Skip to content

Computers aren’t science

As I read the science feeds for various sites, I am struck how often people are reporting on computers and computer techniques. News flash: Computers aren’t science, any more than glass blowing is chemistry or addition is physics.

Computing is a mathematical technique that uses electronic shortcuts. Computation is an aspect of what scientists do (so things like statistics and its implementations like bioinformatics or cladistics are aspects of science) but these are means to an end – the discovery and explanation of phenomena. The glassware aspect of computation is a necessary precondition for doing science, yes, just like experimental design is, but the running of the experiment is the science proper.

There’s a reason why mathematics is a distinct discipline from science in general: it has its own rules and research programs. So does literary criticism. And not only science, but also actuarial work, engineering, economics and astrology use mathematics. Yes, I said astrology – quite a lot of math was developed to help that along early on (the reason why Augustine attacks “mathematicians” somewhere is because he’s attacking astrologers, who at the time were the primary users of mathematics).

A computer is basically a very fast pencil and paper to a mathematician, and research in using these tools in science is a useful, but hardly core, aspect of scientific behaviour. To treat computing advances as science in themselves is as bad as those who treated the social aspects of science as the core of scientific behaviour (and consequently treated science as if it were a village in the Papuan highlands, with as little concern for the actual content as an anthropologist might have for the magical beliefs of that village).

There. I said it. I’ll go have a nice lie down now.

20 Comments

  1. I once butted heads with a local crackpot, and when I said that very few scientists were creationists, he replied with an anecdote about a creationist friend of his who was a *computer* scientist.

    Which is a breathtakingly stupid argument, on so many levels.

  2. Allen Hazen Allen Hazen

    Eamon’s comment is so much better than mine that I should just shut up, but…
    “Computer Science” (or, as some universities and the French say, “informatics”) covers a multitude of sins. “Theoretical computer science” I think I would deem a science: it’s a branch of mathematics with close links to logic. But of course most of the computer news that gushes forth isn’t about THAT end of C.S.!

  3. John Wilkins John Wilkins

    Several things are identifer marks of it not being about Computer Science/Informatics: Any mention of Microsoft, or any other proprietary company (even Apple!); any implementation of a mathematical technique (especially statistics), and any claims to replace empirical science. We can’t simulate a single cell yet; why should we believe that we can replace animal models in biology?

  4. Barry Rountree Barry Rountree

    Ah, well, maybe you could help me come up with a better word to describe what I’ve been doing.

    I’m trying to understand one aspect of a complex system (effect of CPU frequency on execution time for scientific kernels). I started out years ago with a very simple model (increased memory pressure decreases responsiveness to changes in CPU frequency), decided what I should measure to test this hypothesis (Level 2 cache misses), ran lots of experiments, plotted a regression…. and found out the model sucked.

    This was followed by several years of modeling, experiment and refinement to the point where I had a set of metrics that would reliably give an R^2 of 1.0 but still had significant error (which is a nice cautionary tale for interpreting regressions). It looked like this was the best anyone was going to do, I needed to defend my dissertation soon and this was a chapter, so we started writing it up for publication.

    And we got scooped.

    What I had, though, was a model that tried to explain /why/ those particular metrics worked so well, and why error was still creeping in. I moved to a cycle-accurate processor simulation where I could measure anything I wanted and am in the process of showing how a single metric (not available yet on real hardware) can be used to calculate the change in performance directly. In short, I’ve moved from a model of a real system that predicts well and explains poorly to a model of a (slightly) simpler, abstract system that predicts /and/ explains far more effectively. (My prediction error has dropped from ~1% to ~0.01%.)

    So, what I’m looking for is a term that can capture the following:

    1. Simplification of complex systems into models.
    2. Experimental validation of these models.
    3. Models judged on their ability to both predict and explain.

    Obviously “science” is right out, because this work is being done under the auspices of a computer science department and these are just computational techniques.

    Is “qualia” bespoke?

    • John Wilkins John Wilkins

      So, you had data, experiment, theoretical models (which you implemented on computers) and explanatory power. Sounds like science to me. But now suppose that you were Superman, and could do the long division, etc., using only a pencil, get it exactly right, and come up with the same conclusion without a computer. I would still call that science. The fact that you have used a computer is no more significant formally than if you used a clay tablet and abacus.

      Oh, and the fact that your topic of investigation is a computer is irrelevant. In that respect it merely is an aspect of the world that you are studying. You might be doing the same thing in researching any engineering problem.

      • Barry Rountree Barry Rountree

        Well, you obviously weren’t making the point I thought you were making. But I still am not seeing the point you were trying to make.

        Specifically:

        “[T]reat[ing] computing advances as science in themselves [is bad]”.

        KPD would say that this is bad because computer science is entirely mathematics and engineering, but you’re not saying that.

        Might this be that when you see yet another incrementally faster processor hailed as a scientific advance, you’re annoyed by this result being described as scientific?

        No, I’m guessing…. Could you give me an example of what it is that you’re finding annoying?

      • John Wilkins John Wilkins

        It’s not that treating computing advances as science is bad, although I think it is because it shows that doing so is a mistake. It’s like saying that a technical advance in car manufacturing is an event in car racing. The subject is distinct. Now reporting that, say, carbon fibre engine mounts have been developed in a car racing magazine is perhaps appropriate, but nobody should think that it is a win in a car race.

        Reporting on new computer software (including informatics software used in science), on new machines and other technologies – this is not science. It’s relevant in science, of course, but so too are advances in lens manufacture, industrial chemistry and the like. But we should be able to distinguish, nearly all the time, between advances in technologies and formal analytic techniques, and science.

        Maybe what you are doing is science and technology or engineering. There is a scientific discipline of engineering research, for solutions to problems that are not yet solved. But applying existing techniques is just engineering, not science.

      • Barry Rountree Barry Rountree

        Hmmm…. that’s odd. I can’t reply to your most recent comment but I can reply to one above it. Anyway….

        I agree that there’s a distinction to be made here, but I don’t think you have the right terms yet. Let’s put the “reporting” issue aside for the moment — we can agree that this isn’t science.

        “Advances in technologies” and “applying existing techniques” is a much more difficult call, I think. What I’ve done could be described as a technological advance, I think — or at least I’m not sure why it wouldn’t be called that. And I certainly used existing techniques to get there — I didn’t invent any new experimental techniques, for example.

        I was going to say that the creation of novel models is what separates science from engineering, but even that isn’t clear cut. A new building needs several different kinds of models to be constructed.

        Could I sell you on “Engineering models instances, science models classes”?

  5. Bob O'H Bob O'H

    Feyerabend, anyone? One point he made in Consolations for the Specialist was that organised crime had some aspects of science. I think what differs is that science studies the natural world. I would take that thought further, but I don’t want to give philosophers cheap sport laughing at me.

    • John Wilkins John Wilkins

      No. Do.

  6. Lurker #753 Lurker #753

    “Computing is a mathematical technique that uses electronic shortcuts.”

    And all biology is really chemistry (and all biologists know it). Interestingly, since the subjects of chemistry are not alive, obviously the subjects of biology cannot be either….

    Computing is not computer science. Computer science was born from mathematics and logic, but since then it has developed an internal complexity, and a complexity of interaction with other fields such that it has acquired a character of its own.

    At what point does a bunch of atoms become alive…. well, no, it’s the behaviour, not the building blocks. At what point does a mathematics-based discipline become a science…. Are you arguing that the answer is “By definition, never”? Surely the question has to be answered by a “test for science” rather than a “test for mathematics”?

  7. So the point seems to be that technological advances should not be reported as science no more so than advances in mathematics should be?

    An analogy: a new stronger type of girder is not architecture. But if the girder allows architects to do things they hadn’t previously been able to do, does that not make it interesting to architects?

    Similarly, if advances in computing (even if it is just engineering and maths) allow new insights into the physical world, are they not of interest to scientists?

    The LHC is just as much “only” and instrument as a telescope is. It’s much more complex, but by your criteria, I don’t understand how the LHC counts as science?

    Meteorology and climate science rely just as heavily on computers as particle physicists do on particle accelerators. Advances in computers could allow better science.

    So the above is just a series of unconnected thoughts on the topic, but here are two questions I think my other comments suggest:
    1) Can there really be a distinction to draw here between things we use to do science and science “itself”?
    2) What value is there in maintaining that distinction?

  8. Sam C Sam C

    At some universities, Computer Science (the academic subject) is situated in the School of Engineering, which is appropriate.

    Building (for example) big bridges isn’t science either, but (like CompSci), it’s a blend of science, mathematics and technology.

  9. The split between math and science can sometimes seem exaggerated to those who think that the whole universe is simply one big computation.

  10. Soren Soren

    Hmm isn’t this just a flare up in the struggle between mathematicians and philosophers?

    The head of a university is talking with the head of physics about the funding for their new particle accelerator.

    He complains – “Why can’t you be more like the Mathematicians? They only need money for paper, pencils, erasers and waste baskets”.
    – “Or better yet like the philosophers, they only need paper and pencils”

  11. Ian Scott Ian Scott

    I am an industrial computer scientist working in medical imaging.

    My work involves the looking at data and the results of existing algorithms on that data, generating hypotheses about the behaviour of very complicated statistical models of the appearance of anatomy in images. I use those hypotheses to develop new algorithms or the new application of existing algorithms to the data. I then run tests of the prediction that these new algorithms will improve the accuracy of the model w.r.t. the data. Failure and success can both lead to refinements of our understanding, new hypotheses and new algorithms/predictions.

    In what way am I not doing science?

    I will agree that I am mostly doing an engineering science as opposed to a natural science – that is I am a doctor of philosophy in the science of the engineered world rather than the natural world.

    From my point of view Wilkins’s complaint appears to one of “I’m not interested in this field and I wish science journalists would give it less attention.”

  12. m m

    The fact that we can make a conceptual distinction between ‘travel’ and ‘means of travel’ does not make the means of travel less ‘core’ to any given voyage. Likewise, for any given act of experimental science the tools and human relations of science ARE ‘core’ to that science experiment.
    Moreover, a philosopher berates those that focus on the ‘science’ of advancing tools at peril of arguing for their own irrelevance. Any given act of science requires the conceptual tools that philosophy provides, as well as hard and software.

  13. jeb jeb

    “who treated the social aspects of science as the core of scientific behaviour”

    It’s the social aspects that interest me but if I explored it as the core behaviour of science I would be looking at it in the same terms as I do religion and they are not the same things.

    I try rather hard to ensure my method of working is not based on faith or fundamental belief or assumptions.

  14. MPL MPL

    Certainly a lot of the computer news on science news pages is not science—the release of a new operating system, for example.

    I think notably, there are several discernible disciplines in computers at the very least: two mathematical ones, one which studies the theoretical basis of computation, and one which studies the properties of numerical algorithms, at least two kinds of engineering: hardware and software, and various sociological opportunities: people who use computers, organizations that make computers/software, societies adjusting to computerization, etc.

    Of course, the real reason the sites are set up to run computer industry news with the science articles, is that they are really just variations on slashdot: they run stories about science that computer geeks might want to read (there is of course also computer news scientists might want to read, but that would show up in different places).

Comments are closed.

Optimized by Optimole