But if you read the actual details, the picture is subtler and even optimistic. A line from the executive summary of McKinsey’s “Harnessing automation for a future that works” states: "While less than five percent of all occupations can be automated entirely using demonstrated technologies, about 60 percent of all occupations have at least 30 percent of constituent activities that could be automated." To me that does not sound like the employment Armageddon portrayed in some media. In fact, if 30 percent of my job were automated, just think how much more productive I could be.
Furthermore, McKinsey says “the world’s economy will actually need every erg of human labor working, in addition to the robots, to overcome demographic aging trends in both developed and developing economies. In other words, a surplus of human labor is much less likely to occur than a deficit of human labor.”
So how do you distill the truth given varying opinion and decide the impact on your business?
Machine learning, a subset of AI, will almost certainly transform the world. Advances in the last 12 months, demonstrate the potential. In early 2017, Libratus beat the best pro poker players in the world. DeepMind created AlphaGo Zero, which taught itself to become the best Go player on the planet. Then last November, AlphaGo Zero became a superhuman Chess and Shogi player in less than 24 hours. Not only are the algorithms getting more powerful, they’re also becoming more efficient.
It is clearly an exciting period for technology, but for analytics, does this mean firing your team and going all in with AI? I don’t think so, and in fact, advise against it. Tim Harford argues that AlphaGo is actually an outlier, and corporations are getting less and less involved in groundbreaking scientific research. Not only that, finding the talent to successfully run machine learning projects is very hard; the big tech firms swallow up the brightest talent before they’ve graduated, as reported by The Economist (“Battle of the Brains”).
Attend any business technology conference, and the agenda will be full of AI and machine learning sessions. At those events, the message is much more mundane than the bleeding-edge technology described above. The current recommendation is to begin with small, well-defined questions that can be tested against large datasets. This is not because the experts aren’t highly talented, it’s because the real-world business applications aren’t yet as capable as the media hype.
If you can’t make it to conferences such as Hubb in Germany or Strata in the United States, I recommend reading two fascinating books on artificial intelligence, which paint a useful picture of the state of play. Deep Thinking by Garry Kasparov is about his history of playing against computers in chess, with a deep focus on his matches against Deep Blue in 1997. Kasparov interlaces his optimistic opinions about AI and its implications on society and business throughout the book.
For an excellent counterpoint to Kasparov, read Cathy O’Neil’s “Weapons of Math Destruction." There is growing concern that the rampant spread of algorithms into our daily life is happening before we’ve asked ourselves if they actually help society (see, for example, teacher assessments and YouTube’s Kids Channel). Her book is sobering and a wake-up call for individuals and politicians to examine if can we change regulation fast enough to keep up with technology.
Whatever the hype, businesses need to be thinking about artificial intelligence, machine learning, and automation, and it must start with data. Data is the foundation of any AI system. But you need to have sound and mature data management practices in place first. You also need excellent human intelligence to complete this: your employees need to be comfortable analyzing and making decisions with data. Without that in place, moving to AI-focused models is too wide a gulf to cross. People will not trust ‘black-box’ AI if they don’t already trust their data.
Not only should the data analytics platform be robust, the team’s responsible for it must have a good mix of skills. The traditional skill profiles have changed. Now you need “Fuzzies” and “Techies” to succeed vs. traditional analysts.
As you build out this maturity within your organization, identify teams who are ahead of the curve. Ask yourself, which teams have high-quality, large datasets, and tightly defined questions? Of those teams, who are the most analytically mature? That’s where the first forays into AI-based decision making can begin and your journey to embracing instead of resisting AI will grow to support enhanced job production.