Greater Diversity through Evolutionary Algorithms

Recently I read a rather interesting article from a Scientific American blog which hypothesizes about the shape of the human penis from an evolutionary standpoint.

[The methodology employed is described as  “logico-deductive investigative” — meaning that the penis current form is studied within the context of its function and hypothesis are formed, working backwards, regarding why this form came about. I imagine that this is the most natural methodology an evolutionary biologist or psychologist might employ. Understanding this and other methods of explaining evolutionary design might be an interesting exploration for some future post. ]

In a follow-up interview, the researcher Gordon Gallup emphasizes that evolution occurs by selection, not by design. “The raw material for such selection consists of nothing more than random genetic accidents (mutations).” As such, two separate genetic branches cannot be expected to follow the same path of optimization, even if starting with identical initial parameters. This is, of course, very fortunate, for it leads to the great diversity of life where so many radically different methods are employed to solve the same universal problems of survival. 

In contrast, the  solutions reached by purposeful human design are far more limited in diversity. A student of architecture, for example,  might be struck by the number of radically different approaches that humans have employed. 



This range of designs, however, is determined solely by human creativity and contrained by cultural, religious, and other influences. Augmenting human creativity with evolutionary computations could result in an explosion of design ideas.

While one might hesitate to accept that computers could so directly contribute to the creation of art, and before we dive in protest into philosophy and aesthetics, allow me to point out that 

  1. computers and electronic media are already an important contributor to art today, and all that is being suggested is an additional computer-based tool in artistic exploration
  2. we find diversity in nature beautiful, and there is no great difference between natural selection and a hypothetical evolutionary algorithm on a computer.

Architecture’s easily recognizable combination of engineering and art makes it a convenient example. The idea that the increased use of evolutionary methods could lead to a dramatically greater diversity of solutions, however, extends well beyond architecture to many other disciplines, both scientific and creative.

May 10, 2009. Tags: , , , , . Emergence. Leave a comment.

TedTalk: Learning from Evolution

Robert Full describes ways in which engineers can and have learned from evolutionary systems in biology. He notes that we can get ideas from animals and insects, but we cannot copy them blindly.

December 28, 2008. Tags: , , . Emergence. Leave a comment.

Surprising Results from Emergent Systems

One of my favorite examples of a surprising result from an emergent system is an experiment that utilized a genetic algorithm to solve a sorting problem on a FPGA chip. The hope was to arrive at the best (minimal) solution to the sorting problem faster and with more confidence than a human could. The actual result was that the genetic algorithm found a solution that was better than what theory said was even possible!

How did that happen? It turned out that the models humans used and programmed into computer simulations were not in fact complete. They are based in theoretical simplifications that do not fully describe the physical world, in which nature is not limited to the ‘1’s and ‘0’s of electrical theory, in which nature could take advantage of the minute electromagnetic coupling of electrical components packed tightly into a silicon ship, in which higher order effects not considered by models could be utilized to solve the problem. 

By physically wiring an infinitely reprogrammable Xilinx XC6216 FPGA chip to allow the genetic algorithm to use a physical board — instead of a computer simulation of the board — the algorithm arrived at a solution that used fewer steps and fewer gates than the best solution previously offered by theory. 

“What is downright scary is this: the FPGA only used 32 of its 100 available logic gates to achieve its task, and when scientists attempted to back-engineer the algorithm of the circuit, they found that some of the working gates were not even connected to the rest through normal wiring. Yet these gates were still crucial to the functionality of the circuit. This means, according to Thompson, that either electromagnetic coupling or the radio waves between components made them affect each other in ways which the scientists could not discern (Taubes 1997).”

The structure of the genetic solution discarded the need for models or theory and was allowed to experiment with actual physical properties.  All that was needed were rules that allowed better solutions to survive in each generation until the optimal result was achieved. 

Read more: [PDF] 

Not all problems are suitable for such an evolutionary approach; it is often impractical even when it does work. But what appeals to me is the idea that we can  do what nature does. We can use simple rules to find an amazingly efficient solutions to a problem, without understanding how that solution really works. Not far in the future, the world might be full of algorithmic solutions as diverse and amazing to a computer scientist as the animals of the world are to a zoologist.

December 15, 2008. Tags: , , , . Emergence. 1 comment.