The legacy of Evolution

I was reading today that researchers have built software that allows a robot to adjust to changes in its body structure and relearn certain tasks. The software uses neural networks, a technique which mimics the human brain, along with genetic algorithms which allows for learning to take place. These techniques have been used before but until now the task off adapting to changes in its own structure has proved infeasible.

With this new software the robot can detect changes to its own structure and then adds a new collection of neurons to its existing "brain" and uses these to learn its new task, building on what it already knows.

For instance, if the robot starts with two simple, un-jointed legs and six neurons it will use these to identify the best possible way to walk. Once it detects no more improvement in its learning process it will lock these neurons in place. It has learnt how to walk.

Now, if we add a joint to each leg the robot, using its previously learnt algorithm, will suddenly detect its walking is no longer as effective. Since it already knows the best way to walk it realises its body has changed. It now creates a few new neurons and begins to learn again. The key idea behind this is it does not start from the beginning. It still remembers how to walk, and simply learns what to do with its new bits. It builds on its knowledge.

This technique has also been used for sensors and several legs all with success. The key drawback mentioned by other researchers though is that this can result in an unnecessarily large number of neurons. The same tasks could be done with fewer, although relearning everything each time is too complex.

However, this talk of unnecessary neurons immediately made me think of our own brains and the supposed 10% we use. Maybe all that extra brain power is the relic of our evolution. The neurons that originally learnt to breathe in water and swim with fins. Now without purpose they sit idle, the legacy of our past.