It’s been a while since I’ve written anything here, but today I found something that is worth mentioning. It’s the kind of stuff that maybe in 20 years from now will be so big and important in human lives it’s like telling you about computers 60 years ago.
In many of my previous blogs, one of my critiques was that Artificial Intelligence community is wasting an important amount of resources and money going into a path that doesn’t reach anywhere.
In other words, the classical view of computers and computing is a dead end for making intelligent machines that one day would challenge humans.
Don’t get me wrong, super computers like IBM’s “Blue Matter” can simulate very complex neuronal networks that will perform and emulate many of the brain functions. But to do it correctly, most of the computation power goes towards emulating the analog memory elements called synapses.
Have I mentioned that a normal human brain has in excess of 1,000,000,000,000,000 (one thousand trillions – bigger even than US National Debt) of these “computer incompatible” little structures?
So what are we going to do about that?
Well, there is one little electronic circuit that seems to be very promising in taking care of the problem.
It’s called memristor.
Neural networks can learn patterns that are very difficult for engineers to craft as specific algorithms, and that is mainly because they depend on analog memory elements called synapses. Learning occurs when simultaneous voltage spikes are generated from feature detectors in the senses, like edge detectors in the eye. When the simultaneous spikes come in, say from the edge detectors in both eyes, the receiving synapse in the brain responds by increasing its value – a digit used for supercomputer simulations.
Instead, memristors change its resistance value.
Memristors respond to these simultaneous voltage pulses in a manner nearly identical to that of brain synapses, making them a viable alternative to supercomputer simulations. Massive crossbar networks of memristors, could create a more accurate and much faster executing emulation of brain functions than supercomputer simulations.
It doesn’t matter how many zillions of operations per second a computer can execute, it will never be able to become intelligent and not require human derived algorithms that both describe and process information from their environment.
Automatic and autonomous learning of relevant and probabilistically stable features and associations is the challenge that computation will never be able to break through. Computation is a linear process, logic is combinatorial.
Combinatorial complexity of analog logic is infinite.
Welcome to the “Real World”!
If any of this doesn’t make any sense to you right now, just remember this:
One day, nanometer size electronic synaptic circuits (memristor like) capable of adapting the connection strength between two neurons in a manner similar with biological systems, will make the terminator increasingly difficult to distinguish from his human counterparts.
Until then, check these pages once in a while for updates on the subject.







