CAMBRIDGE, Mass. -- Machine learning is overtaking big data in Google searches, but the hype around artificial intelligence systems may not be hyped enough.
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
Erik Brynjolfsson and Andrew McAfee, co-authors of the forthcoming Machine, Platform, Crowd: Harnessing Our Digital Future, said some machine learning algorithms are improving faster than anticipated, thanks to enormous data sets and access to more compute power. And the rapid rate of change is opening up new avenues for machines and new opportunities for CIOs.
Brynjolfsson, director of the MIT Initiative on the Digital Economy (IDE), and McAfee, principal scientist and co-director of the MIT IDE, participated in a fireside chat at the recent MIT Sloan CIO Symposium, where they said they underestimated the disruptive capabilities of automation and discussed what they called the second wave of the second machine age. Instead of humans codifying knowledge for machines, the machines are capable of learning on their own.
"We think it's probably the most important thing that's affecting the economy and society over the coming decade," Brynjolfsson said.
A runaway example, McAfee said, is AlphaGo, a system built by Google DeepMind that bested the world champion of the Asian abstract strategy game Go. "The Chinese Go champion tweeted out on the Chinese equivalent of Twitter a little while back, 'I don't think that a single human has touched the edge of the game of Go,'" he said. In less than five years, machines may have exceeded 2,500 years of study and accumulated knowledge of the game by humans.
And improvement in machine learning systems isn't limited to game strategy. Speech and image recognition technologies, for example, are operating with human-level precision and learning at an unprecedented rate. Google revealed at the Google I/O 2017 developer conference that its speech recognition technology improved from an 8% error rate to about a 4% error rate -- not in 10 years, but in 10 months, Brynjolfsson said.
Those advances are positioning machine learning systems as the powerhouse of the software world, sitting below the user interface of every chat bot, autonomous vehicle and facial recognition system. And the pace at which machine learning systems are evolving presents a unique situation for CIOs, who shouldn't sit passively by and let the technology lead, according to the researchers.
Indeed, Brynjolfsson said companies like Microsoft and Google and renowned machine learning experts like Andrew Ng describe what's happening as "almost a land rush."
"There are so many opportunities that people haven't cashed in on yet," Brynjolfsson said. "And the bottleneck now is actually identifying the problems and opportunities that these technologies can be applied to most effectively."
Experimentation -- such as Google turning a gaming algorithm loose on its data center to reduce power consumption -- is key. "The main error that a lot of companies are going to make is to extrapolate from the past and keep doing what they were doing with a little bit better accuracy or a little bit better precision," McAfee said.
Machine learning systems and the workforce
But grappling with machine learning systems is not just about learning how to exploit the technology. It will require rethinking whole swaths of employment, from menial to highly skilled jobs.
In The Second Machine Age, published in 2014, the researchers wrote about the "the great decoupling" of wages and productivity. "For 200 years, [median wages] rose right alongside productivity, but now they are stagnating," Brynjolfsson said. With dire effects. Brynjolfsson pointed to Anne Case and Angus Deaton's research on the rise of deaths from suicide, drug abuse, alcoholism and depression -- or, as he put it, "deaths from despair."
And with advances in machine learning systems, it is not just rote tasks that are prime targets for automation. Jobs that require years of training are also at risk, such as pattern recognition or image analysis done by pathologists and radiologists. "Machines can now recognize potentially cancerous images as well or better than humans can," Brynjolfsson said.
Just how much of an impact machine learning systems will have on jobs and careers is hard to predict. When Jason Pontin, editor in chief and publisher of the MIT Technology Review and the moderator of the fireside chat, asked the academics to put a percentage on the number of jobs that could be eliminated in the next 10 to 15 years, Brynjolfsson responded by saying, "That's the multitrillion-dollar question."
One thing is certain, the authors added: Machines aren't good at everything. Tasks that require creative thinking, interpersonal connections, large-scale problem solving or complex planning are still better performed by humans -- at least for now. And advances in machine learning systems do offer a career opportunity, especially for people with computer smarts.
In the IT space, forward-looking CIOs will consider how to arm their employees with the skills necessary to remain relevant in the machine learning economy, such as learning TensorFlow, the open source software library for machine learning developed by Google, Brynjolfsson said.
Opening up the 'black box' of machine learning
AI and machine learning in the ER
CIOs, get ready for the cognitive enterprise