AI vs OR

Inspired by two unrelated LinkedIn posts. One was bemoaning that they didn't see much success in using AI for route planning. Another bemoaning that operations research (OR) doesn't get the same time in the spotlight as AI does.

OR is about using the power of math to come up with definitive answers to questions such as how to best pack parcels onto different trucks and then send them to different destinations. Generally it's about finding the solution to a problem that maximizes or minimizes some objective function subject to some constraints.

I've seen first hand how OR is treated as the ugly step sibling of AI. At a previous job, in a pre-sales call, we were looking at a slam dunk OR problem. But the client kept pushing, because they wanted an AI solution. Maybe to impress investors, maybe to access special purpose government funding. Either way, the project didn't make sense that way and the client lost out on actually solving their problem.

In a way, AI and OR complement each other. OR solutions are mathematically precise and require no training data. They're great for "narrow road" problems where and slight departure from the right path leads to an invalid answer instead of just a "good but not great" answer.

Where OR lacks though is ease of use. Taking a messy real world problem and shoving it into the mathematical formulations amenable to common solvers is hard, highly specialized work. You can't just hand a trucking company a license to Gurobi (a popular tool for optimizing so called mixed integer problems) and be on your merry way.

So that's where AI can come up the rescue. I have great hopes for what a generative AI interface into the arcane OR tools could do.

Because many people in industry I talk to have problems much more suited for OR, yet they are looking for solutions among the AI tools, because that's what everyone, from LinkedIn influencers to the big consulting firms, is pushing.