Autonomous vehicles are all the rage lately, from self-driving cars, swarming quadrocopters to self-piloting military drones and boats. Unfortunately, every single one has a problem.
Like any computer, they can only do what they've been told to do. In other words, it's up to the programmer to think up every possible scenario and how the vehicle should respond.
Given how many things can happen on the road or in the air that are impossible to predict, that means your self-driving car might make the wrong choice at a crucial moment.
The military wants drones that can fire weapons on their own in case an enemy jams communications. Of course, you don't want drones that just fire at anything in sight, but you don't want to drones that are easily fooled into not firing on enemies.
Really what you need in every case is a computer that can actually think, learn and possibly even feel, but that's far easier said than done. Still, scientists think they may have found the answer.
It's called a neuromorphic chip and it's set up to work like the human brain. Instead of processors, it has virtual neurons that spike in voltage when certain things happen.
These spiking neurons form groups and pathways that create "memory" and possibly "emotions." That's just what the human brain does.
To really test this out, DARPA has teamed with Howard Hughes Research Labs and AeroVironment to build a custom quadrocopter drone.
The drone carriers a high-end neuromorphic chip with 576 neurons. This is somewhat less than the 19 to 23 billion neurons in the human brain and it actually falls between a roundworm (302) and a jellyfish (800).
Still, it's good enough for basic learning. In fact, the drone can recognize if it's been in a certain place before.
The researchers are hoping it can also be used to recognize people and create emotions that make familiar spots or people feel safe.
While a neuromorphic chip won't be taking the place of traditional computers for processing anytime soon, it might work as a useful addition.
For example, the sensors and traditional computer in your self-driving car might see that there's an object near the road. As long as the object isn't in the road, it doesn't really care what it is.
That's when the neuromorphic chip could identify the "object" as a child and - like any good human driver - be extra careful as it drove past. Making it able to "feel" protective of children would be even better.
If you want a good example of the vast difference between artificial intelligence that can feel and artificial intelligence that can't, watch the 2004 version of "I, Robot." This site will tell you where you can watch it, and other movies and TV shows, online.