Now is the part where we start training the AI that we don't know we're training.
Only now? What about GMail and how they allow themselves to 'automatically' cruch through it? Google Voice and how it transcribes your voicemails? Google Translate which is trained on everything we put on the net...
Any sufficiently advanced computation is indistinguishable from intelligence.
--Arthur Clark, paraphrased
Peter Norvig's talk at startup School 2008 (working link anyone?) suggests how they might have been using this technology internally.
So, how might they have been using it internally?
For regular Image Search. Among images matching the search query, those central in the image-similarity graph are probably more relevant.
I wonder how popular this feature will actually get. From an AI standpoint, its pretty cool. But personally I haven't run into very many instances where if I search for some image, I find it, and now I want another one that looks just like it.
I don't think that's the use case. I think it's meant for when you don't find the image you're looking for, but find one that's close, or at least in the same ballpark. Then you can tell Google to find images that are close and it might find exactly the image you're looking for.
I personally think it's awesome.
I frequently use http://tineye.com to find images in a bigger resolution. Its also very handy if you want to find the author of a particular image/picture. Their bookmarklet is also extemely useful.
This is very nice feature.
Yet, I tried it and it doesn't work for all pictures yet.