понедельник, 4 февраля 2013 г.
Our Predictions Are Bound To Fail
Some years ago, when I did pay much attention to politics, I have found multiple articles by Valeriy Panyshkin (here is some info about the journalist, alas it's in Russian only), who wrote a lot about politics as well as numerous other problems, which one could see in Russia. Despite the fact, that many of his articles, opening my eyes to the things, which I didn't necessarily want to see, were somehow depressing, I read almost all of them. Moreover, even after I had become less concerned about these things - closing my mind once more - I continued to follow the man's writing. Interestingly, Valeriy gradually switched his attention from politics to a narrower set of issues - he got focused mostly on various problems that children and those who try to help them face in our country. He still highlights many other things too, although, as I see it, now, working with organizations that raise money for healing badly ill children or try to protect orphans' rights, he devotes most of his time, efforts and writing to these particular issues.
Recently I have read his article (the link - in Russian too) discussing a set of principles, upon which, according to Valeriy, every charity should base its work. The most surprising thing about the article was the idea that any organization of this kind must aim at ceasing its existence. Although this may seem a bad mission, the idea is quite simple and makes sense when elaborated. Valeriy doesn't claim that every charity should give up helping people and go find a more reasonable niche. What he states is that everyone who works in such a foundation should dream of the day when the problem, that he fights with, is finally solved: a treatment is found, which heals blood cancer in a matter of minutes, orphans receive such a crazy amount of support from government that life is as easy for them as it can be and so on. This way the main purpose of any charity is not only to help people, but to put its best efforts into improving the situation so that one day there will be no need for their help at all. This notion of aiming at becoming obsolete, although looked interesting, seemed to me something local to the field of charities at the first glance. However, it is easy to see that it is inherent (or, at least, should be) to many other kinds of activities and institutions - from medical research centers and police departments to ministries of Justice and parliaments.
Because I am a programmer it was particularly interesting for me to notice a trace of this idea in the world of software. First of all, there is a multitude of software development jobs, which pursue a goal of solving a particular problem, and once the solution is brought to life, its consumers, roughly speaking, don't need the awesome guys, who built it. However, this kind of relation between producers and consumers, although widely present in my field, is quite common elsewhere. Maybe its because it looks so usual, that it does not seem that interesting. It stays hardly exciting even if we consider all the benefits that many of us could get in case we did our best to make working with our products and extending them absolutely independent of our own presence. Still, I have finally stumbled upon another idea that looks somehow reminiscent of the principled proposed by Valeriy.
There is a system at my university that eases collection and aggregates information about the progress of students in the course of a semester as well as their performance during final exams. As I have undoubtedly mentioned in this blog, my diploma project is going to involve analyzing this data and making predictions based on it. What I want to do is to build a system that is capable of predicting the results of exams for a particular student from their performance during some part of a semester. Such a tool may be useful for faculty staff - for instance, they might want to know who's chances to be expelled are high way before exams. Wait a minute... The people, who teach and do different kinds of administrative work at the university pursue the goal of providing every student with quality education - they barely want anyone of their students to be expelled. So, for what sake would they want my software to say that certain students perform poor enough to fail examination and thus put an end to their education? The answer to this question is not subtle at all, because once you know that someone in your responsibility has serious problems - with study in our case - you gain a chance to help them. Thus, my system is not designed to allow teachers and other faculty staff to concentrate only on those students, who work well, but to draw their attention to those who may fail and to do this before it is too late to change anything. This way I am working on a system that will hopefully predict events that are not supposed to happen thanks to these same predictions. Furthermore, I am quite satisfied with this situation and will work hard to make its failed predictions as precise as possible. Clearly the notion of forecasts, that we make with the purpose of making them fail, is not equivalent to the idea of charities aiming at self-elimination, still I feel there is something similar to the nature of these two things.
Not surprisingly, I am not the only one who strives to make precise forecasts to see them fail in the end. The most popular example of applying machine learning techniques, demonstrated in various courses devoted to the subject, is predicting whether a patient is going to have cancer (or has it already.) When such a thing is done by medics outside the lecture-room, their main goal is to save a patient from the disease. The same holds for exploiting data on cardiac arrests or CVAs - healthcare field is rich with examples of this kind, as many other areas are. For instance, analysis of constructions is usually performed to find flaws in them and discover where the thing should be strengthened so that no disaster will happen - not to know where to strike. Aircrafts and vessels are overseen to avoid collisions, which are predicted first. The list is quite long and there is no need to continue it, because my point should be clear already.
Actually activities of this kind took place much earlier than the first computer was built. People always tried to foresee things, which they wish didn't happen, and they did this not only because of the desire to know the future, but because they knew that as soon as one is aware of bad things ahead, one can escape or prevent these. So I am not opening something new - I just feel fascinated by the simplest idea that was here since the first days of humanity or even earlier. The invention of computers merely gave us new powerful tools and techniques to foresee things, and it seems astonishing to me, how frequently we try hard to make precise predictions that are bound to fail just because we do succeed.