A difficulty with DoubleQuotes
[ by Charles Cameron — from artificial intelligence to the Council of Nicea in one easy blog post ]
It seems fairly easy for a human to tell likes from dislikes, but for a computer to tell likes from unlikes appears to be a far trickier business. Consider the following DoubleQuote in the Wild, which I found in David Berreby‘s Nautilus piece, Artificial Intelligence is Already Weirdly Inhuman:
You might think these two images are the same. Or that they’re a little different, as the images from your left and right eyes always are, but that if you squint at them just right they will merge into a single image with a vivid sense of depth, like a movie seen with 3-D glasses. You mivght even think the differences between them are a matter of steganography, encoding some IS battle plan under cover of a diggie pic.
But you are unlikely, I suggest, to think the image on the left is of a dog, while that on the right is of an ostrich. Which is what “artificial intelligence”, in the form of a neural net, figured out.
And how different are they “in fact”?
The middle image here shows the amount of variation in pixels between the two outer images:
The image on the right — the one the neural net iodentified as an ostrich — is an example of what the researchers, Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow and Rob Fergus, call an “adversarial example”.
It’s not my intent to dismiss neural nets by any means: I have one myself.
What interests me, though, as someone preoccupied with analogy and metaphor — with likeness and unlikeness — is the deep question of what likeness and unlikeness mean.
That question lies at the heart of my DoubleQuotes and HipBone Games.
Back in my Oxford days, my tutor in Dogmatic Theology had me thinking about the difference between the two words Homoousion and Homoiousion, homoousion meaning of the same essence, and homoiousion of similar essence. The distinction was important in Patristic theology, the questionn being whether the Son and Holy Spirit were of the same essence as the Father (one God in three Persons) or of similar essence (three Persons in one God).
You’ll see from the way that I’ve phased the distinction in brackets (one God in three Persons vs three Persons in one God) that I find the distinction itself less than helpful — and I said so in the essay I read my tutor. Those who hold the Three Persons are the same God (the homoousios doctrine) are saying they are both similar as to the recognition of their common Godness and dissimilar as to the recognition of their separate Personhood, whereas those who hold that they are of similar essence (homoiousios) are, perhaps unexpectedly, also saying they are similar but different: it’s all a matter of emphasis.
My tutor, much to my surprise and delight, mentioned that he had made the same point in a paper he had recently published in, if I recall, the Journal of Theological Studies, and gave me a signed offprint.
Similarity and dissimilarity, likeness and unlikeness appear to me to find themselves on a spectrum which approximates closely to identity at one end — but if two things are identical, how can they be two? — and absolute distinction at the other.
Yet the difference beween homoousion and homoiousion was decided in favor of homoousion at the Council of Nicea, a decision which one writer calls a “bloodless intellectual victory over dangerous error” and “of far greater consequence to the progress of true civilization, than all the bloody victories Constantine and his successors.”
And okay, there’s more to it, as always…
Dogs and ostriches, apples and oranges — what’s the diff, eh?
And G*d knows best.
August 10th, 2015 at 12:34 am
And, yes, this is the article I linked to in my post. I suppose we both got there from 3QD.