According to the Financial Times, Facebook is working on producing artificial intelligence (AI) chips for a virtual assistant to be smarter than Apple’s Siri or Amazon’s Alexa
Facebook is working with Intel, other tech giants and semiconductor makers to make this goal come true.
It’s goals in developing a more specialized and powerful AI chip is so that a digital assistant will have enough ‘common sense’ to communicate with a human on any subject as well as making it a practical tool in regards to controlling its social network; for instance in the area of monitoring videos and deciding what content should be allowed on its service.
Facebook has in the past been known to make its own hardware when necessary and is currently developing its own custom ‘application-specific integrated circuit’ (ASIC) chips which would be a particular kind of transmission protocol designed to specifically work with the company’s AI programs.
Yann LeCun, Facebook’s chief AI scientist and one of the pioneers of modern AI, says there is a need for more specialized AI chips to speed up tasks at lightening speed but with lower power consumption. He says that the company is doing anything and everything it can do in this area especially in monitoring videos flowing through their site in real time but that it will essentially require neural designs which are computer systems modeled on the human brain and nervous system.
In particular, the company is looking to new neural network architecture that will mimic aspects of human intelligence to make its systems more natural for users to interact with. Especially systems that would operate on a ‘common sense’ level, particularly for digital assistants which would have background knowledge in order for users to be able to have discussions on any topic with them, says LeCun.
Another area of research and development is in the area of actually making these systems work just like the brain; networks ‘firing’ up neurons to solve a specific problem which would greatly reduce power consumption, he said.
All in all, these advances for the future function of neural networks will have an impact on the design of the AI chips that power them which will potentially open up the competitionto all the companies that manufacture today’s top selling AI chips.