The problem is in #2. Imaginary cheese is not a kind of cheese.
Edit: I’m not entirely sure this is where I saw it first, but this forum post (ironically, on a Catholic forum of some sort, apparently discussing whether certain games such as Magic are evil...) makes the argument excellently.
Edit2: In fact, I daresay an excerpt from said post is good enough to post as a rationality quote on its own, which I will now do.
I’m not sure I like this phrasing although the essential point is correct. I’d say rather that generally when one uses a word one implicitly has “actual” or “real” in front of it. Adding the word “actual” at the relevant points in the argument makes the problem more clear.
What is the position of imaginary cheese in thingspace, relative to the position of the cheese similarity cluster?
Along most dimensions (those relating to physical properties, most causal properties, etc.), imaginary cheese is quite far removed from actual cheeses. Along a couple of dimensions (verbal description, perhaps something like “what sorts of neutral firings are involved in perceiving it”), imaginary cheese is closer to actual cheeses.
To take a two-dimensional example, perhaps gouda is at (4,6), cheddar is (5,3), mozarella is (3,7), provolone is (3,5)… and imaginary cheese is, say, (100,4). Within the cluster if you look only at the y dimension, quite distant from it if you look at all dimensions. And if we actually plotted cheese and imaginary cheese in some suitably higher-dimensional space, there’d be a lot of dimensions like x in my toy example (along which imaginary cheese is far from actual cheeses), and few like y (along which imaginary cheese is close to actual cheeses). Out of those dimensions in which cheeses form a cluster, most would be like x, few like y.
Edit: the basic issue is that things cluster in thingspace; categories into which we place things are reflections of that clustering. What things do not, in fact, do is fall neatly into classes and subclasses that might seem natural to us, like objects in Java, where if you have e.g. ImaginaryCheese extends Cheese (i.e. the ImaginaryCheese class is a subclass of the Cheese class), then ImaginaryCheese is guaranteed to inherit any and all properties of its superclass Cheese. All we really have is approximations of this behavior, to a lesser or greater extent, e.g.:
GoudaCheese behaves more or less like a subclass of Cheese; most relevant properties of Cheese (that is, properties shared by all things within the main body of the Cheese similarity cluster) are in fact inherited by GoudaCheese… because, of course, GoudaCheese is within the main body of the Cheese similarity cluster.
Conversely, ImaginaryCheese is not within the main body of the Cheese similarity cluster, so we shouldn’t expect it to behave like a subclass of Cheese… and it doesn’t.
So an alternate response to the logic in the great-grandparent (Nisan’s comment) might be:
Yes, some cheese is imaginary. You can’t get it anywhere because, unlike most cheeses, imaginary cheese isn’t “a thing you can get”. This is not a problem because reality doesn’t (apparently!) feature strict class hierarchies.
In fact, the problem with the reasoning is that while you could construct a strict class hierarchy, the properties you could assign to the Cheese superclass would be only those shared by all cheeses… and if you’re drawing the boundary around the similarity cluster such that ImaginaryCheese is within the boundary, then “existence outside of minds, and therefore ability to be ‘gotten’” would not be one of those shared properties.
The problem is in #2. Imaginary cheese is not a kind of cheese.
Edit: I’m not entirely sure this is where I saw it first, but this forum post (ironically, on a Catholic forum of some sort, apparently discussing whether certain games such as Magic are evil...) makes the argument excellently.
Edit2: In fact, I daresay an excerpt from said post is good enough to post as a rationality quote on its own, which I will now do.
I’m not sure I like this phrasing although the essential point is correct. I’d say rather that generally when one uses a word one implicitly has “actual” or “real” in front of it. Adding the word “actual” at the relevant points in the argument makes the problem more clear.
What is the position of imaginary cheese in thingspace, relative to the position of the cheese similarity cluster?
Along most dimensions (those relating to physical properties, most causal properties, etc.), imaginary cheese is quite far removed from actual cheeses. Along a couple of dimensions (verbal description, perhaps something like “what sorts of neutral firings are involved in perceiving it”), imaginary cheese is closer to actual cheeses.
To take a two-dimensional example, perhaps gouda is at (4,6), cheddar is (5,3), mozarella is (3,7), provolone is (3,5)… and imaginary cheese is, say, (100,4). Within the cluster if you look only at the y dimension, quite distant from it if you look at all dimensions. And if we actually plotted cheese and imaginary cheese in some suitably higher-dimensional space, there’d be a lot of dimensions like x in my toy example (along which imaginary cheese is far from actual cheeses), and few like y (along which imaginary cheese is close to actual cheeses). Out of those dimensions in which cheeses form a cluster, most would be like x, few like y.
Edit: the basic issue is that things cluster in thingspace; categories into which we place things are reflections of that clustering. What things do not, in fact, do is fall neatly into classes and subclasses that might seem natural to us, like objects in Java, where if you have e.g. ImaginaryCheese extends Cheese (i.e. the ImaginaryCheese class is a subclass of the Cheese class), then ImaginaryCheese is guaranteed to inherit any and all properties of its superclass Cheese. All we really have is approximations of this behavior, to a lesser or greater extent, e.g.:
GoudaCheese behaves more or less like a subclass of Cheese; most relevant properties of Cheese (that is, properties shared by all things within the main body of the Cheese similarity cluster) are in fact inherited by GoudaCheese… because, of course, GoudaCheese is within the main body of the Cheese similarity cluster.
Conversely, ImaginaryCheese is not within the main body of the Cheese similarity cluster, so we shouldn’t expect it to behave like a subclass of Cheese… and it doesn’t.
So an alternate response to the logic in the great-grandparent (Nisan’s comment) might be:
Yes, some cheese is imaginary. You can’t get it anywhere because, unlike most cheeses, imaginary cheese isn’t “a thing you can get”. This is not a problem because reality doesn’t (apparently!) feature strict class hierarchies.
In fact, the problem with the reasoning is that while you could construct a strict class hierarchy, the properties you could assign to the Cheese superclass would be only those shared by all cheeses… and if you’re drawing the boundary around the similarity cluster such that ImaginaryCheese is within the boundary, then “existence outside of minds, and therefore ability to be ‘gotten’” would not be one of those shared properties.
Wonder what the author of that post was banned for.