wrong expectations in AI research

For quite a while humankind works on understanding the human brain. While we know quite a bit about what single components do, we are more or less innocent about how any cognitive process actually works. You see that on the lag of “brain like” algorithms that are actually use full. The technological pinnacle like Siri is performing great, but is not at all brain like, but a rather large Markov Process. Most AI algorithms might perform a human task, but they do that using their one algorithm and have been invented by some computer scientist and not because they were reverse engineered from the biological counterpart. At the same time we have a lot of anatomical data on nervous systems, why is it so hard to reverse engineer it?

I think the main reason for that is that it wasn’t “engineered” in the first place. You and I would design some kind of top down architecture made from modules and single components that are understood, and build the logic using already working components. The result would be a rather well organized layout similar to what we would do when designing a computer chip (see the title image for an example). This regular layout is almost like an additional requirement to ease the design process. The designer “off loads” some of her cognitive load into the layout. It makes constructing things much easier if you have reference points and group components of similar function together. But how was this done in a brain? Evolution opportunistically placed things were they were needed, co-opted functions when possible, and most importantly did not need to design something with the purpose of being understandable or portable to other systems. Imagine you have to build something complex from scratch similar to something you have seen on a picture. Your success will very much depend on how the machine in the picture was designed in the first place. If someone planed well, made it modular, and easy to access, the machine layout makes functions obvious to you. However, if the original designer cobbled together a big pile of wire that performs the function but does not have any rhyme or reason to its design, you will most likely need to reinvent the wheel. The same analogy hold for reverse engineering the brain or brain like algorithms; your biological blueprint is not transparent to you and was purely optimized for functionality and wasn’t make portable or to be understood – because evolution doesn’t need to understand anything.

Too bad you might think, we will never be able to reverse engineer a brain … but there is hope! We know the process that made brains in the first place: Evolution. And we are very capable of making systems evolve (or at least I am)!

Leave a Reply