Caltech creates first artificial neural network from DNA
By Ben Coxworth
July 22, 2011
One of the things that our brains excel at is the ability to recognize what things are, even when presented with an incomplete set of data. If we know only that an animal is sold in pet stores and stuffs food in its cheeks, for instance, we can be pretty certain that the animal in question is a hamster. Now, for the first time ever, researchers at the California Institute of Technology (Caltech) have created a DNA-based artificial neural network that can do the same thing ... albeit on a very basic level. They believe that it could have huge implications for the development of true artificial intelligence.
The neural network is made up of just four artificial neurons, as opposed to the human brain's 100 billion real ones.
To test the network, the scientists played a game with it. That game started with the network being trained to "know" four scientists, each one identifiable by a unique combination of yes/no answers to the same four questions (such as "Is the scientist British?"). Human players then chose one of those scientists, and provided the network with an incomplete set of the identifying answers. They did this by dropping DNA strands that were programmed to correspond to those answers, into water in a test tube that contained the neurons.
Communicating through fluorescent signals, the network would then either correctly identify the chosen scientist, it would indicate that it didn't have enough data to identify just one scientist, or it would state that the data didn't match any of the scientists.
This was possible because of the manner in which the added strands of DNA paired with the strands already present in the network. This, in turn, was determined by whether or not the sequences programmed into both strands complimented one another.
It should be noted that the network took eight hours to come up with each of its responses, and that new DNA strands had to be created for each game. Still, the Caltech team state that the technology could ultimately be used for creating biochemical systems with artificial intelligence, which could revolutionize fields such as medicine, chemistry, and biological research.
The research was recently published in the journal Nature.