By Charles King, Pund-IT, Inc. November 22, 2016
There are natural synergies between the evolution of artificial intelligence (AI) and Intel. After all, the company’s microprocessors and other technologies power the vast majority of the enterprise systems and cloud computing infrastructures where a great deal of pioneering AI research and development is being performed. Intel-based PCs are also the targets of or necessary for many of the initial business and consumer grade AI services that are coming to market.
However, at the Intel AI Day event the company hosted last week in San Francisco, Intel stepped forward to clarify what it will do to foster future AI innovation. What I saw there, as part of a sizable audience of analysts and reporters, is that Intel understands the key role that it and solutions based on its products will play in successfully bringing AI to commercial markets. The company also described next steps and advancements that should help move AI fully into the mainstream. Finally, Intel illustrated its points with business- and consumer-focused use cases where AI technologies are and should continue to be welcome.
Setting-up the AI taxonomy
Intel’s CEO Brian Krzanich began his opening keynote with a short video explicating the interdependencies between AI and related machine learning (ML) and deep learning (DL) technologies and processes. That suggests an admirable level of care within the company for getting its message and messaging right. But it also underscores the “squishiness” that pervades many marketing efforts and public discussions around AI research, products, services and themes.
Intel’s approach also reflected on competitive issues and differences within the vendor community. “AI” is a term that can be and is used to describe often subtle and sometimes radically different projects pursued by major vendors, including Amazon, Facebook, Google, IBM, Intel, Microsoft and NVIDIA, along with a host of smaller specialty players. Though these efforts certainly overlap one another, there are also significant, even unique differences between them.
So how does Intel define AI? To quote Bob Rogus, chief data scientist for Intel’s analytics and AI group, as technologies that can “Sense, reason, act and adapt.” That places the company pretty much in-line with other vendors. For example, IBM describes cognitive computing solutions, including its Watson platform as those capable of “Understanding, reasoning, interacting and learning.”
Intel follows a different path than some competitors in that it bifurcates AI into machine learning (including DL processes) and what it calls “reasoning systems.” Both leverage foundational technologies provided by recent Intel acquisitions. In the former case, Krzanich highlighted Movidius’ efforts in computer vision and Nervana’s work in DL and learning networks, while Saffron Technologies is doing pioneering work in reasoning systems.
AI and IA together –
On one level, the call to battle in Krzanich’s keynote – that “Intel is guiding the AI computing era” – is a statement of remarkable ambition but also makes sense given the company’s role in developing and delivering innovative tools for OEM’s and other customers. It also casts light on how Intel plans to further expand and enhance the massive strides that have been made in computer perceptual technologies, particularly the reduction in error rates in vision and speech recognition, during the past five years.
Diane Bryant, SVP and GM of Intel’s Data Center Group, detailed the company’s AI-related data center offerings, including new and upcoming silicon solutions that significantly improve the performance of ML, DL and advanced analytics workloads. For example, the new Intel Xeon E5 processors (shipping in production versions to select cloud providers and slated for general availability next year) have delivered up to an 18X improvement in Apache Spark analytics workloads over the past 3 years.
Similarly, the next generation “Knight’s Mill” version of Intel Phi solutions for highly parallel processing is optimized for deep learning workloads, delivering up to 4X better performance than current Intel Phi. AI is also the central target for the “Lake Crest” solutions that the company will begin testing in early 2017 (with availability planned for later in the year) and “Knights Crest” which will combine Xeon silicon with Nervana Engine technologies and a high performance interconnect fabric to support neural networks.
Bryant and other company executives said that by 2020, Intel will be delivering silicon enabling reductions in DL training times that will be 100X faster than today’s best GPU-based solutions. If the company achieves that goal it will provide a significant competitive advantage. Plus, it will advance Intel’s larger goal of “democratizing” AI by lowering dramatically associated costs and allowing vendors, developers and businesses of every sort to pursue AI development.
One company that notably shares this Intel vision is Google, as evidenced by the appearance of Diane Greene, SVP of its cloud business (and formerly co-founder and CEO of VMware). Greene joined Bryant onstage to announce a new strategic alliance between the companies that emphasizes “AI Everywhere.” As part of these efforts, services and efforts, including Google Cloud, Google Deep Mind, Tensor Flow and Alpha Go will all be optimized for the Intel Architecture (IA).
So what were my primary takeaways from Intel’s AI Day event? First and foremost, that the company is bringing a sizable number and variety of assets to bear on achieving its goals. This includes Intel Xeon and Phi silicon solutions that have been engineered for specific artificial intelligence, machine learning and deep learning processes, of course. In addition, the company is doing impressive work in software development and toolkits, and making significant contributions to AI- and advanced analytics-related open source projects.
Intel’s role as a purveyor of component technologies for OEMs and other manufacturers means that the company has fewer competitive complexities to deal with than do vendors developing and promoting their own AI solutions, services and platforms. Interestingly, that may explain some of the unique messaging points and terminology Intel used during the event.
For example, Gayle Shephard, formerly Saffron’s chairman and CEO and now a GM within Intel’s New Devices Group, described Saffron as a “cognitive computing provider” and her group’s role in building cognitive tools and capabilities on Intel’s hardware framework. The difference between cognitive computing and what other executives called “reasoning systems” may simply be rhetorical. Or it could indicate how Intel is differentiating its own AI solutions from competitors, like IBM and Microsoft, that are promoting their own homegrown cognitive computing platforms and services.
Overall, Intel’s AI Day offered the opportunity to learn about its vision of the current condition of artificial intelligence technologies and their future promise. More importantly, the company provided analysts and reporters deeper insights into how the company plans to help customers get from here to there, bringing AI capabilities fully into commercial markets, applications and use cases.
The goals laid out by CEO Brian Krzanich, SVP Diane Bryant and other Intel executives, along with partners like Google’s Diane Greene, were massively ambitious and wide reaching. As a result, the company’s progress is likely to be examined closely and dissected by competitors hoping for signs of weakness or outright failures. That said, if Intel can deliver on the details and plans it shared in San Francisco, the company’s vision of democratized AI Everywhere seems imminently achievable.
© 2016 Pund-IT, Inc. All rights reserved.