My instincts tell me we need a sense of urgency around the use of artificial intelligence (AI) in manufacturing.
The urgency is driven by how quickly technology can move today, and how an unexpected breakthrough can quickly dominate. AI is used in facial recognition, converting speech to written word, and in winning chess matches. Surely, there must be a horde of potential applications in manufacturing.
While I’ve written before that I think the reality of AI’s “intelligence” is complex mathematics, I got a more enlightened vision when I posed that view to a true expert. In his opinion, we must think of AI in broader terms. “It is fair to say it is mathematics, but it is easy to get too hung up on the math because it is really more about the data,” explained Oliver Christy, founder of Foxy Machine, an AI consulting and strategy firm based in New York City. “AI lets us ask new questions using data in any situation.”
But even limiting oneself to data and mathematics is, well, limiting. A third consideration must be the business problem—what is the situation you are considering and what tools are available, according to Christy. “You need to look at any given problem from all three approaches,” he explained. “Mathematics, data, and the problem itself. That holistic viewpoint [can] give you a robust solution.”
He also thinks manufacturing is ripe for AI applications. “Some of the easiest problems for AI is in solving safety and manufacturing risks,” he said. “We have AI systems now that can understand on a huge scale what risks there might be and how to improve safety.” Using image recognition techniques in quality control is another easy win. One application Christy worked on recently trained an AI system to recognize the quality of steel after heat treating, flagging samples that might be of concern. The samples were then examined in more detail by a human.
He stressed a key point about how AI could best be used—augmenting rather than replacing the quality control worker. Using digital images, an AI system can look at and flag many more samples than a human can without fatigue. But in the steel quality case, it could not provide that final human touch needed to ensure it was good or not. He related how a similar system used to find skin cancers performed better than a trained oncologist, but the AI system combined with the oncologist performed even better than either alone. “Exactly the same approach should be used in manufacturing, where machine systems and humans could work hand in hand,” he said.
Back to my sense of urgency. Much of today’s AI technology is not new. The concept of using computers to mimic human abilities instead of as just calculating machines has existed as long as computers themselves. But an inflection point in computing power and data is making it urgently relevant. “When I started 20 years ago, the cost of computing power and data limited [possible] applications,” said Christy. “Now we have huge amounts of mixed data and very cheap computing. I have access to million-dollar computers, direct access to technology like IBM’s Watson that costs almost nothing.” To this add open source AI software like Google’s TensorFlow that is effectively free to use, and we have a “perfect storm of all the components needed to build and use AI,” said Christy.
He believes that AI can easily become the next competitive advantage. However, it can be daunting. His advice? Start small. Build a small team, start collecting data around a pilot business problem, and learn how the best approaches can solve a manufacturing problem. “But start today,” he said.