Overview

Artificial intelligence implemented in all our devices has become a part of our modern life without going unnoticed. It is in entertainment applications, medical diagnoses, and our bank’s security system, to name a few. But the incredible thing about this technology is its capacity to reproduce the human brain’s neural processes faithfully. And now, thanks to a team of researchers who designed a super-fast processor, it can do it a million times faster than people. Has the day come when a computer can think faster and better than us?

The science and other stuff to know

Standard machine learning works by feeding a bunch of data into a database and setting specific parameters so the algorithm can process and differentiate between them. After analyzing a particular amount of data, the algorithm begins to recognize patterns, to learn. So it is capable of improving the effectiveness of its tasks by evolving.

A new type of machine learning called analog deep learning proposes a more power-efficient and mind-bogglingly faster way to achieve this goal. A team of MIT researchers developed a processor comparable to the brain synapse, but it can process information much faster and uses less energy than ordinary processors.

In their article published in Science, the team ensures that the durability of this processor is far superior to others because it uses a resistant organic material that does not deteriorate when the proton flow that emulates the synapse passes over it. In addition, it can operate at room temperature — which means huge savings in cooling and fabrication of superconducting materials — this processor simultaneously computes a massive amount of data in a millionth of a second.

Murat Onen is one of the authors of the article. He said in a press release for MIT News: “The speed was certainly astounding. Normally, we wouldn’t apply such extreme fields to the devices so as not to burn them to cinders. But instead, the protons ended up traveling at immense speeds through the stack of devices, specifically a million times faster than we had before. And this movement doesn’t damage anything, thanks to the small size and low mass of the protons. It’s almost like teleporting.”

So what?

With cheap, efficient materials and a fast and robust processing network, complex data processing can be achieved in real-time, whose applications in medical images could be revolutionary for the clinical field. Another possible application is in autonomous cars and even aircraft.

“Once you have an analog processor, [you’re no longer training] networks that everyone else is working on. You’ll be training networks with unprecedented complexities that no one else can [afford], and [thus, you’ll] vastly outperform them all. In other words, [this isn’t a faster car; it’s a spaceship],” tells Onen.

What’s next?

Now that they have proved the materials’ benefits and the techniques implemented in the hardware, the experts consider that the next step is to plan larger-scale manufacturing and study the device’s potential for other systems integration. They have also started collaborating with the MIT Quest for Intelligence to apply this knowledge to neuroscience. “The collaboration [we have] is going to be essential to innovate in the future. [So] the way forward will continue to be very challenging, but at the same [time,] it is very exciting,” says Jesús A. del Alamo, co-author of the article.