Google trained artificial intelligence to design computer chips. Not only is it comparable to those designed by humans, but AI can do it in hours instead of months.
Artificial Intelligence has long been a topic of debate. Initially, the question was, do we want to go down this road and teach machines to make decisions? Now that AI has been around for a bit and we’ve seen it make poor or biased decisions, the question becomes, can AI really be trained to think like a human and make decisions? In a way, the answer is both yes and no, but Google’s AI recently designed a chip in a fraction of the time it takes a human to do so. And while the layouts might be different, the optimization of components by AI was comparable or better than those designed by humans.
Earlier this month, a paper was published in the scientific journal, Nature. In it, researchers outline their processes and procedures for their purposes. We’re not going to get into the nitty gritty of the technical details on this, you can find a great breakdown of it on Venture Beat, but the idea is around AI performing chip floorplanning. Chip floorplanning is “the engineering task of designing the physical layout of a computer chip.”
Somehow, through all of the innovation around automation the world has seen over the last decade, chip floorplanning methods have escaped evolution. Currently, it takes months for physical design engineers to come up with layouts that can be reproduced and manufactured. This is because chips have to be designed with components laid out in the most optimal way. We all want our chips to be fast, work right and be efficient, right? Placement of components matters. Not only that, but replication of the design is imperative if we don’t want to design every single chip we manufacture. It’s a painstaking, laborious and often frustrating process for humans.
So, researchers took it upon themselves to see if they could train AI to speed up the process. Turns out, they can. And they can do it in under six hours, shaving months of time and hours of labor costs. The team used deep reinforcement learning to achieve this goal, and successfully designed the next generation of Tensor Processing Units, which are Google’s AI processors. AI designed a chip that will help advance AI.
Using software to assist in chip floorplanning is nothing new, but it’s never been done this quickly. Humans often overcome the limitations of our brains by shortcuts we’ve learned over time, but we aren’t good at tackling difficult tasks in one chunk. We think and function better when we work through parts of a complicated task. Our brains automatically find the order of operations that makes the most sense, break it down into pieces and then we get started. But AI doesn’t need to do that. The limitations of AI are nowhere near the same as those of our brains, it can tackle the whole problem at once because that’s what it’s designed to do. This doesn’t make it “better” than humans, but it does make it more efficient.
Now, with all of that said, let’s discuss the implications of this. We are currently enduring a massive chip shortage. This technology won’t help us with that right now, but if AI can lessen the time it takes to design a chip, then it will take less time to produce from start to finish. Which means that as we advance AI to design other types of chips, we can use the humans who were tasked with that to make knowledge-based decisions and complete tasks a computer cannot complete. All of this leads to higher efficiency rates, and probably shorter wait times for companies who need said chips.
While some may see this innovation as “robots taking over,” it most certainly is not. AI is not going to replace humans, certainly not anytime soon. Why? Because innovations in technology, any technological space but especially AI, requires more outside-the-box thinking. It’s about creating AI algorithms that enhance the knowledge that humans already have. As Venture Beat said, “It is rather a manifestation of humans finding ways to use AI as a prop to overcome their own cognitive limits and extend their capabilities. If there’s a virtuous cycle, it’s one of AI and humans finding better ways to cooperate.”
Artificial Intelligence isn’t going anywhere, nor should we want it to go anywhere. AI, along with other technological advancements, is furthering our abilities to foster knowledge, to find new ways to make renewable energy, to find answers to problems we’ve asked ourselves for centuries, to bring solutions to problems like pollution. The more we learn from technology and the more we learn how to make it work for us, the sooner we will be on a better path to keep our planet and its people healthy and happy.