Designing Autonomous AI

Part I. When Automation Doesn’t Work

I recently participated in a workgroup organized by the National Science Foundation to investigate the use of AI in manufacturing. About 20 of us in Workshop 2 Roundtable 4 started by pondering why “software hasn’t had more impact on the manufacturing industry.” Some industries like finance have embraced software and data science in a more significant way than what we see in manufacturing. We came to two conclusions. First, the long list of software algorithms that are available to solve problems are not accessible without training in programming and using algorithms. Many times I’ve spoken to software engineers about industry problems, who asked: doesn’t this industry know about this algorithm? No, they don’t. And even if they did, they would not be able to manipulate it the way a software engineer can and even then there’s no guarantee that the algorithm will solve the real-life problem in all of its complexity without getting to know the process.

My second and more important takeaway from the workgroup was that the manufacturing industry runs on the exchange of skills, not algorithms. In manufacturing, operators, supervisors, and engineers practice building, refining and teaching high-value skills for controlling equipment and processes. Just a few weeks ago a manufacturing executive told me that the most important asset of this multi-billion dollar company was the expertise (skills) of its operators and process controllers. This suggests ...