Google’s Next Frontier – a New AI Chip
Artificial intelligence, AI for short, is almost overtaking software to become the preferred technology for all operations. This is not a surprising trend because AI is the next generation of machines that come with the ability to learn and implement certain behavior without needing a constant intervention from humans. AI is becoming a reality sooner than we may have guessed, and Google wants to lead the way.
During the company’s annual conference, its CEO, Sundar Pichai, announced that the company will be releasing a new computer chip that can perform advanced machine learning and AI tasks. Called Cloud Tensor processing Unit, this chip is named after Google’s open-source machine learning framework called TensorFlow.
Machine learning, deep learning, supercomputers and AI technologies have been transforming tech companies and its clients over the last few years, and this announcement from Google has made this trend official now. In many ways, AI is transforming Google and its operations too, so it is only natural that this company wants to set the trend and capture the AI market even when it is in its nascent stages.
So, what’s this new chip all about? What’s new in it? Well, lots!
This chip is being dubbed as the first one that can work at blistering speed not only in its executions, but also in its ability to learn. In other words, this chip can be trained incredibly fast and this can be a vital difference that can set this chip apart from others in the same category.
Let’s take a simple example. You want a machine to identify a pizza from other foods such as hot dogs, salads and burgers. To do this, you’ll have to feed in hundreds of different images of pizzas along with other foods to help the system to learn. The sheer amount of calculations needed to train such a model is mind-boggling and complex, so it can take days or even weeks for a system to identify a pizza from other foods.
Google wants to simplify this process, so the learning process is shortened greatly. To do this, the company plans to create machine learning supercomputers called cloud TPU pods. Many Cloud TPUs will be wired together using high-speed data connection. As a result, the learning will be split across different TPUs and they will happen in parallel, thereby leading to a shorter learning curve.
According to Pichai, Google wants to create thousands of TPUs and make them available over the Internet so that it can benefit researchers and developers in a big way. To start with, Google announced that it will share 1,000 TPUs with artificial intelligence researchers to help them in their studies.
In addition to this AI chip, Google plans to create algorithms that will fine-tune other machine learning algorithms and was developing AI tools for advanced studies such as genome analysis, molecular discovery and more.
In all, this is a great initiative from Google and one that could potentially transform our society in a big way. In return, Google can emerge as the leader of the AI chip market, right from its very beginning.