DrJosh

Roboticists wish to create adaptive machines that can alter their mode of operations in the face of a continuously changing world.

Natural selection is the only mechanism we know of that can produce adaptive machines with human-level intelligence. So, it seems to make sense to follow evolution's lead....

Ah ha, this question had to surface. :)

I think Rod Brooks, the former director of MIT's AI lab, hit the nail on the head. He argues that people will gradually merge with machines, so that it won't be as much an 'us-vs-them' scenario in future, but rather a 'we-together' hybrid species that will gradually emerge. We already have teenagers glued to smartphones, cochlear and retinal implants, and, soon, Google Glass. In a few short years, it might not be possible to tell the machines from the humans, and vice versa.

Risky if you're an academic and need to produce a regular stream of new ideas, but not if you're a hobbyist and just wish to see where your ideas will take you.

Oh, that's a tough question. I make the following suggestion to students and hobbyists alike. Follow your intuition, and sketch out your idea without worrying about whether someone's already done it. Then, try to create a `prototype' of your idea. This might mean writing a computer program or creating a simple robot. Once you've got something working, then do some searches on Google Scholar to see if the idea has already been tried. If not, go back to your program or robot, and make it a bit more sophisticated by introducing more of your idea into code or metal. If your prototype continues to show promise, dig more deeply into Google Scholar, or send a video of your robot/code in action to a researcher. Rinse and repeat...

Another reason we tend to work with evolutionary algorithms is that they provide a relatively straightforward way to automatically improve both the body and brain of a robot. Most machine learning algorithms only improve the brain of the robot, and leave the researcher to manually design the robot's body.