"Scientists are still suspicious of AI"

Naveen Rao  photo: Intel

Intel VP and Artificial Intelligence Products Group general manager Naveen Rao explains in an exclusive "Globes" interview why the company has marked AI as strategic.

In March 2016, Google's Alphago artificial intelligence (AI) program stunned the world by beating the human world champion Go player in front of 200 million spectators. This was living proof of the potential in AI technology and the level of maturity reached by neural network and deep learning technologies. Those astounded by the success included quite a few engineers and managers who have been leading the AI revolution in the world in recent years. One of these was Intel VP Naveen Rao, general manager of the company's Artificial Intelligence Products Group, which was founded last year.

"When I studied at college in the 1990s, we regarded artificial intelligence as 'creative work'," Rao relates. "The game of Go was then considered an example of something akin to magic in the capabilities of the human brain - something impossible to do in computer program because of the enormous number of ramified possibilities. Back then, no one thought that it would ever be achieved."

Even now, when AI technology has proven achievements to its credit and is improving computer capabilities and performance in areas like image and voice processing, security, finance, etc., quite a few academics still tend to view it suspiciously as borderline mysticism. The typical attitude towards deep learning is "It's inelegant, because we don't understand exactly why it achieves the results that it does."

Rao, who has degrees in computer science and neurology, is familiar with this attitude, and takes a businesslike approach to it. "We're constantly improving our understanding of the neural network and how its internal mechanism works. When you teach a network, however, it devises the presentation of the information for itself - a kind of 'understanding' in non-human terms. We find it just as difficult to follow the presentation of information in the human brain, but there's no need to understand everything and program everything. When you play tennis, you don't think about how every neuron operates during the game. You just play. Just because you don't understand exactly how a neural network achieved the final result doesn't mean that it's not useful."

A veiled swipe at Nvidia

Like most of its competitors, Intel is entering the AI field now in order to supply the market's growing need to process the huge quantities of information being accumulated in the online digital age and produce value from it. "This is the next evolution of intelligence," Rao says. "Our brain has developed biologically, but it has now encountered the brick wall of information processing. There is so much information available now stored on the Internet and in information storage technologies - every click we make is saved somewhere - that we are unable to utilize it. AI, which expedites the process, is the right way. This is the next generation of evolution, and it's natural for Intel to be there."

One of the examples is Facebook, a partner and customer of Intel in developing AI capabilities. Facebook's AI engines process images and information put on the social network by almost two billion active users. They can identify images across the network, connect them, and produce commercially useful information, "and we're only scratching the surface," Rao adds.

Intel is a relatively late participant in the AI trend now taking over the global hardware and software industry, but it is determined to catch up in this arms race, in which it is going up against giants like Qualcomm and Nvidia. Intel is striving to establish a leading position in this strategic sector. One of the ways it is doing this is through massive investment in mergers and acquisitions. One prominent acquisition, on which Intel spent $600 million, was Nervana Systems, which Rao founded in 2014.

On the basis of Nervana's technology, Intel has developed a new family of chips called neural network processing (NNP) chips, officially launched this month, which is scheduled to reach customers before the end of the year. The new chips' capabilities are still very mysterious, not only because the products are based on completely new architecture, but also because Intel is not sharing many particulars about them, and is refraining from publishing performance data. Rao too offers few details about their future performance, saying, "There's no point right now in publishing data comparisons, because there's actually nothing to compare them with. Comparison with performance of other AI technologies is like comparing apples and oranges."

Nevertheless, anyone who wants to can sense a veiled swipe at Nvidia, Intel's great rival in this field, which also recently announced a breakthrough chip for AI, based on its graphic processing unit (GPU). "AI processing has hitherto taken place mostly using GPUs that were not originally designed for these tasks. Our product is the first designed from the beginning for parallel AI processing, and it has undergone a great many specific optimizations for this purpose. The architecture is non-modular and completely different. We're building a family of models for different purposes," Rao declares.

"Turning vehicles into robots will take time"

Intel is initially aiming its chips at giant data centers in order to speed up the process of filing and analyzing the data. Rao says that a development team at Intel's center in Israel is also currently engaged in this task. Other products from the same family will be designed for the "training" market of computers for self-teaching. "Training is very important," Rao remarks, "The eventual goal is for the computer to learn independently from the data and experience, instead of having to being programmed. The human mind also improves when it practices the same task over a prolonged period. Radiologists, for example, become more expert in reading X-ray and CT results during their careers. A computer can become much more expert by repeated training and 'self-training' at high speeds, when the economies of scale are enormous."

What about mobile applications, such as autonomous driving, on which Mobileye, recently acquired by Intel for $15 billion, is now working? Will we see an interface between these two worlds? Rao says that autonomous driving is a typical area in which AI is needed for processing a prodigious quantity of visual information produced from the vehicle and drawing immediate conclusions for driving, and for later analysis of the data streamed from the vehicle for tasks such as mapping, data commercialization, and so forth.

"When you're driving, you need to not only see road signs and the road situation; you also have to 'understand' and analyze the intentions of the other drivers around you. This is a task that requires a large amount of high-speed computing," Rao says.

According to Rao, the group he is responsible for will work closely with Mobileye in the future. "We already met with their team in Israel. They're very focused, and there is an overlap between the fields of activity. The job of making the vehicle into a robot will take time, but we're on the way to it. It's a process of training and acquiring confidence that will get better as the training continues. In the US, a 16 year-old can get a driving license. Do you trust a 16 year-old driver in front of you on the road?"

What about achieving a breakthrough through more mergers and acquisitions of Israeli companies operating in relevant sectors, even after the $15 billion spent by Intel on acquiring Mobileye in Israel? "Israel has played a very important role in Intel's history because of the innovation capabilities here," Rao says. "It won't stop in the near future. We're constantly considering the acquisition of new technologies, including through acquisitions. We're also one of the most competitive multinationals in Israel in personnel recruitment."

"There will always be human backup"

At the end of the interview, in a slightly more philosophical moment, we asked Rao for his opinion on public concern that uncontrolled AI technologies will make human beings redundant and have a negative influence on human society. "AI is only a tool for our use. People have always been afraid of new technologies," Rao says. "Human skills in some areas will have to change, but there will a need for new skills in order to produce value from the new technologies. The number of jobs will increase, and mankind will not become redundant. People will get used to the new technologies and develop expectations from them, just as they got used to smartphones and got used to expecting new performance from them.

"Right now, we're not yet as good as biology and evolution (which created the human brain), but we're nearing a stage in which computers will be able to perform many functions and leave people time to be creative. Furthermore, there will always be human backup for AI decisions, just as is the case with airplanes. The automatic pilot now flies the plane better and more safely, but it is still supervised by human pilots."

Published by Globes [online], Israel Business News - www.globes-online.com - on November 5, 2017

© Copyright of Globes Publisher Itonut (1983) Ltd. 2017

Naveen Rao  photo: Intel
Naveen Rao photo: Intel
Twitter Facebook Linkedin RSS Newsletters גלובס Israel Business Conference 2018