Artificial Intelligence

Google Is Teaching an AI to Create More AI

Whether you're fascinated by thoughts of a future where robots make our lives better or your dystopian visions of a world overrun by mechanical beings keep you up at night, there's no question that artificial intelligence is here to stay. At the recent annual Google I/O conference the company introduced what may be its most jaw-dropping AI innovation yet, AutoML—a technology that can utilize neural networks to teach itself.

The Rise of Neural Networks

Inspired by the human brain and nervous system, neural networks are mathematical systems designed to analyze enormous amounts of data. Like a brain, neural networks rely on large numbers of interconnected "neurons" work at the same time to solve problems. But instead of following instructions like a traditional computer, they follow examples and "learn" to improve the rules and methods they use. That's where cutting edge fields like artificial intelligence, machine learning, and deep learning come in. Many companies are already using this technology to recognize images, recommend things you might like to watch or buy, or even predict the results of legal proceedings.

Machine learning (ML) is becoming so popular, in fact, that there's a shortage of experts who can make the software. It's an intensive process for scientists and engineers, and that has limited the use of neural networks and machine learning to small pools of computer experts and academics. That's where AutoML comes in. Its machine-learning software that can help create machine-learning software. With it, Google aims to address a shortage of machine-learning experts and further "democratize" AI technology across the development community and beyond.

AI Accelerated

Google's early testing of AutoML shows that the technology may be faster and more efficient at writing the layers of code necessary to create machine-learning neural networks that beat humans. For example, when researchers Quoc Le and Barrett Zoph tasked a machine-learning system to pinpoint which architecture to use to have software solve tasks such as recognizing images and languages, the system came up with a solution comparable to the best human-created solutions we have—and totally surpassed them on language recognition.

Google's AutoML is still in its early stages. But if the successes seen in the tests by Le and Zoph can be replicated, it has the potential to accelerate the entire field of AI faster than what was ever imaginable. And the more that people that have the ability to create customized neural networks, the greater ability the technology has to improve human lives.

Artificial Intelligence Explained

Written by Jamie Ludwig June 6, 2017

Curiosity uses cookies to improve site performance, for analytics and for advertising. By continuing to use our site, you accept our use of cookies, our Privacy Policy and Terms of Use.