The future of AI may look a little less I, Robot and a little more Blade Runner.
Researchers at Caltech have built an artificial neural network made exclusively from manmade organic matter. One day, a more advanced version of the network could potentially diagnose diseases, make decisions, and even forge its own memories. For now, it is able to categorize handwritten numbers, as shown in a study published in the journal Nature.
Like all neural networks, it mimics processes that occur naturally in the human brain.
“Humans each have over 80 billion neurons in the brain, with which they make highly sophisticated decisions. Smaller animals such as roundworms can make simpler decisions using just a few hundred neurons,” Lulu Qian, assistant professor of bioengineering at Caltech, said in a statement.
“In this work, we have designed and created biochemical circuits that function like a small network of neurons to classify molecular information substantially more complex than previously possible.”
Unlike the vast majority of neural networks, it was developed in a test tube, is made from synthetic DNA, and resembles a “smart soup”.
Why DNA? Single strands of DNA are built from the same four molecules (or nucleotides) – A, T, C, and G. This makes their reactions extremely easy to predict, whether or not they evolved naturally or were built in a lab.
To show that artificial intelligence can be “programmed” into synthetic biomolecular circuits, the team tested their creation using the handwriting challenge – a test that’s extensively more difficult than it first appears thanks to the many variations and, occasionally, sloppiness of human handwriting.
Instead of using “visual handwriting”, the team used a technique called “molecular handwriting”. This means that the writing does not take the shape of a number or a letter. Rather, every single molecular number is comprised of 20 unique DNA strands, each selected from 100 molecules representing different pixels in any 10-by-10 pattern, that have been mixed together in a test tube. The neural network is able to identify the molecular number as one of nine digits between 1 and 9.
The neural network was first put to work on a simple model that only required it to separate two digits, 6s and 7s. Using a so-called “winner takes all” approach and a type of DNA molecule nicknamed “the annihilator”, it correctly identified 36 out of 36 handwritten numbers.
“The annihilator forms a complex with one molecule from one competitor and one molecule from a different competitor and reacts to form inert, unreactive species,” first author and graduate student Kevin Cherry explained.
“The annihilator quickly eats up all of the competitor molecules until only a single competitor species remains. The winning competitor is then restored to a high concentration and produces a fluorescent signal indicating the networks’ decision.”
Next, it upgraded to a more advanced model requiring it to sort between nine digits between 1 and 9. Again, it was successful.
While it may be some time before we see anything vaguely Blade Runner-esque, it could pave the way for a future where AI is more organic than mechanical.
“Though scientists have only just begun to explore creating artificial intelligence in molecular machines, its potential is already undeniable,” Qian added.
“Similar to how electronic computers and smartphones have made humans more capable than a hundred years ago, artificial molecular machines could make all things made of molecules, perhaps including even paint and bandages, more capable and more responsive to the environment in the hundred years to come.”