Time： 2018-11-28 16:14:27
Kurt Sievers, president of NXP Semiconductors, said at a CEO roundtable at electronica in Munich that AI is overhyped today . “It's this miraculous thing where nobody knows what it is," he said. "Most people wouldn't even know what the definition is."
Sievers and the CEOs of STMicroelectronics, Infineon Technologies and Mentor Graphics, along with the CTO of Alibaba’s IoT unit and the automotive segment managing director of Elektrobit, discussed the reality of AI and its actual utility. Most concurred that AI is more of an enabling tool for applications such as IoT and autonomous driving.
“The IoT was the hype thing two years ago. It hasn't really happened so far," Sievers said. "But we believe with connectivity — and with sensor technologies — AI is actually going to make the IoT work.”
Reinhard Ploss, CEO of Infineon Technologies, said, “AI is a fusion of the data you have available and all the algorithms behind it.”
The CEOs tried to bring AI down to earth, explaining where it fits and separating hype from reality.
“AI at the edge is an incremental innovation," said Jean-Marc Chery, president and CEO of STMicroelectronics. "It’s not the radical innovation that everybody thinks it is."
Chery talked about how ST is already using some form of AI in manufacturing, explaining how software with a neural network component collects wafer defect and process parameter data, looks for patterns, and can anticipate failure mechanisms. “You need very strong engineering skill to understand the root cause of a failure," he said. "So, we know at ST what you can do with AI and what you should not expect from AI.”
Chery said AI is an accelerator and enabler, making systems more efficient and accelerating decision making. But AI will not take over the role of a human.
“AI can’t take away the innovative mindset [of humans],” said Dean (Xianfeng) Ding, CTO of the IoT unit at Alibaba.
This sentiment was was echoed by Ploss. “AI won’t be creating new wisdom," Ploss said. "It can help or hint to where a problem is. Or, in R&D, AI can be used to accelerate the development process. A system trained in one area can help us improve in that area.”
On this aspect, a key point made by the panel was that AI involves rule-based inference, and that anything requiring the use of unknown data cannot be driven by AI. In other words, the computer cannot be better than a human.
The panel discussed various geographic differences. The thinking was that the U.S. has been leading AI from a scientific perspective, but China is leading from an implementation perspective, simply because of the sheer number of people and the depth of data available through all those people using things like social media. Panelists said that where Europe can make a difference is by providing a security and privacy perspective to AI-based applications.
Automotive is clearly a key application for AI. But visions and expectations of fully autonomous Level 5 capability are now being toned down, according to Lars Reger, NXP’s automotive CTO. We sat down with him at electronica to understand where he was coming from.
Reger said that the "level" discussion in automotive was moving away from Level 5, with everyone suddenly calling for Level 3.
“Reality is kicking in," Reger said. "We’re going through a perfect hype cycle, and now on the downward slope.”
In both automotive and IoT, customers are asking NXP to build appropriate functionality — for example, in IoT they are looking at energy efficient processing at the right performance point.
Reger also talked about building platforms that can be utilized across different domains, as part of NXP’s strategy to move both up the value chain as well as work with new market entrants.
“We are moving to an on-demand culture in society," Reger said. "Look at car sharing for example – it makes sense in a new age where 'ownership' is looked down on. But how far does an on-demand culture go? You can have on-demand cars, on-demand healthcare, and so on. But what is the end point?”
This is where Reger believes there is a need to be nimble enough to service new customers that don’t have domain knowledge. These customers may be creating new business models in smart homes, smart factories or smart mobility. But they won’t necessarily have the capability to deliver the technology or systems needed to form the end points.
Reger says this is why NXP is moving to create more reference designs. “We can have the same platform but for different domains," Reger said.
A reference design for automotive with its high-performance compute engine can easily be translated to other applications, Reger said. Hence NXP is setting up reference design centers and providing reference materials to enable smaller, less experienced customers to create the advanced systems that previously only players with expert domain knowledge were able to.
So is AI overhyped?
The bottom line is that in the world of electronics systems design, AI is probably not exactly overhyped. As the CEOs articulated at electronica, there’s a good dose of reality in the ecosystem, and the chip vendors clearly know the real challenges faced by their customers and what type of AI engines or chips can tackle them.
Engineers are a smart (and pragmatic) lot, and the wool can’t be pulled over their eyes. Where the hype does come is probably more in the mainstream media, where imaginations have run wild about AI being equivalent to smart intelligent computers and robots with minds of their own taking over in everything. That’s just not going to happen — at least not for a few years yet.