TP

Philosophical questions about programming

Combining philosophy and computer science might appear a bit odd. The disciplines have very little overlap. Both philosophers and computer scientists get taught formal logic at some point in their undergraduate courses, but that's probably as close as they get.

But the fact that the disciplines do not overlap much might very well be the reason why putting them together is interesting. In an article about Design and Science, Joichi Ito (from MIT Media Lab), describes the term antidisciplinary and nicely summarizes why looking at such unusual combinations is worthwhile:

Interdisciplinary work is when people from different disciplines work together. But antidisciplinary is something very different; it's about working in spaces that simply do not fit into any existing academic discipline.

[When focusing on disciplines, it] takes more and more effort and resources to make a unique contribution. While the space between and beyond the disciplines can be academically risky, it (...) requires fewer resources to try promising, unorthodox approaches; and provides the potential to have tremendous impact (...).

As you can see from some of my earlier blog posts, I think the space between philosophy and computer science is an interesting area. In this article, I'll explain why. Unlike some of the previous posts (about miscomputation, types and philosophy of science), this post is quite broad and does not go into much detail.

At the danger of sounding like a collection of random rants, I look at a number of questions that arise when you look at computer science from the philosophical perspective, but I won't attempt to answer them. You can see this article as a research proposal too - and I hope to write more about some of the questions in the future. I wish antidisciplinary work was more common and I believe looking into such questions could have the tremendous impact that Joichi Ito mentioned.

Thoughts? Comments? This is very much a draft and I am very interested in feedback! To make this easier, I also posted the article on PubPub, which is a nice platform for reviewing and commenting. Please share your thoughts there!

How we work

I grouped the questions into three fairly general categories. The first one is mostly a philosophical reflection on how research is done in computer science with, given my own background, focus on programming languages research.

This is perhaps where combining philosophy and computer science could have the most direct influence. If we take a step back and think about what is it that we are doing, perhaps we can discover that our ordinary way of working is not the only (or the best!) option.

How do you tell (computer) science from pseudoscience?

The question how to distinguish between science and pseudoscience is known as the Demarcation problem. This is harder than it seems - programming language research is published in academic conferences or journals, but the criteria for this is that other scientists think it is scientific. Saying that "scientific" is what is "scientific" is not particularly helpful!

Typical programming language paper might include a performance evaluation (if it is implementing something, like a garbage collector) or it might include simple mathematical model of a language feature with a proof (programs in the model do not have certain bugs). Those might be the most common ways of evaluating work in programming language research - but many other disciplines work differently.

Perhaps more importantly, did we choose the above criteria of being scientific because we think that this is what science should be, or did we choose them just because they are easy to assess? Asking whether a programming language is more intuitive or easier to use is not "scientific" - but is that just because our field (somehow) converged on demarcation criteria that exclude such questions?

You could also ask differently - medium is the message and the fact that computer science research is published as papers changes what questions we ask, because we only worry about questions that can be answered in the paper format. Should we be making screencasts and demos or interactive essays instead?

What formal models tell us about the world?

Work on programming and programming languages often involves three layers. There is some intuitive idea, which is turned into actual program and formal reasoning is done about a simplified mathematical model. In particular, programming language research often uses simplified models (take a look at every other POPL paper for example).

One question that is almost never asked is, what is the relationship between these three vertical layers? We simply assume that proving properties in the mathematical model tells us something about the (significantly more complex) program or perhaps even the original idea - but this assumption is rarely made explicit. Is focusing on the mathematical model level leaving out many important questions about the other levels?

In philosophy of mathematics, similar problem appears in the context of informal mathematics and proofs (see Proofs and Refutations). Informal entities allow only imprecise reasoning, while fully formal versions are precise, but disconnected from the original entity. In computer science, we often sacrifice some of the original intuition about problems in favor of working with formal entities that can be treated more precisely. But do we lose something essential about the original problem by doing that?

Unreasonable (in)effectiveness of mathematics

In a famous paper The Unreasonable Effectiveness of Mathematics in the Natural Sciences, physicist Eugene Wigner points out how mathematics often works not just as a model, but also leads to new discoveries in natural sciences like physics. In natural sciences, this is based on a long history of the field.

In other disciplines such as economics, the effectiveness of mathematics has been a lot less unequivocal. In computer science, we also often take the effectiveness of mathematics as granted, but we use it in a very different way than physicists. We do not use mathematics as an analysis tool, but we try to use it as a construction tool. And unlike physics, I'm not sure we have long enough history to support the idea that this method works well.

For example, one big difference between the use of mathematics in natural sciences and in computer science is that physicists cannot change the world to make the mathematics work. They have to tweak the models so that they get close to how reality works. In computer science, when we build something that follows our intuition, but does not quite work mathematically, we can just change it so that the mathematics works. But does this take us further from the original idea, just because we choose mathematics as our construction tool?

What we think

The previous three questions were mostly reflections over how we (as computer scientists) do things. Philosophy lets us take a step back and think why that is the case and whether this is a good way (or, at least, what would be the alternatives).

In this section, I'll take one more step back and focus more at the thinking behind what we are doing rather than at the concrete scientific outcomes. Just like scientific practice has implicit assumptions (the "right way" of doing things), so does thinking have its hidden assumptions.

What we can not think?

This will get a little meta - but when we think about a problem (or even when we think what is a well-formed problem in the first place!) we rely on some broader underlying "apparatus" that makes this thinking possible.

Our modern scientific thinking is certainly very different than the thinking of people in the middle ages. This is not (just) because we are smarter - it is because our thoughts are based on different foundations. In philosophy, Michel Foucault calls this episteme and in science, the concept is similar to Thomas Kuhn's paradigms.

Why this matters? The interesting question is whether there are some things that we cannot even think because they do not fit with our episteme. In computer science, our focus on proofs, measurements and other forms of evaluation might be arising from a single episteme. Would it be possible to think about problems differently? Perhaps in a way that would give more space to linking the three vertical layers (ideas, implementation, mathematics) and less space to moving horizontally (translating abstractions between different mathematical models or comparing language features)?

How we invent abstractions?

One of the core ideas in computer science is abstraction. The idea that we can find common patterns that are more general than a concrete structure, yet capture all its important properties. (This thinking is very likely part of our episteme!) But one question that is almost never asked is where do these abstractions come from? Are we just looking for structural patterns, or do abstractions arise from some intuitive ideas?

Many common abstractions arise from intuitive metaphors. For example, in the First Draft of a Report on the EDVAC, John von Neumann described modern computer architecture. Rather than calling the individual units components or units, he called them organs. You can see a biological metaphor here, right at the foundations of modern computing! Similarly, how did we end up calling programming languages languages? It is yet another metaphor - faithful in some respects and lacking in others.

If we want to understand abstractions, should we be mathematicians, or should we instead become literary critics? And isn't category theory just a source of mathematical metaphors?

Is computer science discovered or invented?

The question whether mathematics is invented or discovered is a fundamental topic in philosophy of mathematics (and there are several other positions too). Are numbers abstract entities that exist independently of humans that we discover? Or are they just things we constructed based on our environments?

You might think this does not have many practical consequences, but I just had to put this question on the list after seeing Phil Wadler's talk Propositions as Types. In the talk, he uses the idea that some mathematical objects (lambda calculus, in particular) are discovered while other programming models are invented to hint that the former are better.

The position in the talk is a bit unusual in that it mixes both philosophical positions on mathematics into a single one (how exactly do we tell which programming model is invented and which is discovered?), but nevertheless, it is an interesting idea that combines computer science and philosophy. And if there was a sound philosophical argument for demarcating between two kinds of objects, should we treat one kind as better than the other?

Historical reflections

In philosophy of science, many of the arguments about the success of particular scientific methods are based on the history. By looking at the long history of natural sciences, we can understand what makes the scientific method so effective and try to replicate the method in other disciplines. Although the history of computer science is not very long, there is certainly enough interesting material there that we can examine when searching for answers to some of the questions above.

How we misinterpret the history?

One thing makes looking at history very difficult. When we try to analyze history through our modern perspective, it is easy to see it through the modern eyes and forget about the original context in which the work was done.

For example, when we talk about Ada Lovelace as the first programmer, we are using a term that not only did not exist in 19th century but, in fact, had a very different meaning even in 1950s when first electronic computers were created! However, precisely because programming did not exist back then, Ada Lovelace's work and thoughts were even more interesting! And if we see her in the context of 19th century, it might be even more fascinating than through the modern perspective of a "first programmer". (Science of Operations has a great chapter on Babbage and Lovelace.)

The other important point when looking at history is that our work is in the hands of its later users (see Science in Action). When you read a foundational paper, it is important because of the work that builds on top of it which often takes the idea in another direction - possibly quite different than what the author intended. What would we learn about our discipline if we approached its history with respect to the original context, rather than reinterpreting it through the modern perspective (using what is known as Whig interpretation)?

How paradigms shape our thinking?

History can also teach us about hidden assumptions in our thinking. Scientists (including computer scientists) often do not question all assumptions in their work. When something does not work as expected, we blame auxiliary assumptions rather than the hard core of the research. This is the idea behind Research programmes as described by Imre Lakatos.

In computer science, one example is the Algol research programme (again from Science of Operations). While the Algol language never got popular in practice, it defined a hugely influential set of core assumptions for academic programming research - the idea of using mathematical logic for ensuring program correctness can be traced back to Algol.

Seeing the "correctness through logic" idea as a core assumption of a particular research programme explains why this is rarely questioned in programming language research, but it also makes it easier to see alternative perspectives. For example, many of the Future Programming Workshop demos follow a different research programme. They are not just early (not yet formalized) works - they are works that may never need to be formalized.

Conclusions

As a scientific discipline, computer science often paints the picture that it is gradually progressing towards some ideal goal - perfect programming language, provably correct programs and so on. Academic papers support this illusion of continuity by building on previous work, even when they completely change the direction of what the original author intended.

I believe that using the antidisciplinary method and combining computer science and philosophy is a great way to bring more innovative ideas to the discipline - be it programming language research or other areas.

A little background in philosophy lets us understand that things are not always as simple as they seem. It is not that easy to distinguish between good and bad science; we often build models, but rarely try to understand how and what they tell us about the world. We use certain methods without worrying about the consequences they might have - both on the results we get, but also on questions we can ask.

This article certainly does not aim to tell you that everything we are doing is wrong. That is not at all what I believe. But I hope to inspire the readers to think about what we do in a slightly different, more reflective, philosophical way. Reading a great book on history of computing or philosophy of science is a great way to get started.

Thoughts? Comments? As mentioned at the beginning I am very interested in feedback! If you have related ideas or comments related to this article, please add them to the PubPub version of this article!

Published: Thursday, 26 May 2016, 1:33 PM
Author: Tomas Petricek
Typos: Send me a pull request!
Tags: philosophy, programming languages