Machine learning is poised for mass adoption

Artificial Intelligence Data, and its limits Machine Learning Technology analysis

Every few years, it’s exciting to witness a nascent technology emerge as a credibly disruptive influence around the world.  Personal computing exploded in the 1980s, the web in the 1990s.  Today, machine learning for the masses is on the brink of a similar explosion.

Remember when it was a stretch to think that Granny could understand the difference between software and hardware?  These days, she’s got her own computer—inconceivable in 1983!—and uses it for lots of tasks. The same will be true of machine learning, which is the technology that “connects the dots” between data and what she’s interested in. And if she can do it, so can you.

Looking backwards, we’ve already been through the AI/ML hype cycle (remember The Fifth Generation of the 1980s?). This second time around, it’s serious.  From this point of view, I disagree with where Gartner places Machine Learning on its 2015 emerging technology curve, as shown in the graphic above.  We’re not just past the “Peak of Inflated Expectation”: to the contrary, we’re well on the way towards the “Plateau of Productivity”

My evidence comes from (a) my experience living through the Inflated Expectations curve; (b) years living through the Trough and (c) in 2015, I’m receiving signals that feel much more like mass adoption than disillusionment from inflated expectations, as the curve above would have you believe.  Buyers are buying, it’s not just about sellers and marketing.

The hype

Back in the 80s and 1990s I was privileged to contribute machine learning expertise to a number of projects, including the Human Genome Project, DOE hazardous waste analysis, forensic hair analysis with the Colorado Bureau of Investigation, and more.

The trough

Starting in the mid-90s, things cooled down a bit, though, so much that I couldn’t say “Machine Learning” or “Artificial Intelligence” to refer to my work.  So for a while, it was all about “data mining”, then “analytics”, then “predictive analytics“.  Even as recently as Gartner’s curve last year, “machine learning” was nowhere to be seen:

2014-gartner-emergingEntering mass adoption

Today, the explosion of data is creating an adjacent desert: a lack of systems that help you use that data to drive business value, whether it’s customer experience, revenue, or cost savings.  This is the gap that machine learning is filling today.

My evidence: I’ve won a lot of new projects this year, with clients that recognize this: data alone is chocolate, not chocolate cake.   It’s just one ingredient: necessary but not sufficient to obtain business value from many data sources.  My clients have also come to realize that technologies like machine learning, artificial intelligence, and decision intelligence are needed to take extract the full value from investments in big data.

This year, I’ve built machine learning systems for clients in financial modeling, visual pattern analysis, marketing, time series prediction, network optimization, and more.  And their expectations of world-class performance are solidly in line with reality.

Driving the machine learning car

Along the way, the discipline of machine learning practitioner is emerging.  This is in contrast to the researchers and academics that have dominated the field until now.

It’s like the comparison between a car driver and an auto mechanic.  Just as you can drive a car without needing to be able to explain the chemistry inside a carburetor, this separation is emerging in machine learning as well.  And expert drivers are just as important as expert mechanics.

I’ve written already about Amazon’s machine learning environment; it’s one of a handful of new players (including Microsoft’s Machine Learning Studio) that’s aimed at an entirely new audience: graduates of community college programs in data science, rather than PhD machine learning developers.  But mass adoption will go much further: data science experts represent just one stepping stone along the way.

In 2015, for the first time, advances in Deep Learning, in combination with new user interfaces (in stealth today), the availability of GPU and cluster compute environments, massive data stores, and open data from many sources, means that, this year,  we’re well past the hype and have fully entered a world of genuine value from this important new technology.