Machine learning is being touted more and more as a cornerstone for the future of business and industry. But will it be enough by itself to help us all move forward?
If 2016 was the year artificial intelligence entered the general conversation, 2017 might be the year machine learning follows suit. SAP co-founder Hasso Plattner discussed it in a keynote earlier this year, plenty of other big names have announced ventures into the space, and a pair of industry analysts have published a new book on the subject.
The Mathematical Corporation: Where Machine Intelligence and Human Ingenuity Achieve the Impossible is the product of more than a year of research and reporting from Josh Sullivan and Angela Zutavern, SVP and VP, respectively, at Booz Allen Hamilton. The duo founded a data science practice back in 2011 — they originally focused on cloud analytics — and have evolved since then to jump into perhaps the hottest tech topic today.
Before we go any further, a short note: Machine learning and artificial intelligence are, for the record, very different areas. Artificial intelligence is the subject of so much science-fiction over the last six or seven decades, and is the idea that machines can perform tasks, either physical or, increasingly, audibly. Machine learning, meanwhile, is more of an AI application where we humans give machines so much data and basically tell them to sift through it and … learn.
About a year ago, Sullivan and Zutavern were deep into the Internet of Things. They stepped up, proverbially speaking, for a broader view “and saw that there were two things happening in the world,” Sullivan said. “One was that everybody was talking about the technology and nobody was talking about its application. The other was that nobody was talking about leadership. How will we lead differently when we have human workers and machine workers in the future, and machines are doing cognitive-level work? What does that mean for a leader? How do you think about optimizing your organization?
“When we got into it and started thinking about it, people weren’t even talking about privacy or ethics, anything like that, they were just doing experiments. So we spent a year talking with leaders who are really trying out different things, trying to get a sense of what they were doing, then distilling that into some top takeaways.”
The result is one of the more important tech books published this year. The Mathematical Corporation offers a look at dozens of companies, including some big-name manufacturers like Merck. Where have they struggled? How did they right themselves? And how is their leadership entwined now with machine learning?
“The promise of using machine intelligence is that we will overcome our narrow-minded thinking by using rules derived from a fuller understanding of how humans act,” Sullivan and Zutavern write early on. “In turn, we will be able to predict and shape behavior, whether for profit, competitive gain, protection, or social good.”
In a survey conducted just last year, fewer than two of every five executives said their companies were “highly data-driven” — and manufacturing execs are often even further behind the tech curve — with most relying on machine intelligence and learning for only basic descriptive and diagnostic purposes, not predictive purposes. That was one of the big reasons Zutavern and Sullivan wrote the book they did.
What follows is part of our conversation about machine learning, how it might affect manufacturing, and just how much it could become a part of everyday business. Seven weeks still remain until Labor Day and the end of the summer. If you’re looking for another read — on the beach or for the boardroom — this is a good one.
IW: First question, really broad: How do you see machine learning affecting manufacturers?
JS: In manufacturing, experimentation through knowledge creation is going to be huge. And I think we have to be honest, there will be some job displacement. That is going to happen. I’m more and more convinced that jobs that are very narrow are the most susceptible, and the one that cross different domains or disciplines won’t be threatened by machine learning for a long time.
One of the other things we learned from the book is that some of the leaders, especially in manufacturing and especially people working with physical machinery, were embracing complexity as something that is a boon, not a burden. There’s a real big mindset change. The U.S. Army has 212 different systems using GPS, and so every piece of machinery they build and use has this GPS receiver technology on them. One aircraft alone has 14 different GPS receivers on it, so very complex. Rather than simplify it, or make one kind of GPS system, the leader, Kevin, just said, ‘I’m going to build machine learning algorithms that take advantage of that complexity so I can understand real signals and spoof signals. The world is full of complexity, and I’m just going to build my machine learning algorithms to work in the real world.’ I thought that was a real interesting leadership take on how to think about complexity as an asset.
IW: What does the future hold for manufacturing?
JS: Angela and I disagree on this topic.
IW: Perfect! It’s point/counterpoint.
AZ: We debate this all the time.
JS: I absolutely think there will be job losses. I think we’re heading into the next few years of automation, of certain types of knowledge workers, and that’s going to happen, and that will give way to a broader set of job displacements. It won’t happen overnight. We are adaptable to change and will certainly figure out training programs. But I do think we have to accept there will be some job displacement. Now, that being said, if you think of the workforce at large in the world, most of the rote jobs, the jobs that are ripe for machine learning to automate or to assist in some way, are in India and China. So I think we’ll see it happen in other countries first; we won’t see it happen on a broad scale in the U.S. But we are headed for some amount of automation and job reduction.
AZ: I believe we’re headed for a net gain in jobs. The jobs will shift, certainly, and some jobs will no longer need to be performed by people — machines will be much better at those jobs — but people will shift into new roles and there will be new products, new services, new business and new functions launched as a result of the innovation that comes from machine intelligence. People will have to shift their skill sets and some of their job functions.
JS: You often argue with me there will be net new business created, too.
JS: The new insurance company that only insures semi-autonomous vehicles — that only insures the software sensors and the hardware, not the driver.
AZ: And in manufacturing, there will be new businesses that start out because of new approaches to industrial control systems and other new technologies. There will be demand for new products we’re not even envisioning today, and those will all need to be manufactured.
IW: This seems like a pretty major difference of opinion for two people who worked together on a book for the better part of the year. I imagine it made the book stronger, you two going back and forth. Did either of you manage to swing the other, or are you both pretty resolute?
JS: I think we’re both pretty resolute, but we’ve been able to sway some of our clients and those around us. Angela, I think, has a much more optimistic view of the future than I do, just by default. She often backs me into logic corners where I really have to defend my position. It’s strengthened us. This tension, though, is with lots of people we talked with for the book. There are boards and CEOs that told us, ‘I don’t know how to talk to my workforce about the future. It’s so nebulous, and I don’t even understand it.’ They’re really grappling with it.
AZ: We like that we challenge each other, and it points to one of the lessons learned in the book: We talk about these asymmetric teams. You don’t want people on a team who all have the same way of thinking and the same perspective and the same background. So much more creativity comes out of differences of opinion and disagreement. As long as people are open-minded about it, that debate really takes everything to a higher level.
IW: One of the manufacturers you talked with and wrote about that seemed to have used machine learning to go to a higher level was Merck.
JS: What Merck did is, they had a really complex manufacturing process they had perfected over years. It was four stages, it took 40-plus hours. They had optimized each step of their manufacturing process really well, and they had optimized locally. But they had never looked over the entire four-step process, because it was thousands of variables, this really high-dimensional data set, really complicated to look at, and every run would look a little different. When you optimize something locally and so specific as the manufacturing process, you’re almost giving up the ability to optimize globally. … For Merck, it was something like 15 billion different calculations. They said, ‘You know, it’s complicated, so let’s send in the machine learning algorithms to go figure it out.’ And they had a pretty successful time of it.
AZ: They learned so much just from combining that one manufacturing process, it inspired them to spread it throughout entire plants. When they opened a new plant in Singapore, they implemented it across the entire plant, every manufacturing process.
IW: How high can you see machine learning and intelligence going? Obviously, it can replace and reallocate a lot of the workforce — and you stress a few times it’s best to combine human and machine — but is there a point where we might have machines basically managing whole groups of people and machines? Maybe even a robot CEO?
AZ: We don’t believe in robot CEOs, but we do believe machine intelligence will be an additional seat at the table in the boardroom. That’s how high we think it goes. The collaboration between leaders and machines is going to be key to new breakthroughs. But it will need to be at the highest level. It can’t just be some specialty technical area. It will affect how companies interact with customers, how they manufacture, how they perform their internal and back-office operations. It goes all the way to the top.
JS: We still believe people should be the ones who drive the breakthroughs. The detailed calculations and the work can be left to the machines and the AI algorithms, but it has to be a human who has the wherewithal to say, ‘Let’s go figure out how to do something differently in this space, let’s go see what’s possible.’ Hopefully, CEOs and other leaders are still the drivers of breakthroughs.
IW: As you point out, you still need to imagine, still need to create. Robots don’t really ask very good critical questions. What’s the most interesting question you’ve been asked about machine intelligence that you really didn’t consider while researching, reporting and writing?
JS: We were in Silicon Valley and one person was very fearful of the general artificial intelligence, that we were racing toward AI inside machines we couldn’t control. They were almost fixated on that. I can’t predict the future, certainly, and we may see shades of that, but right now, we can barely get these machine learning algorithms to drive on a straight highway. We’re a ways away from that, but there’s a lot of hype.
Nothing we’re seeing is actually new. We have more data so we can train convolutional neural networks better, we train these algorithms better. We haven’t seen anything new in this space, we’re just figuring out how to apply some of these technologies that have been around for 10 or more years. I get a lot of questions about the Terminator and extreme sci-fi, about how far we are. Some people are obsessed by it and are really fearful. I just don’t think we’re anywhere near that.