AI Reality Check: We Humans Shouldn’t Step Aside Yet
Is artificial intelligence (AI) going to take away our jobs? Will AI be able to think like humans? Are we on the verge of a robot apocalypse? Whoa, we’re getting way ahead of ourselves, said AI experts at last week’s Pure//Accelerate 2018 in San Francisco.
Let’s first befriend AI and get it to work for us.
“What goes hand-in-hand with [robotics and AI] is some hype and some fear mongering” stirred up by the media, sci-fi movies and pop culture, said Dr. Kate Darling, a research specialist at MIT Media Lab and keynote speaker at Pure//Accelerate. “It distracts from a lot of the potential that some of this technology has to actually benefit us.”
There’s no question AI’s potential to benefit businesses is massive. Four out of five enterprises will deploy AI by 2020, Gartner says. IDC estimates storage for AI workloads will be a $10 billion market by 2022. A new survey from MIT Tech Review found that 82 percent of business and IT leaders believe AI will have a positive impact in areas of analytics, greater efficiency and reduction of human error.
Channel partners, too, can grab a piece of the AI action. Storage vendor Pure Storage, which sells exclusively through the channel, unveiled AIRI and AIRI Mini, a hardware-and-software solution tuned specifically to handle AI workloads.
At Pure Accelerate, experts talked glowingly about AI’s benefits today.
Ziff, a provider of advanced AI solutions, helped a scrapbook company develop deep-learning models that let customers create scrapbooks. A deep-learning model looks at a photo library, determines which pictures are, say, family photos (such as photos with babies) and which ones aren’t (such as photos with receipts), then groups similar photos for a customer to select the best ones.
“The scrapbook company has more deep learning in production than most Fortune 100 companies do,” said Ben Taylor, Ziff co-founder and chief data officer. “The fun thing with AI, when you’re dealing with big amounts of data, is that things that seem very subjective actually become less subjective.”
The watch phrase, of course, is “big amounts of data.” AI needs to train on copious amounts of data, such as images, to become better at its job.
Processing all this data has been an AI bottleneck, but this is rapidly changing. A few years ago, a million photos seemed large, said Taylor. Today, a lot of Ziff customers start at 10 million images. Some customers have a billion images, which, even last year, seemed impossible to handle.
“Pretty soon, the petabyte won’t seem too big,” Taylor said.
As AI gets better at training on data and performing tasks, the fear is that it’ll replace human workers – but Darling disagrees. She contends that AI will augment humans and likens AI to a working animal.
People have spent thousands of years domesticating animals to perform simple tasks that take advantage of their natural abilities, she says. Today, ranchers use a border collie to herd sheep in the field. It’s not unlike lifeguards using an AI drone to deliver flotation devices to drowning swimmers.
“Animals are these autonomous, sometimes unpredictable entities that we’ve partnered with because their skill set is different and supplemental to ours,” Darling said.
Darling is quick to point out that one of the biggest differences between AI and humans is that humans can perform a myriad of tasks, as opposed to a single one. Humans can also switch contexts on the fly, which AI has trouble doing. For instance, if you asked an early version of Apple’s natural language processing system, Siri, to call you an ambulance, Siri would literally call you an “ambulance.”
“AI is a lot dumber than people,” Darling said. “We’re not anywhere close to developing the type of artificial general intelligence that would be required to context-switch effortlessly the way a person can. We’re not going to be there anytime soon. We don’t even understand our own intelligence.”
(Artificial general intelligence, or AGI, is defined as the intelligence of a machine that could successfully perform any intellectual task that a human can. Many governments, companies and organizations are pursuing AGI development.)
A more tempered view comes from Dr. Andreas Liebl, managing director at Unternehmertum, a nonprofit organization in Munich, Germany, helping small and midsize companies get started on AI. Unternehmertum’s goal is to make sure German businesses don’t get left behind in the AI age.
Liebl sees a near future of AI augmenting humans and helping them do their jobs. In the distant future, though, he understands AI might force knowledge workers to reinvent themselves. A doctor, for instance, might not be prized so much on her ability to diagnose cancer, rather her tactfulness at relaying bad news to the patient and managing follow-up care.
As for AI one day being able to do anything a human can, Liebl doesn’t count it out.
“I would agree we’re not even close to AGI,” Liebl said. “Yet what we’ve seen in the last couple of months in terms of development in AI – we no longer can trust our eyes and ears – we’ve already underestimated how fast we’re going.”