Google seems to have managed to design an artificial intelligence program capable of designing other AI chips. Advancements in AI, machine learning and machine intelligence have been coming in fast recently with tech companies heavily investing in the industry to help cement their position.
AI, in one form or another, is actively shaping the world. Most people understand this, and the futuristic take on AI has proven popular thanks to shows like Westworld and movies like Ex Machina. Hollywood aside, AI and machine learning currently play a major role in driving essential devices like computers, cars and cameras. The technology is also being used for fighting crime as well as for surveillance. With digital activity today producing massive amounts of data, big tech companies like Facebook are even employing AI to regulate their massive content-generating platforms, as well as helping to further personalize it for the users.
Google's chip-making AI system is the latest in a long line of recent AI developments and, as IEEE reports, is an attempt to make chips best suited to the latest AI algorithms. The design, published in a research paper this week, employs deep learning and has been developed by Google Brain researchers - the tech giant's deep learning artificial intelligence research team. Google has been a major investor in the technology and uses some of the most powerful machine learning algorithms available to manage the popular search engine and regulate videos on YouTube. Similarly, the new AI system is now expected to have an impact on the chip-making process.
How An AI Chip Makes An AI Chip
The new system drastically simplifies the process of making chips that go into Google's Tensor Processing Unit AI integrated circuits, by taking over the process of Placement. This is the process by which components that makes up a chip have to placed in a circuit in such a way that it results in maximum power and performance, while keeping the chip as small as possible. For integrated circuits, this has to be done in a synchronized fashion and that makes it even more challenging. Placement is the task of a human expert and can take weeks to get right. Google researchers designed an AI program to study and replicate this process using a deep learning algorithm so that it can be done within 24 hours. This huge reduction in time can create a big difference in the performance of various AI algorithms as up until now, chip designs have been lagging behind machine learning advancement. By using AI algorithms to make chips used for AI projects, researchers hope to keep the chip design on par with algorithm development.
This path-breaking achievement might accelerate AI development while improving efficiency and reducing cost. However, Google is not the only company working on AI chips. Intel recently developed an AI chip capable of identifying various smells for improved chemical detectors. Both of which are prime examples of the different, but similar breakthroughs people can expect to see in the artificial intelligence space going forward.
Source: IEEE Spectrum