Say hello to Kyrre Glette, Jim Tørrenson, Jeff Hawkins and one of your future robot overlords. The first three have developed two different ways to create independent machines. The fourth is just looking to destroy you.

Glette and Tørrenson, from Oslo University, have come up with what could be the next generation of hardware—machines that evolve on their own. This means that, rather than having to install new software to update the machine, the evolving hardware inside changes in order to better complete whatever task needs doing. How the hell is this done? Well, while I would trap underpaid little people inside the machine, apparently this is all about pairing up genes in the hardware to find the design that is most effective to solve a specific problem.

This approach avoids the limitations that efforts at artificial intelligence have had so far—precisely the ones that Hawkins claims to have smashed with his new venture, Numenta, which has created software that mimics the human brain. Or so he says.


More on how these two technologies compare to each other, right after the jump.

The Norwegians' first step in 2004 was to make Henriette, a robot chicken that used evolution software in order to learn how to walk (you can try a program that does precisely this in your PC or Mac here). Rather than trying to comprehend the world and create solutions through AI, the evolution hardware tries random variations starting from a seed, which are measured against the problem in order to select the better options. Following the best paths, the hardware reaches an optimum solution for a very specific problem.


In this case, they want to make a robot that will evolve to install underwater oil pipes on its own at 2,000-meter depths, where you either need miles of cable or echo signals to convey orders to the machine, which means long delays between command and execution.

Until one day they finally rise from the seas to command and execute us, that is.

Hawkins' Numenta solution, on the other hand, tries to actually make a machine think as a human brain, feeding it with sensorial information that the artificial intelligence will use to build a model of what surrounds it. This new AI concept, developed after Hawkins had one too many beers with the neuroscientists from his nonprofit Redwood Neuroscience Institute, tries to identify elements, establish relations between them, learn from how things work and, just like a child, predict outcomes based on this understanding of the world.

So, to summarize, here's the drill:

Step 1. The thing, probably to be called SkySomething or HAL-whatever, understands the world.
Step 2. Someone gives it the keys to the ICBMs.
Step 3. The thing destroys the world in a massive nuclear attack.
Step 4 (optional). Someone sends somebody else back in time to smack those two crazy Norwegians and that Jeff dude, who apparently don't see enough sci-fi movies, and try to save the planet.

As you can see, both efforts look to solve the same problems, as if we didn't have enough already.

First Hardware to go through Evolution Developed [Bits of News via Dizzy Thinks]
Project page [Oslo University]
The Thinking Machine [Wired]