DRMacIver's Notebook
Ubiquitous Incommensurability
Ubiquitous Incommensurability
We were talking in bookchat about favourite books the other day, and how difficult some people find the problem of listing, say, your top ten favourite books.
I claim the reason that this question is hard is that it’s actually impossible, and not in a good way, because even within a category such as books, many things are incommensurable - it doesn’t really make sense to compare them in such a way as to say that one is definitively better than another.
In order to illustrate this, let me tell you about two of my favourite books. Sorting Things Out: Classification and its Consequences and A Civil Campaign.
Sorting Things Out is a nonfiction book about the way classification systems are developed and how this affects the people interacting with or being classified by them. A Civil Campaign is a book late in a space opera series that is mostly a romantic comedy about marriage and edible bug byproducts.
Both of these are favourite books of mine in the following senses:
- My life is significantly improved by having read them.
- I would unambiguously recommend them to anyone who thought they sounded appealing.
- There is no book I strictly prefer to them.
Which of these books do I like better? I don’t think the question makes sense. Each book serves a very different role for me, and those roles are not at all interchangeable.
There is an idea in economics of revealed preferences, which suggests that preferences can be inferred from behaviour. Von Neumann-Morgenstern Utility Theory uses a similar idea to attach a single scalar utility value to a set of things being chosen over. These may both be true in some sense, but a naive adoption of them hides something very important: How incredibly context sensitive choices between objects are.
You might imagine that you can find out which of these books I prefer by asking me to make decisions, but the problem is that which book I choose will be heavily influenced by how you ask the question.
Some possible variants:
- If I had to give away one of these books, which would I pick? A Civil Campaign, because Sorting Things Out costs more to replace. (Advantage: Sorting Things Out)
- If I were given an extra copy of one of these books, which would I pick? A Civil Campaign again, because I know many more people who would enjoy it than the former and I’d gift that on. (Advantage: A Civil Campaign)
- If I could only ever read one of these books again, which would I pick? Sorting Things Out, because I use and refer to its ideas much more often. (Advantage: Sorting Things Out)
- If I were stuck on a desert island with one of these books, which would I pick? A Civil Campaign, because it’s much more emotionally comforting to reread and I’m not likely to be up to much more than that on a desert island. (Advantage: A Civil Campaign)
One way to think of this is in terms of pareto optimality: You can only count a book as better than another if it is better in all degrees. I think this is too strong - for example I’d consider either of these books better than, say, The Tao of Pooh, because I consider the latter to be a dreadful book that I would not recommend to anyone. Are there senses in which The Tao of Pooh is better than either of these books? Yeah, probably. It’s more accessible than Sorting Things Out for example, but I would still say Sorting Things Out is the unambiguously better book.
Nevertheless, I think it is still true that many pairs of books are incommensurable - you cannot readily say that one of these things is better than the other, because “better” is so context dependent. Further, nothing about this is book specific - the same is true for meals, films, people, and almost everything else. Context is important a driver of preference between objects as the objects themselves.
You can certainly demand that I reduce those contexts to probabilities, and make decisions about what I would prefer in each context to reveal my preferences, in order decide which one I strictly prefer by comparing the resulting subjective utilities.
To this demand, I offer the following rebuttal: Nope. Shan’t.