Justin Li

Invisible Colors


Often, when I have conversations with non-computer-science people, these "conversations" often resemble "arguments". One recurring argument is how we decide whether something has a particular property. To give a concrete example, we might try to figure out whether a critical review (of a book, say) is authentic, that is, that it was derived from the true beliefs of the literary critic. It turns out that the disagreement runs deeper than literary criticism, and I want to address that in this post.

Sidenote: the core ideas in this post were inspired by another article. I think I push the idea a little further here, and also give better examples, so feel free to read the article and come back. At the very least, it’ll plug any explanation pitfalls in this post.

The big takeaway of this post is that sometimes we care about properties of objects that cannot be found in the make up of that object. To use a simple examine, my name is “Justin”, but if you take me apart atom by atom, there is no “Justin-ness” property to be found anywhere. Note that this is not simply the distinction between a system of components versus individual components; the “circulatory system” may not reside in any individual atom, but we can point to some group of atoms (ie. the heart, blood vessels, etc.) and say that they make up the “circulatory system”. In this case, the circulatory system is the result of the interactions of atoms; the same cannot be said of the name “Justin”.

Maybe a real-world example would make the point clearer. In business, when you receive funding from some source, sometimes there will be stipulations for how you may use that money; this is the case for the US government. Closer to home, the Computer Science and Engineering Graduates (CSEG) student group gets money from the University of Michigan, but they can’t use that money to buy alcohol. At the same time, CSEG also collects money from graduate students; this money can be used to buy alcohol. To keep these funds separate, CSEG actually has two bank accounts, and is not allowed to arbitrarily move money from one to the other (for some definition of arbitrary). There is no difference between the money in these two accounts, of course; money is fungible, and merchants would happily accept money from either. But to abide by university regulations, CSEG must distinguish between the two, and only use one of them to buy booze. In finance, this distinction is called the Color of money, a term I will adopt here (and also used in the source article above, although for a different reason).

If we want to get technical, Color-like properties are ones that do not exist in the object itself. We can take apart CSEG’s two bank accounts (they won’t be happy about this), and nothing in the account will say that one can be used to buy beer. Nothing in the external world can tell us the Color of the accounts; instead, Color is something that we need to keep track of ourselves. In memory research, properties like Color are called “non-projectable”, because the outside world cannot give us this information (if you want to be poetic, it cannot “project” the information into our heads). What this usually means is that Color is not about the object itself, but about the history of the object. When we say that this account has a Color of “cannot buy booze”, what we mean is that the money in this account came from the university and not from students. It’s how the money got there, not what the money is, that defines this distinction.

Since Color is not an intrinsic part of an object, we can do strange things with it. For example, CSEG can do tricks to change the Color of money, by paying for some social event (for example, rock climbing), then charging grad students to attend. The money with the “no booze” Color is now spent (notably, not on booze); the money that CSEG now has came from students, and so has the “yes booze” Color, even though these two may be the exact same amount. We have now changed the Color of money, and can do different things with it (eg. a bar crawl). This is, of course, a form of money laundering, but no one seems to mind.

I want to return to computer science now, and give another example of Color: whether a file is subject to copyright (this was the example in the source article). Let’s say I have three copies of a song (say, this one), which I got from different places: one I got from the artist as a gift, one I got as a digital download from iTunes, and one that I got from a BitTorrent file-share. From the computer’s perspective, these three files are exactly the same, down to every 1 and 0 at the binary level. (This is not strictly true, but we’ll run with it for now.) The computer can’t stop me from making a copy of my iTunes version for a friend – not because it doesn’t have the power to, but because it doesn’t know which one came from iTunes. There’s nothing in the bits that say whether I’m legally allowed to copy a file; the legality of that operation is a Color.

For obvious reasons, this is a big problem for lawyers. To make sure everyone is paying for their music, they want to impose restrictions on what people can do with their files, so they come up with digital rights management (DRM). They require iTunes to add extra 1’s and 0’s to their files, so that my computer knows it’s from iTunes, and can stop me from making copies; in computer science jargon, this is called metadata, since it’s data about the data itself. The problem is that metadata doesn’t change the file in any intrinsic way; it merely adds some numbers to the end. If I want, I can decide to add this metadata myself – or more likely, remove this metadata – and all of a sudden my computer can again copy files from iTunes.

(In reality, there’s the issue of encryption – changing the file in some way that requires special knowledge to read. This is a bigger problem, since encryption is designed to make its reversal – “removing the metadata” – difficult. But none the less, if the encryption can be reversed (ie. if the file can be decrypted), the file can again be read and copied.)

If all this computer science stuff seems complicated, it turns out that the same problem exists in meat space. Copyright, ultimately, is about the ownership of an idea, and ownership applies equally well to real objects. Let’s say you and I both own the same coffee mug, which are identical in every way. I can take both, move them around behind my back, then bring them back out and ask you which one is yours. You can’t tell, of course, because the mugs are identical; you are now in the same position as the computer with the three files.

“But wait!” you say, “I can just stencil my name onto my cup, and now my cup is different! What do you think of that?” You’re right, of course, but that’s the point: your name on your cup doesn’t really mean that you own the cup. I can, for example, also stencil your name on my cup; nothing stops me from doing this, and the cup is still mine, even though it has your name on it. The ownership survives regardless of what changes you make to your cup and what changes I make to mine; it does not depend on the physicality of the cup, and is therefore a Color.

(Encryption in this case, I suppose, is equivalent to locking up your mug. Even then, nothing stops me from picking the lock and claiming the cup is mine.)

As the source article pointed out, it’s not that computer scientists don’t care at all about Color. When we need a random number (excuse me, a randomly-generated number), we are a lot that the numbers are actually random. Theoretically, rolling a die and getting five 6’s consecutively (6, 6, 6, 6, 6) is just as likely as getting any other sequence (say, 4, 6, 5, 1, 6); they both have (16)5=17776=0.000129 probability of happening. But, like any Color, randomness is not in the sequence itself, but in how the sequence is generated. Which is why we care about how a random number generator works, precisely because we can’t tell just by looking at the numbers. In the ideal case, we’ll have the source code, so we can mathematically prove that the generated numbers are random; in practice, we leave this job to a small subset of people, and trust that whatever random number generator we use is good enough.

I am finally ready to go back to my opening example about authenticity in literary criticism. In real life, the discussion came about because my friend mentioned that in the school of New Historicism, critics frame the piece of literature in its historical context; they might, for example, comment on how colonialist ideas show up in Shakespeare’s works. This is notable because the idea of colonialism wasn’t around during Shakespeare’s time; it is only in hindsight that we can point out these influences. I then asked the question of whether a critic can pretend to be from a different time period, and review a work from that perspective; an example might be to review Shakespeare as someone from a inter-stellar civilization. My friend was rather horrified, and said that such a review would be science fiction and not literary criticism, that it would not be authentic. I then modified my example: what if a critic is familiar with modern psychology, but pretended not to be, and wrote a review using psychoanalytic theories? Would such a review be accepted into literary journals? My interest at the time was more about the need for scientifically-accurate literary criticism, but it also touched on the issue of authenticity. Given my friend’s previous vehement reaction, I was surprised to hear that such a review would be publishable, despite, of course, it also being inauthentic.

As might be obvious by now, the problem is that the authenticity of a review is not a property of the review itself, but a Color. Two critics can have completely difference view points but write the exact same sequence of words in a review; nothing in the words would distinguish which one is authentic and which isn’t. To be fair, this is not just a problem for literary criticism, but for academia as a whole: any manuscript to be reviewed for publication should be true, but that truth is not a property of the manuscript, and therefore cannot be determined 100% accurately every single time. The difference between literary criticism and (say) computer science is that in computer science, the manuscript contains experimental results; this allows the experiment to be replicated and the results reproduced, thus checking the correctness of the manuscript (albeit after it has already been accepted or rejected for publication). This path to determine correctness (or the criticism equivalent, authenticity) is not open to literary criticism: there is nothing to compare a review against, since a review is by nature subjective, and a different critic cannot “reproduce” the views of the original author. Of course, we’ve actually already encountered this problem before. Correctness in the sciences is like having the random number generator available; if we don’t believe the sequence is random, we can look at the random number generator itself and “reproduce” the results. Authenticity, on the other hand, is like being given a black-box random number generator; we might be able to say something is not authentic if it is painfully obvious (like if the generator always produces 6’s, or if a critic claims to be part of an inter-stellar civilization), but for less egregious violations, there’s simply no way to tell.

I didn’t mean this post to be an attack on literary criticism; it was merely a convenient example (and a true and authentic one; you’ll just have to trust me on this…). The issue of Color comes up in many other places, including within computer science itself, and I’m sure has sparked many debates similar to the one between me and my friend. Color is therefore a useful thing to keep in mind, especially since it makes an appearance in both the sciences and the humanities. Instead of arguing over why it is unrealistic to care about some property, we might invoke the concept of Color and move on to whether and how accurately we can determine that Color instead.

PS. I do want note the irony in how literary criticism as a field cares so much about authenticity, given that they spend a lot of effort separating the creator and the work created; this is the (famous?) “death of the author” view of literature. Somehow, this philosophy does not extend to criticism itself.