Is Artificial Intelligence (A.I.) Ready to Run a Factory?

Abstract

With smart factory investment expected to increase 20% year-on-year over the next five years and total investment expected to reach $275 billion worldwide by 2027, the use of Artificial Intelligence (A.I.) to manage operations is receiving considerable attention.  This paper takes an in depth look at how factory data is being generated, stored, processed, transferred, trained and ultimately validated using A.I.  The conclusion is that deep machine learning is more than capable of controlling devices.  Yet, research shows only 14% of smart manufactures would describe their A.I. efforts as successful.  The problems are cost and application.  Smart manufacturing is almost exclusively done by multi-billion dollar operations.  Is this money well spent?  Factories aren’t closed, linear systems. In these chaotic systems infinitesimal changes in any one of the myriad of input variables are capable of producing disproportionate changes in output values. As a result, no matter how much scrap, downtime, sales or on-time delivery data a company collects actual values will diverge exponentially from what existing A.I. algorithms are predicting.  Until more research is done predicting dynamic, nonlinear systems A.I. will not be capable of running a factory without human involvement.