Operations research and its discontents 2

Man versus machine
The most unsatisfactory aspect of operations research is that the activity of man is reduced to maths equations. To be sure, other social sciences also do this: like economics. Therefore I never completely liked economics either.

A big problem with economics has been the homo econimus assumption. Greedy and self-interested optimisers is a fairly good description of man, since we see from the collapse of communism that relying on altruism simply doesn’t work. It’s the best simple explanation of how people behave. But it is terribly inadequate. The most exciting research that has come out of economics / psychologists have to do with behavioural economics. Yes, people behave like that but there’s more to it.

Another big problem is the use of statistics to model human behaviour. The range of human behaviour is never perfectly described by maths equations. The standard probability functions are not a very good fit for how people will behave.

One tactic of using mathematics to attack the problem is, like I mentioned earlier, assume the system is a pinned down butterfly, and then use mathematics equations to describe it. But the system is a living entity, and often the usage of mathematics fails to capture the dynamicity of the entity.

Things have gotten better with computers. Unlike the old days when you scratched out equations on pieces of paper, computers can handle an arbitrarily large amount of complexity. It’s still not easy to code out everything, you still have to make simplifying assumptions that may be detrimental to the realism, but it’s much better. That being said, a lot of OR, especially the mathematics part, was conceived in the days that predated PCs. So a lot of models are similarly inflexible and unrealistic.

The spirit of optimizing operations of a firm is also totally wrong. What often happens is that you have a CEO, and he always wants to optimise the profits of the firm. Therefore: pay your workers only enough so that they don’t want to leave. Screw your customers. Screw your suppliers. Screw your environment. In reality you will soften this stance, especially in the face of the need for better PR. But the main thrust will still be there. What Marx means when he complains about the endless accumulation of capital. Anyway I’ll bitch about this in the proper place, which is when I’m bitching about capitalism.

OR believes that global optimisation is always better than local optimisation, which tends not to take into consideration the states of other parts of the system. First, centrallised control is quite akin to communism. In fact, a lot of the pioneers of operations research came from the Soviet Union. It doesn’t always work because the big boss in the centre does not have perfect control of the system. Then there is the problem that your information is not perfect, and that can screw up your entire planning. The real solution, then is to have a balance between piece-wise optimisation of the subsystems, and considering the needs of the system as a whole. Sometimes it’s better for each smaller part to act myopically and greedy

The other problem is the engineering of human beings. You just can’t mandate how people are going to behave from above, unless you use extreme measures. My boss was wondering out loud how on earth they got McDonald’s to get every employee to make the burger in the same way every single time. I had already read “Behind the Arches” so I knew the answer but I kept quiet because he never pays attention to me anyway.

McDonald’s owns the property that houses the restaurant. It rents it out to the franchisee. The franchisee is entirely at the mercy of McDonald’s. Do things your own way, run operations your own way, and boom, you’re out of business. It’s that simple. This is a benevolent dictatorship. I don’t know why Americans always complain about Singapore being a nanny state and shut one eye towards McDonald’s being a nanny employer. But I do know why the Singapore government and McDonald’s get along so well.

In the absence of such drastic measures, you have to deal with conservative Singaporeans who are as stubborn as mules, and who don’t like to change for the better. In general, though, my personal reasons for not liking the engineering of human beings does not have to do with how difficult it is to get people to do your bidding. My personal gripe with there being only one way or the highway is: I like my freedom. You could say that a really high degree of discipline temperamentally disagrees with me. However: I must add, I am probably mellowing with age.

Which is why I was so happy to discover chaos theory. Mathematics, cold, rigid and hard mathematics, actually shows you that things can get really screwed up and unpredictable – without pre-supposing the existence of randomness! Randomness arises from determinism. Determinism arises from randomness. Chaos arises from order. Order arises from chaos. This is the true mathematical derivation of the yin and yang philosophy, of opposites deriving from and begetting each other, rather than the western dualist idea of two irreconcilable poles.

There’s so much more wonderful things I can say about chaos theory, but for me, it has made mathematics much more lifelike and intuitive. It has softened the hard edges and surfaces, but at the same time it has also removed a lot of what was great about mathematics – its predictability. Its ability – some say its conceit – to be able to make predictions with – well – mathematical certainty. Chaos theory doesn’t make precise predictions. It shows you what can go wrong with your maths theories. It seldom has a lot of concrete results to show. What it amounts to is one big spanner being thrown in your direction. But it does bring a long overdue sense of reality.

Other rubbish
In spite of my profound skepticism about operations research, I feel that it is still a valuable tool, so long as we don’t blinker ourselves about its infallibility. There are so many times when I see some people walk around believing that operations research, by virtue of being more scientific, is closer to the truth than stories that workers tell you on the ground. They believe that what’s on a process map is a higher truth than workers describing the million and one exceptions to the rules that take place in real life. They believe that the view from the passenger’s seat is better than from the driver’s seat, and if the driver would just sit up and listen every now and then the world would be an irrevocably better place. Sometimes I just have to roll my eyes at that.

For me, operations research is a tool. Possibly the best tool we have in our mission to improve operations. But at the end of the day it is still inadequate on its own. In the end, I thought I would have some other skills. I was equally interested in studying what human beings are like. My future bosses may or may not have been fully appreciative but I wanted a more fully nuanced view of what a firm was like. I took all sorts of liberal arts courses. In the end, I read a throwaway comment by Peter Drucker that said that business studies is the ultimate liberal arts subject. It was at that point that I felt that my decision had been vindicated.

Recently there was a long overdue sentiment that we were expending too much of our effort in the wrong direction. There was way too much emphasis on operations research solutions, and the attitude is “OR is superior, we’ll just wait for them to wake up and catch up”.

There has been a slight shift away from that. Instead, what we have is conducting information sessions and roadshows that try to persuade operations staff to embrace more scientific and analytical methods of conducting operations. That was a welcome change, in my opinion.

What I also wish to happen is that we have some more capabilities to obtain intelligent information about what goes on in operations. The current practice is that we identify that there are demons of inefficiency somewhere, and we’ll just prescribe medicine to cure what we assume are the problems. It’s always the usual suspects, but we’re never really sure. What I wish to happen is we improve our diagnostics.

A good analogy for what we do is that we are doctors (not surgeons, because surgeons are the ones who work on the ground.) We look after a patient whose health is sometimes good and other times not so good. We’ve got medicine (typically OR) which is sometimes painful to use but we believe in it so much that we push it all the time. But firstly we need to improve our diagnosis so that we can prescribe more finely focused solutions that solve the real problems, rather than wholesale invasive procedures that have negligible effect outside of a hot zone. Secondly, we need a keener, more systems oriented approach, where we realise that reacting directly at a problem will have side effects that might be worse than the existing problem. Some problems have counter-intuitive solutions. For example, cutting down the number of cars going through a road can actually increase the throughput of that road.

We need to improve the diagnostics. Too often we conflate measuring performance with identifying problems that need to be solved. These are different tasks, they need to be solved with different means. We need to look closer at local conditions, rather than assume that what we have will fit a model that we bought out of a box from somewhere else. We need to understand how and why our problems are unique, and perhaps be mindful that the solutions will also be in some ways unique.

In academia there is this undue focus that a solution doesn’t really have much legitimacy until some publisher somewhere else deems it fit to be published. Then what happens in the end is that we graft a solution on top of a problem where the fit is not perfect – sometimes, more than imperfect, there is no fit at all. (Six sigma, I’m looking straight at you.)

A former boss said that he saw a pyramid, from an IBM article. (I saw that pyramid too). He was commenting that he was now doing analytics, whereas what he used to be doing, ie optimisation, is at the top of the pyramid. Well I’m thinking, he doesn’t understand that pyramid very well. If analytics is at the bottom of the pyramid, that is the first thing you must achieve, and do very well. You must concentrate your energies on the stuff at the bottom of the pyramid before you go to the top. You just can’t go straight to optimisation simply because you happened to have a PhD in those things.

Anyway, what I’ve described is mostly the theoretical problems with operations research. The problems that gave me an uneasy, queasy feeling. It got even worse when I became a real life operations research practitioner. I don’t intend to blog yet about what it’s like being an operations research practitioner because it’s blogging about my job. (Talking about operations research is merely dangerously close to talking about my job.)