From Shaun Saunders comes a very interesting article in New Scientist:
Up to this point robotic scientists have used a try it, fail, succeed system for robot learning which is often slow and time consuming. MacLeod's team took this idea a step further, however, and developed an incremental evolutionary algorithm (IEA) capable of adding new parts to its robot brain over time. This new system can learn and adapt to changes and relearn functions by adding on "brains" to deal with the changes.
Read the complete New Scientist article
- Christopher MacLeod and his colleagues at the Robert Gordon University in Aberdeen, UK, created a robot that adapts to (environmental) changes by mimicking biological evolution.
Up to this point robotic scientists have used a try it, fail, succeed system for robot learning which is often slow and time consuming. MacLeod's team took this idea a step further, however, and developed an incremental evolutionary algorithm (IEA) capable of adding new parts to its robot brain over time. This new system can learn and adapt to changes and relearn functions by adding on "brains" to deal with the changes.
Read the complete New Scientist article
5 comments:
Shades of Terminator.
Would you agree, though, that although the brain (i.e., neural network) can grow in complexity, it only, of course, can make use of existing nodes etc (i.e., it cannot physically expand or add new neural components) which is still the big difference between machines and biological organisms. organisms.
An interesting question, but honestly they didn't limit the growth to added complexity. The way I read this article as saying the new algorithms see that the present system is not working and the program has been locked into not evolving any further because and such changes were viewed by the system as less efficient. To counter this block, the new program sidesteps the existing system and adds a whole new processing area where the unit once again relearns to operate in its present configurations. Its like growing a whole new brain. Now you would ask, how is this like biological life? Well, consider humans. All through gestation, the fetus goes through all the major animal types and with each change a more complex brain is laid over the older. So, the way I am seeing it with the robotic version is that the older system is completely bypassed for a newer system capable of addressing and controlling the news systems.
Yes, but, it cannot 'grow' new nodes - only use what is already available.
I.e., the physical aspect of the hardware that comprises the neural network is fixed.
Thats kind of splitting hairs really. In reality we don't have that feature either. With enough time that could take place but its not a given. We have fixed amounts of material to work with as well. We can utilize it in different manners....but then thats what the robots are doing. So to be fair, if we are saying that we would have to evolve new material...then we have to expect at some point the same advantage would have to be afforded the machine.
Fair point :-)
(that's one drink for you when I visit)
Post a Comment