This week’s BEACON Researchers at Work blog post is by MSU graduate student Daniel Couvertier.
We have all seen the wonders of evolution in the biological world. We have marveled at the great variety in the creatures that share the planet with us, and we have admired how they have adapted to overcome even the most extreme environments. We have also been surprised by the cooperative capabilities that many creatures, from lions to ants to bacteria, have proven themselves capable of performing. Of course, as humans we are no strangers to cooperation. We know that cooperative behaviors can allow a group of individuals to flourish and achieve things that would not be possible otherwise. We have, and continue to experience the benefits of this cooperation every day of our lives. However, every time I bear witness to cooperative acts among members of other species, I can’t help but be fascinated by them. I recognize that human cooperation requires elaborate communication, and issues of “trust” come into play. Is cooperation in the wild defined by the same (or similar) rules? Are creatures as simple as ants truly capable of understanding them? If not, how could such apparently structured and complex behaviors actually evolve without the individuals actively planning them out?
Wondering about these questions and researching their answers was an entertaining hobby, but being a computer scientist, I could only ever take a passive role in them. After all, what could a code-cruncher like me ever contribute to this field of natural sciences? I was involved in computer technology, which is probably the farthest thing from biology. I worked with sensor systems, networks, and robots, all of which seemed to have a large disconnect from the natural world. However, it soon became apparent to me that we were reaching the limits on human design potential. Problems were becoming too big and too complicated for us to come up with solutions on our own. Naturally, we allowed computers to start doing a lot of the work for us, but even that is bounded by the creativity of its human designers. Finally, we took a very important step—we applied evolution to the search of solutions for useful applications in technology. This effort, called evolutionary computation, has had many achievements. We have improved existing systems and developed new ones with little to no human intervention in the decision process. Also, we have been able to use digital life platforms to further our understanding of biology. This field bridged the gap between computer science and biology, and it is where I decided I wanted to be. In particular, I study the evolution of cooperation in groups of artificial agents in virtual worlds.
In my research I get to apply the principles we believe gave rise to cooperation in biology to artificial systems. However, since these systems are not natural, I am not necessarily constrained to all the restrictions that exist in the natural world. For example, in the real world, organisms cannot communicate across extremely large distances (such as miles), while in artificial systems, with the use of radio or data transmissions, this is not a limitation. My goal is to explore evolutionary applications in this new artificial world, where the rules and limitations of the environment are different, thus allowing evolution to take new, never-before-seen paths in its innovations.
In particular, my work started off with my study of Biased Group Selection. In a group of real organisms, say lions, reproducing has a very high cost in terms of time and energy. As such, members of a pack may opt to allow misbehavers and cheaters to coexist with them simply because replacing these individuals with more well-behaving ones would not be cost-effective. However, in an artificial digital world, replacing an organism has no cost at all. I wanted to explore how a world in which the composition of a group is determined by different sets of rules would work. I studied this by establishing several group structures to evolve cooperative predation behaviors. In some group structures, all members were clones of each other (homogeneous groups) and in others, individuals varied amongst themselves at different levels (heterogeneous groups). We explored these variations by considering them to be part of a general spectrum that ranges from a homogeneous extreme to an entirely unbiased heterogeneous one. What we found was that evolving a group of clones results in solutions that are very refined, yielding a high performance. However, it turns out that it is hard for these groups to actually find a solution at all because all individuals are following the exact same evolutionary path. Heterogeneous groups, on the other hand, have more variety and can find solutions quicker. However, this variety does not allow the groups to get as good at solving their problems as their homogeneous counterparts because some individuals waste their time doing other things instead.
For my more recent work, I’m hoping to explore a different aspect of evolution in groups, in particular, self-evolution. While in the real world organisms cannot alter the genetic code that defines them (with the exception of some bacteria), this can be easily set up in a digital world. If I give my digital organisms the ability to alter themselves, how quickly could they adapt to a drastically changing environment? Will they be willing to share their genetic material with other group members? Can this type of self-altering behavior be sustained in the long run? These are questions that I wish to address in my new work for which I am currently designing and developing my own original evolutionary system. Currently, I am exploring the evolution of cooperative foraging where a group of individuals have to forage for food quickly enough to be able to reproduce, or risk getting wiped out by a competing group instead.
For more information about Daniel’s work, you can contact him at couverti at msu dot edu.