10 February 2011

Genetic drifting

Okay, so neither genetic drift nor drifting are involved in this, but I couldn't resist the stupid joke, because it's about evolving cars:

Play it here!
It's called BoxCar2D and could be a fantastic tool for teaching a biology class about how natural selection works.

It happens in real time right in front of you, and uses an easily understood measure for evolutionary success: distance traveled. I also think it could help draw in those students whose interests lean toward the mechanical rather than the biological.

Basically, it starts by throwing together a bunch of random combinations of wheels and polygonal bodies for each "car" and then lets them loose on the track you've chosen. After twenty cars have gone the next generation starts, but instead of being entirely random like the first, it uses the builds of the most successful cars from the previous generation (those that traveled the farthest) to influence the body plans of its 20 cars. And so on an so forth.

I'm not sure about any of the numbers, like what number of the top performers are chosen and to exactly what degree their body plans influence the choice of parts for subsequent generations, but you can see the trends right in front of your eyes, and that's the important part.

Some important comparisons and contrasts that I think would be important to point out to a biology class: without knowing the details, I'm not sure whether going the farthest is analogous to reproducing the most or if the top x cars all reproduce an equal number, in which case it would be analogous to being able to reproduce at all; I'd present both as possibilities.

In this model the environment is static, which is of course not true in real life. If one of the tracks (I haven't tried them all yet) happens to have two differing portions to it, you could possibly demonstrate how being well-adapted to one portion isn't necessarily an asset in the next one (and perhaps postulate that one of the designs you saw earlier might have been the more successful had the second portion come first).

In my experience while using the max of 3 wheels, most cars will be two wheeled, and occasionally a car with three may show up; this can be compared to a mutation introducing new information into the genome, and you'll be able to see whether it's successful or not.

It's important to realize that this doesn't do the best job of modeling an entire organism, given the random nature of the initial cars and in the amount of times they turn out completely unable to function. It's an okay level to see it on if you think that's as far as the class will be able to follow you, but if you can compare it to a single aspect of an entire organism, such as the ability to thrive in a very specific set of conditions or the ability to digest a particular nutrient, then I think that'll give an overall better understanding of the concept.

I think perhaps the best thing about this program is that you get a real feel for the unguided nature of it all. You and your class can see as rational observers that making one small adjustment to a certain body plan could make it quite successful, but just like nature, this model isn't based on potential, just results, and a small but fatal flaw can cause the information from an otherwise good genome to be lost from the population.

I'm already certain that I'll be using this once I have a bio class of my own. I'm thinking I'll show the class a few generations while explaining the basics, then take it out of view but keep it running while I get down to the nitty-gritty and the particulars of real world natural selection, and then bring it back up at the end to see how the cars have changed.

There's also an option to design a car, but don't feel any pressure to give it equal time. ;)

(By the way, I found out about this site from Dinosaur Comics, which I linked to in an earlier post. Funny AND educational!)

No comments:

Post a Comment