Each time analyst group Gartner unveils a new edition of its Hype Cycle chart, it inspires either schadenfreude or a sinking feeling. Your competitor has banked on a technology that's mired in the Trough of Disillusionment, and you were wise enough to cash out on the Slope of Enlightenment. Or maybe it's the other way 'round.
The most curious detail about the 2016 edition of the Hype Cycle is not where any one technology shows up. It's how multiple incarnations of one underlying technology -- machine intelligence -- are spread out across several points on the infamous trough-and-plateau chart.
Gartner's label for the rise of machine intelligence is "the perceptual smart machine age," and it predicts that such machines will be "the most disruptive class of technologies over the next 10 years."
The benefits of what Gartner calls "radical computational power, near-endless amounts of data, and unprecedented advances in deep neural networks" are on the rise, but none has yet ripened to the point where it is boringly useful. Furthermore, Gartner doesn't see any of the bunch becoming mainstream before at least two years go by, with most of them in the "wait five years or more" category.
At the start of the curve, in the "innovation trigger" section, are normative, world-changing concepts like general-purpose machine intelligence, smart robots, and neuromorphic hardware (such as chips that simulate neurons). This slice of the curve includes related technologies like brain-computer interfaces; it doesn't take much work to imagine how connecting computers to brains can aid the former in behaving more like the latter.
"Machine learning" is at the peak of the curve and is predicted to become a mainstream item within two to five years. That sounds way off -- if anything, machine learning arrived a long time ago and made itself comfortably at home in the form of cloud services, software toolkits, and custom hardware.
The real issue -- and the reason Gartner may be pushing machine learning's heyday ahead -- is not that machine learning faces technical barriers. Rather, the primary obstacle is that machine learning isn't a cure-all, so it'll take time to reveal its genuinely beneficial applications. (Word had it the inventors of the laser couldn't figure out what to do with it at first either.)
Careening headfirst into the Trough of Disillusionment are self-driving cars (labeled as "autonomous vehicles") and "Natural-Language Question Answering." Both are examples of machines expected to act like humans but without mistakes. There's little point of having a self-driving car if it hops the curb and forgets to brake for pedestrians, and there's little reason to have a machine respond to plain-English questions unless it can produce precise and coherent answers.
In both cases, it's not hard to see why Gartner thinks the bloom may be coming off their roses. The reality of both projects is that they're messier and more complex than anyone could anticipate. It's one matter to make a car that can maintain speed and distance from other cars on the highway, and another to make a car that can parallel-park without getting a ticket -- for now. Ditto natural language, especially since natural language is ambiguous by nature.
Sensible discussions about machine intelligence hold it as an augmenter, not a replacement, for human insight and understanding. In that sense, machine intelligence is another variant of what computers were invented for in the first place: To automate the boring stuff (as the title of a popular book on Python puts it) or to transcend the limitations of human observational capacity.
If any variety of machine intelligence makes its way to Gartner's Plateau of Productivity, it'll be the most dialed-down and immediately practical variety -- for example, using machine intelligence to find patterns inside petabytes of data, then making educated decisions based on the findings. Making sensible, unhyped use of that technology is a big enough ambition for the time being.