Sometimes You Don’t Need Deep Learning: Eye on A.I.
Ibrahim Gokcen, the digital chief technology officer for industrial giant Schneider Electric, has some words of caution about deep learning—the latest craze in artificial intelligence. Sometimes, conventional data crunching works just fine.
All of the technology sold by Schneider that warns corporate customers when their industrial equipment may fail uses basic analytics or statistical analysis to make predictions. Although the software incorporates machine learning, it doesn’t use deep learning, a technology that has led to breakthroughs in image and language translation.
But that’s okay, Gokcen explained.
Traditional data analytics and machine learning work perfectly well for Schneider Electric, the 180-year old company that is known for its circuit breakers and other electrical equipment. Additionally, using older data crunching technology makes it easier for the company’s data scientists to understand how those systems reach their conclusions—a challenge for even the best deep-learning experts.
Schneider Electric’s strategy is to update its existing products, used in power plants and factories, to be more digital. The goal is to siphon data from industrial equipment and then apply basic analytics and machine learning, so it can sell customers newer services, like warning them when a circuit breaker will likely die.
Companies have been marketing these so-called predictive maintenance services for a few years, but it’s a tough market to crack. Although businesses are interested in predictive maintenance, it’s hard for executives to justify spending a lot on it because the technology is typically targeted at only one thing, like gas turbines, instead of an entire factory, Gokcen said.
Many executives expect “amazing results to the business,” he explained. The reality, however, is that the savings are often too modest to justify the high cost. For big savings, companies must have all of their industrial equipment “digitized” so that the machinery can be analyzed together, Gokcen said. The potential for costs savings therefore multiplies.
Schneider Electric is exploring how deep learning could help it create new predictive maintenance nirvana. But the technology isn’t ready.
Still, the excitement over deep learning is infectious, and Schneider Electric will continue testing the technology, Gokcen promised. After all, there are a lot of free, open-source deep-learning tools available that a younger Gokcen could only have dreamed about.
“Suddenly you have a tool that I would have spent two years building as a PhD,” Gokcen said. “As an A.I. researcher, I couldn’t have imagined the breakthroughs we have today.”
A.I. IN THE NEWS
Fake it ‘till you make it. The Wall Street Journal examined the small A.I. startup Engineer.ai and discovered that that the company “exaggerates its AI capabilities to attract customers and investors.” Instead of using A.I. technologies to help companies build apps, the Journal reported that “the company relies on human engineers in India and elsewhere to do most of that work, and that its AI claims are inflated even in light of the fake-it-till-you-make-it mentality common among tech startups.”
Understanding language. Researchers from Facebook, the University of Washington, New York University, and Google’s DeepMind unit debuted the SuperGLUE benchmark, a series of tests intended for researchers to examine the performance of A.I. systems that can understand language. The new benchmark will succeed an older version and include “a new set of more difficult language understanding tasks, improved resources, and a new public leaderboard,” the group wrote.
Bernie Sanders is not a fan of facial-recognition. Democratic presidential candidate Bernie Sanders said he wants to ban law enforcement from using facial-recognition technology, CNN reported. A spokesperson for Sanders’s presidential campaign told the news service, "Police use of facial recognition software is the latest example of Orwellian technology that violates our privacy and civil liberties under the guise of public safety and it must stop."
The workers who label data for a living. The New York Times visited the Indian and New Orleans offices of data labeling company iMerit to shadow the human workers responsible for prepping data that will eventually be used to train A.I. systems. As the Times notes, “A.I. researchers hope they can build systems that can learn from smaller amounts of data. But for the foreseeable future, human labor is essential.”
HE WOULD SAY THAT, WOULDN’T HE?
Nvidia CEO Jensen Huang talked to tech news site VentureBeat about the booming A.I. market and his belief “that AI is the largest technology force of our time.” Nvidia is trying to capitalize on the A.I. market by selling companies its graphics processing units (GPUs) used to power deep learning. “The automation of automation, the automation of intelligence, is such an incredible idea that if we could continue to improve this capability, the applications are really quite boundless,” Huang said. “When you think about the size of the intelligence market — how big is the intelligence market? It’s measured in trillions of dollars.”
EYE ON A.I. TALENT
Microsoft hired Bill Stasior to be corporate vice president of technology and will lead an A.I. group, tech publication The Information reported. Stasior was previously an Apple vice president leading a team responsible for the Siri digital assistant.
Graham Media Group picked Anthony Plosz to be the broadcasting company’s vice president and chief technology officer. Plosz was previously a vice president of engineering and operations at NBCUniversal Media.
EYE ON A.I. RESEARCH
How bias creeps into hate speech A.I. tools. Researchers from the University of Washington, Carnegie Mellon, and the Allen Institute for Artificial Intelligence published a paper detailing how A.I.-powered hate speech detection tools can develop racial biases. The paper, presented in early August at the Association for Computational Linguistics conference in Florence, Italy, probes the Google-developed Perspective tool, which would classify phrases made in the African-American vernacular as toxic.
One of the problems, the researchers noted, is that the kind of A.I. that helps computers understand language often fails to understand context. Dan Keyserling, the chief officer of Alphabet-subsidiary Jigsaw, which helped build the tool, told Fortune that Jigsaw is working with the researchers to improve it’s A.I. content-moderation tools. “The technology will never be perfect,” Keyserling said. “That is the nature of machine-learning research.”
FORTUNE ON A.I.
Genpact CEO: Companies Have a Responsibility to Reduce A.I. Bias – By Damanick Dantes
Microsoft Will Continue Letting Workers Listen to Skype, Cortana Recordings – By David Z. Morris
That is one big chip. Journalist Tiernan Ray writes in Fortune about the startup Cerebras Systems, which has developed a huge computer chip that is intended to speed up the lengthy time it can take to train neural networks—the foundational software designed to loosely mimic how the brain learns. Time will tell if the startup’s chips will work as claimed. Currently there’s no benchmark to compare Cerebras chips with competing A.I. chips, and the startup isn’t disclosing performance statistics. But, the chip is giant by chip standards, and that’s interesting in itself. Ray writes:
Four hundred thousand little computers, known as “cores,” cover the chip’s surface. Ordinarily, they would each be cut into separate chips to yield multiple finished parts from a round silicon wafer. In Cerebras’s case, the entire wafer is used to make a multi-chip computer, a supercomputer on a slab. Companies have tried for decades to build a single chip the size of a silicon wafer, but Cerebras’s appears to be the first one to ever make it out of the lab into a commercially viable product.