You may have heard about the Land and Jarman study on the creativity of children vs those of adults. Disregarding the unnecessarily flashy terms like "genius-level" etc, we know that solutions found by children are often highly diverse and creative while those of older and more "educated" children start to converge. They all start to learn to think alike.
As do OR practitioners.
Remember that first optimization course you took where the prof asked you to model a categorical decision, like what color to choose for a car? You were probably thinking about an integer first, 0 for this color, 1 for that, 2 for that, and so on. Am I right? I bet not many people's first choice was to use a bunch of binary variables for this task. Or when the prof asked to model a permutation and, again, used binaries to model those rather than, well, ask the solver to please create a permutation.
You have been trained to model one way. And that was the best way to model problems that would be solved by dual solvers (meaning, solvers that compute bounds and search systematically).
Implicitly, you were taught that not best solution within the affordable time was important but lowest time to optimality was the holy grail of OR. You were taught that multi-objective optimization was to be avoided: aggregate or constrain one and optimize for the other - both is utter nonsense in practice. And you were taught that stochastic optimization does not scale and anyway was for academics with too much time on their hands. Alas, most forecasts used in practical applications are estimates and come with inherent forecast uncertainty. Which you now ignore, of course, because you were trained to do so.
The result?
Anyone who has a real industrial problem to solve can literally go to *any* optimization consultant and they all come up with basically the same approach.
No creativity.
No differentiation.
Your work is absolutely replaceable by anyone out there who was trained in mainstream MIP modeling.
When I had implemented the first version of InsideOpt Seeker and I began using it, it felt like someone opened the window for first time in two decades of living in the MIP world. Suddenly, I had to unlearn what I had been taught. To stop using very unnatural and cumbersome techniques that had been essential before.
Suddenly, the models felt and were much closer to the real world. Possibilities opened: Throwing massive compute at a problem. Optimizing 5, or even 20 KPIs at the same time. And of course, creating solutions that work against a host of potential realistic futures rather than picking one and overfitting my solution.
Now that InsideOpt Seeker is available to you, you have a choice: Stick to your training and do replaceable work. Or become a trailblazer for the new world of optimization. Become a child again with unbounded creativity to highly diverse ways to solve problems.
What will it be?
