
Google’s AutoML
In May 2017, MIT Technology Review reported on Google’s Annual Developer Conference and how Google CEO Sundar Pichai unveiled the company’s latest project–AutoML (1). This AI software is the creation of the Google Brain Team (2). This research group focused on developing AI that is “deep learning”.
Deep Learning is the newest phrase for neural networks also known as “deep structured learning, hierarchical learning or deep machine learning”. This type of AI has the ability to learn algorithms for specific tasks, such as “object recognition in the field of computer vision.” (3).
Many of the tools designed by Google Brain are currently used by other applications, such as Google Search, Google Maps and Street View, Google Translate, YouTube and other products.
The Future Is Now
AutoML, simply put, is Artificial Intelligence (AI) that can create other AIs. This is done through the automation of complex processes. The AutoML can be used to help developers that might not have all the necessary skills to complete a project. The project can be accomplished by assigning specific tasks for the AI to develop and solve.
MIT Review states, “In some cases, their [Google Brain Team] automated system came up with designs that rivals or beats the best work of human machine-learning experts.”
CEO Pichai told the conference attendees that Google’s AI promises to accelerate the AI field by giving researchers the ability to complete “some of the most challenging problems we face today.”
Making Do with Fewer Experts
Perhaps the biggest advantage to using Google’s user-friendly AI is the potential to “expand the number of developers able to make use of machine learning…” This will be possible since the AI learn during the process and grow in its knowledge of specific tasks.

Google’s goal is to offer AutoML as a cloud service where it can be used “to build and host with machine learning.” For example, a developer could theoretically create a model software with the help of the Google AI. The AI would supply the expertise the developer might require so the project can be successfully completed. With a limited number of human experts, the AI could fill in that much needed skill level.
While the AI would essentially learn as it worked on the project, it could then evaluate the processes and even offer solutions and improvements. This AI collaboration means developers could possibly take on projects they’d normally not be equipped to tackle.
Deep Learning Technology
MIT Review explains that the AutoML uses a technique known as “deep learning”. This technology is used in speech and image recognition software as well as robotic technology and even translation programs.
The AI deep learning actually serves to teach the software how to be smarter. The complexity of the system is summed up as, “…passing data through layers of math loosely inspired by biology and known as artificial neural networks.”

MIT Review explains that it’s up to developers to decide which architecture is best for the “neural network’s web of math.” The architecture has to be right or it simply won’t work.
AI Surprises Team
In one experiment, the Google Brain Team presented the AI with the task of deciding which architecture was the best to use so that the software could learn “to solve language and image-recognition tasks.”
The team didn’t expect the kind of results they received. The AI performed far better than anticipated. MIT Review explains that for the image task of the challenge, the system “rivaled the best architectures designed by human experts.” When it came to the language task of the challenge, the team reported that the AI beat them.
MIT Review states that when the AI approached the architectures and decided on the best one that it presented “a kind that researchers didn’t previously consider suited to those tasks.”
Endless Possibilities
With deep learning AI like Google’s AutoML, the possibilities of where this technology can take science seem endless. If AI can create other AIs and find solutions to problems by out-thinking humans, there is the real possibility that AI-driven machines might one day make humans obsolete in the process (4).
References & Image Credits:
(1) Technology Review
(2) Google Research
(3) Wikipedia
(4) AI May Kill off Humans