Can you drive a car? Speak a language? Look for clues in reams of data? People do those things all the time; now, a growing number of tech companies are trying to build software that can do the same things with a similar degree of effortlessness, thanks to artificial intelligence (A.I.) and machine learning.
The rise of A.I. will necessitate an accompanying increase in specialized programmers who can craft the algorithms that enable machine learning. At the moment, such skills are in short supply, but universities and online course-providers are trying to fill the gap.
Now that companies comprehend the potential of A.I., they are willing to invest billions of dollars to transform abstract concepts into hard products. Analyst firm McKinsey and Co. estimates that these companies invested between $20 and $30 billion in A.I. last year; the bulk of those bets are being made by digital giants like Google and Baidu, while venture-capital firms also contributed a few billion.
Meanwhile, global revenues from A.I. will hit $12.5 billion in 2017, according to analyst firm IDC. That’s a 59 percent increase over last year; the compound annual growth rate (CAGR) through 2020 is expected to reach 54.4 percent.
Yet despite all the money and market growth (or perhaps because of it), the number of data scientists needed to advance A.I. will fall short of demand. There’s also a severe shortage of managers and analysts. IBM has gauged that job demand for “data-savvy professionals” will reach about 2.7 million in the next few years; average salaries should be around $80,000, and require between three and five years of experience.
“The number of data scientists is growing by 7 percent every year through education, but the demand is growing 12 percent year-over-year.” said Mat Leonard, the Deep Learning Foundations lead at Udacity. “We have a lot of work to do, clearly.”
The potential of A.I. has only grown stronger. “Thanks to recent improvements in deep learning, computers can now do far more than a few years ago. Everything ranging from accurate face recognition, to practical speech recognition, to a growing ability to process and spot patterns in terabytes of data,” said Andrew Ng, Coursera’s co-founder. “That’s why we say A.I. is the new electricity—similar to the rise of electricity 100 years ago, A.I. now is poised to transform every major industry.”
Once again, the tech industry faces the steep adoption curve of a new technology. Universities respond by issuing new programs and degrees, while online course providers produce nano-degrees and certificates.
Need to Know
The nano-degree or certificate approach is aimed at imparting very specific skills. “The vast majority of our students are employed during and after participating in Nanodegree programs—and many of our students are sponsored by their current employer,” Udacity’s Leonard said. “Overall, we have 53,000+ Nanodegree students enrolled (4x growth over last year) and we’ve had 18,000+ Nanodegree grads to-date (6x growth over last year).”
For Udacity’s Deep Learning Foundations Nanodegree program, students need experience in Python, calculus, and linear algebra in order to grasp the underpinnings of modern frameworks. “For career-ready Nanodegree programs like the Machine Learning Engineer Nanodegree program, you’ll need intermediate Python, statistics and knowledge of calculus and linear algebra,” Leonard added.
But the basic coursework for other programs may not necessarily require previous experience. As Coursera’s Ng points out, first-time students in a machine-learning MOOC (ml-class.org) can get by with basic programming skills.
“If you want to learn state-of-the-art techniques, the Deeplearning.ai specialization is taught in Python, and also assumes you already have basic programming skills. Prior knowledge of basic linear algebra will be very helpful,” Ng said. “Previous experience with machine learning, for example, through the Machine Learning MOOC, would also be helpful but is not required.”
Nor are universities idle when it comes to A.I. Case in point is the University of Washington, which supplements its academic offerings in computer sciences with “career accelerator certificates” for specific skills, including machine learning. “Our data programs are in high-demand, particularly via our Career Accelerator program, where we’ve increased overall capacity by 93 percent and online, specifically, by 152 percent,” said Andrew Hoover, Senior Director of Program Strategy, UW Professional & Continuing Education.
The machine-learning course “teaches over twenty different types of supervised and unsupervised machine learning models using Python and R programming languages,” Hoover added. “We teach feature engineering, model selection and model fitting to make the most accurate prediction, optimal classification or best clustering of data possible.”
“Students will need to learn the basics of optimization methods to configure the settings of available tools and to produce the best model for the assigned task.” Hoover continued. “We’re training people to train machines.”
And if you train to become one of those people training the machines, the potential opportunities are endless.