happens to arise and accumulate, and if populations do not resist such change, then of course species can evolve to new designs. Indeed evolutionists often refer to Darwin’s idea as a truism. It necessarily follows from the premises.

And so it is hardly surprising that we can develop numerical methods that solve problems in this manner. Indeed, Newton and Raphson’s work predates Darwin by almost two centuries. By the nineteenth century there was no question that, given the right properties, iterative numerical solutions could move toward a solution in a stepwise manner. And today our methods are all the more sophisticated. But none of this justifies the evolutionary premises.

Consider the robot research in the paper referenced above. The evolutionary robot experiments took place in an engineered environment. Most of the experiments were computer simulations. That is, the experiments took place in computers created by people, developed in programming languages developed by people, running on electricity provided by people, and interpreted by people. Some of the experiments used actual robots in a testing environment that reflected the computer simulations.

Furthermore, the robot’s capabilities were pre arranged and provided. The algorithms that provided the instructions for the robots, as well as the robot configurations, evolved. But such evolution was within the context of the provided capabilities.

For instance, the algorithms that provided the instructions for the robots are controlled by weighting factors and these weighting factors are varied in the experiments to “evolve” the robot behavior. The weighting factors alone do nothing. They rely on the advanced algorithms and software to have their effect.

And sophisticated sensors were supplied that provided the inputs to the algorithms. In turn, the algorithm outputs controlled electronic motors that set the robot into motion. Then the experimenters provided simple criteria (which they referred to as the robot fitness) to select the best designs at each iteration. Just as the Newton-Raphson method iterates toward a solution, the robots were iterated toward their solutions. Here is how the paper described one experiment:

The sensors were connected to eight input neurons that were connected to two output neurons, which each controlled the direction and speed of rotation of one of the wheels. The genome of the robots consisted of a sequence of bits encoding the connection weights between input and output neurons. Mutations allowed the strengths of connections between neurons to change over generations. Experimental selection was conducted in three independent populations each consisting of 80 individuals. The performance of each robot was evaluated with a fitness function describing the ability of the robot to efficiently move in the maze.

It may have been a nifty bit of engineering work, but this is hardly evolution in action. If you randomize aspects of pre supplied functionality, and select for certain outcomes, then you will end up with those outcomes. Here is how the authors described one outcome:

the best evolved individuals across all replicates moved in the direction corresponding to the side with the highest number of sensors. This was because individuals initially moving in the direction with fewer sensors had higher probability of colliding into corners and thus had lower probability of being selected for reproduction.

It seems to have been a fine piece of work, and such approaches are undoubtedly useful in robot design, training and control. But this has very little in common with biological evolution.

And the experimenters even solved the problem of large-scale change by designing a “gene” to supply different algorithmic topologies:

The genome of each robot consisted of two chromosomes, one encoding the topology of a neural network and the other encoding the shape of a body composed of rigid blocks linked by controllable articulations. This led to the coevolution of different types of robots capable of moving towards the cube and preventing access to its opponent. For example, some robots consisted of a cubic block with two articulated, arm-like structures, which were used for moving on the ground and holding the cube. Other robots were composed of only two articulated worm-like segments where one segment was so large and heavy that, once placed over the cube, it prevented the opponent from displacing it.

Given these engineering heroics it is hardly surprising that the experiments worked so well:

Just a few hundred generations of selection are sufficient to allow robots to evolve collision-free movement, homing, sophisticated predator versus prey strategies, coadaptation of brains and bodies, cooperation, and even altruism.

Yes the robots evolved in this engineered environment, but it is simply a misrepresentation to claim this is evidence of biological evolution. Nonetheless this is precisely what the authors claimed:

these experiments revealed how the coevolution between brain and body morphologies can produce various types of adaptive behaviour and morphologies. … These examples of experimental evolution with robots verify the power of evolution by mutation, recombination, and natural selection.

Verify the power of evolution? A spectacular demonstration of the power of natural selection? This is, to put it diplomatically, a non scientific misrepresentation. Unfortunately this is what we can expect from evolutionists. If there is anything spectacular here, it is the steady stream of myth-like claims made by evolutionists. When not overtly religious, the evidence they produce for their theory is, frankly, absurd. Religion drives science and it matters.