Massively parallel supercomputing hardware and advanced artificial intelligence algorithms are being harnessed to deliver powerful new research tools in science and medicine, according to Dr. France A. Córdova, director of the National Science Foundation.

Córdova spoke Oct. 26 at the GPU Technology Conference organized by Nvidia, a company that got its start making video cards for PCs and gaming systems and now manufactures advanced graphics processors for high-performance servers and supercomputers.

Córdova, who is directing long-term research in AI at the NSF, said the research there is being used already in the Cancer Moonshot project currently spearheaded by Vice President Joe Biden, whose son Beau Biden died of brain cancer in 2015 at the age of 46. The Cancer Moonshot is a major effort to focus resources and funding on the fight to cure cancer on a scale similar to the original mission by NASA to land on the moon.

The future of AI is a major focus by the Obama administration, which recently released a multi-year report, "Preparing for the Future of Artificial Intelligence." Dr. Córdova said similar efforts are moving forward in the study of quantum physics and a related field, quantum computing, which she said has the potential to revolutionize research by bringing to bear computing resources with unprecedented processing power.

The problem, she said, is that industry needs to do more than it has so far. "We're under-investing in AI," Córdova said. She added that all of the work in AI depends on the availability of big data to do its work. "AI feeds on data."

The level of effort being put forth by the NSF on AI was explained later by Lynne Parker, NSF division director for information and intelligence systems. "NSF has been making significant research investments in AI systems that are more robust and flexible in complex, realistic environments. One existing NSF program focused on this research challenge is the Robust Intelligence program," Parker said in an email to eWEEK.

"This program has specific goals of creating AI systems that are characterized by flexibility and resourcefulness, among other traits. Of course, it is very difficult to create truly robust AI systems, so much additional research is needed, as is called out in the National AI R&D Strategic Plan," Parker wrote.

Córdova illustrated common uses of AI by referring to Apple's Siri, who she briefly interviewed in preparing her presentation. She used Siri to demonstrate both how good AI has become, but also to demonstrate its limitations. When she asked Siri to dance, for example, the digital assistant demurred, saying it would "sit this one out."

The overall focus of the GPU Technology Conference was on computing systems built around GPUs built by NVidia. These processing units are designed to complement a computer's CPU by providing vast numbers of cores capable of very fast floating point calculations. According to NVidia Chief Scientist Bill Dally, the GPUs being produced by the company are reaching levels of 20 teraflops of 16-bit floating point calculations per second in a new server built by IBM.

Dally said the company's new Pascal DGX-1 supercomputer is 65 times as fast as the immediately preceding generation of GPUs. This level of computing, he noted, is providing new capabilities to internet of things (IoT) devices such as Amazon's Echo. He also said that because these supercomputing devices can be installed in vehicles, they're playing a significant role in supporting autonomous vehicles.

One company that's leveraging the raw computing power of GPU-based devices is MapD, which has built a set of GPU-optimized databases to handle mapping data. But the company also found another big data application that might strike fear into the hearts of official Washington: an AI application that can tie diverse information on political contributions with voting records.

A spokesperson for MapD quietly showed me how the new application may be able to ferret out legislators who are voting based on the contributions they receive, despite efforts to funnel the money through political action committees and other third-party groups. They were able to show me where such donations originated on a map and demonstrate how the money made its way to the lawmakers.

Of course, there are other ways to mine such vast quantities of data. The company was able to show me geo-tagged tweets in real time, including the tweet from a White House visitor as he wandered into the West Wing for the first time.

While some of those capabilities are whimsical, others, such as the work with autonomous vehicles, will have important social and economic impacts. Still others, such as the computing done in support of the Cancer Moonshot, can transform lives.

While the folks from NVidia were careful not to overstate the capabilities of this previously unimaginable level of readily accessible and reasonably priced computing resources, it's clear that any attempts to guess what the future might be already are falling short.

Already there are computing resources that build their own neural networks so they can learn how to perform activities simply by watching someone do them. While there are still some limitations—after all, they still have to be programmed—those limits are falling away surprisingly quickly.

While it's impossible to know exactly where this new combination of data, computing resources and AI algorithms will end up, it's clear that the pace at which it's proceeding is gaining speed.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved