Eat your heart out Steven Spielberg — turns out artificial intelligence is not just a figment of your imagination.
A team of researchers lead by Lulu Qian from the California Institute of Technology (Caltech) have for the first developed an artificial neural network — that is, the beginnings of a brain — out of DNA molecules. And when quizzed, the brain answered the questions correctly.
They turned to molecules because they knew that before the neural-based brain evolved, single-celled organisms showed limited forms of intelligence. These microorganisms did not have brains, but instead had molecules that interacted with each other and spurred the creatures to search for food and avoid toxins. The bottom line is that molecules can act like circuits, processing and transmitting information and computing data.
The Caltech used DNA molecules specifically for the experiment, because these molecules interact in specific ways determined by the sequence of their four bases: adenine (abbreviated A), cytosine (C), guanine (G) and thymine (T). And what's more, scientists can encode the sequence into strands of DNA molecules, essentially programming them to function in a predetermined way.
Without getting too complicated, Qian and her team created four highly simplified artificial neurons in test tubes comprised of 112 strands of DNA, each strand programmed with a specific sequence of bases to interact with other strands. The interactions resulted in outputs (or not), basically mimicking the actions of neurons firing. In order to see the DNA neurons firing, the scientists attached a fluorescent molecular marker that lit up when activated.
Next, the researchers played a trivia game with the neural network to see if it could identify one of four scientists based on a series of yes/no questions. Basic information related to the identity of the scientists was given to the tiny DNA brain in the form of encoded strands of DNA.
To quiz the brain, a human player placed DNA strands that hinted at the answer into the test tube. With these clues, the neural network was able to produce the correct answer, which was visible thanks to the fluorescent markers.
In this way, the network could also communicate when it lacked enough information to correctly identify one of the scientists, or if any of the clues contained contradictory information.
The research team played this game using 27 possible ways of answering questions and the neural network in the test tube answered correctly each time.
The team published their results in the July 21 issue of the journal Nature.
[Via Laboratory Journal]