At the heart of artificial intelligence lies the question of whether we might be able to create artificial systems that behave and compute in the same manner than human beings do. This would obviously be a mind-blowing breakthrough were it ever accomplished – it would give us new applications for computers, change the nature of work in our society, and force us to redefine the very nature of being human. Perhaps it is no surprise, then, that such a feat has proven to be incredibly difficult to achieve. While it has grown in complexity and scope, artificial intelligence is still quite far from any kind of accurate human resemblance. However, this may change very soon.
Back in 2008, the world of electronics was abuzz with excitement over a new invention – the memristor. This is an electrical component that behaves very similarly to a resistor, but with one key difference. Memristors impede the flow of electricity, but the amount that they do so is dependent on the current that has passed through the memristor in the past. Now, this might not seem like such a big deal, but think about the implications. Essentially, such a piece of hardware has the ability to store some information about its previous input. It has the electrical equivalent of memory. With that in mind, let’s venture into the realm of cognitive science.
The problem with traditional artificial intelligence is that it is based on a computer architecture that is inherently different from biological brains. Computers have a specific place where computations are carried out (CPU), a specific place for short-term memory (RAM), and a specific place for long-term memory (the hard disk). This means that any time a particular bit of information needs to be altered, it has to pass through a number of bottlenecks that drastically reduce the efficiency and speed of the system.
For those of you who are familiar with brains, you know that they don’t work that way. There is no central processing unit embedded within your skulls, and there is no “hard drive” area that stores all of your memories. Instead, there are only millions of neurons in an interconnected and never-ending chorus of electrical activity. Such a system does not need to separate its various functions into discrete locations because, broadly speaking, every location in the brain carries out every function that a normal computer would. The neurons (and possibly their neighboring cells, glia) both carry out computation as well as store information about the past.
And so all of our efforts to simulate brains have hit this fundamental roadblock – it is incredibly difficult to create machines that act like brains without being built like them. This is where memristors come in. By allowing memory to be embedded directly within artificial networks, we are one giant step closer to mimicking the way that biological neural networks compute and store information. Such a revolution in information technology will allow us to create systems that behave very differently from those currently in use and perform tasks that are too difficult for most computers.
For more information, check out this article from IEEE Spectrum. It’s written by a team from Boston University working with HP labs to create one of the first “neural” artificial computers.