Putting together an expression of interest document for Brighton to host the 2016 Design Research Society Biennial conference I naturally Googled for a few images of Brighton, mainly for inspiration but perhaps to add (with, ahem, appropriate permissions) to my document. One of the things I was looking for was a decent photo of the Brighton University Grand Parade Annex building, completed in 1969 and evaluated by Niklaus Pevsner in his Buildings of England series on East Sussex as:
“one of Brighton’s better postwar buildings and remarkably considerate to its site. Concave front following the curve of Grand Parade with glazed ground floor set behind pilotis. Two bands of fenestration above, articulated in a rhythm which reflects the width of the Regency terraced frontages which follow”.
The search I asked Google for was ‘Grand Parade, Brighton’. I clicked on the image link to see what I had. The photo below is what grabbed my attention, and that forced me down a route of mild diversion.
The image is of a black and white postcard, from around 1930, and which shows an aerial view of the area of Brighton near where the Grand Parade building is now located (top right). The hand-written caption suggests that this is a view from an aeroplane and of Grand Parade, but once you know that St Peter’s church tower stands about where the photo is taken from it’s clear that this wasn’t taken from a (scarily low) aeroplane and isn’t really of Grand Parade. The camera looks down on Richmond Place, behind which is dense Victorian housing. Tramcars make their way along the road, past shops with awnings pulled down on outside displays.
The curve of the road and urban division between Victoria Gardens on the right, and buildings on the left is still preserved today, though the streetlight in the middle of the road is gone. Most of the buildings that line Richmond Place still stand, as do some of the houses in the background but many have been demolished. The ones in far-view to the left of Victoria Gardens as we look have given way to new office space. Car parks, a covered market, a police station, a dole office, the law courts, and the American Express headquarters now account for the other demolitions on the rising slope behind.
The postcard invited a visual comparison between past and present, ideally from the same position. Once again I turned to Google, but looking down the previous search I couldn’t find anything similar. Google now does an image search, where you can search for an image with an image, and I figured with the structure of the urban environment similar between past and present, an image search might throw up a modern version.
All the search results were in black and white of course, that’s the most obvious thing about the original, and something I’d forgotten to take account of. The image below was the number one returned image.
The image shows a photograph, apparently from the holocaust, of about 50 women, most dressed in the ‘stripped pyjama’ concentration camp uniform, though around 10 are nuns, dressed in habits. The women are funneled between what looks like barbed-wire fences left and right. Perhaps they are in an open train carriage? There is no uniformity in the attention of the women standing, suggesting that this might be a photograph taken from a film. Some look sideways, some to camera, others backwards. They seem to be going somewhere, but either going there very slowly or stopped altogether. There is enough blur in the far background to suggest that the image might have been digitally contrived or altered, but in most respects it appears convincing. The question is, why might Google consider it is anything like the first image above?
Once one has read William Mitchell’s excellent book The Reconfigured Eye one tends to think of digital images slightly differently. As collections of optimised pixels, graded and toned, or as scripted algorithms that allow a computing device to ‘make’ an image. One tends to think less of image content, and more of the patterns of pixels; about what a computing device can technically assume about an image in recomposing it to a level where you won’t notice the bits that are missed off or averaged out. If one was to average out the greyscale of both images into one uniform tone – and that involves taking the writing on the postcard into account, something that we tend to ‘see through’ as content – the two images might be similar, might even be exact copies of one another in terms of a block of grey. But it seems unlikely, the overall tone appears different to my eyes; very slightly brown in the top image and very slightly green/blue in the bottom image.
What seems more likely – discounting any conspiracy theory – is that the holocaust image is simply one that is, or has been, very popular in search terms. Google is saying to me: have a look at this one, which a lot of other people have clicked on, you might find it is similar to your other one. And perhaps a lot of people do, like me, click on this image in disbelief (and perhaps horror), that it could be anything like the image that they are searching for. For Google though, that click means that it has got it right. Your click means that you think it is like the image you are searching for. So next time someone searches for a similar black and white image, Google can be reasonably sure that the concentration camp image might stand a good chance of being like it. In fact it stands a good chance of being like quite a few black and white images. In Google’s mind all images might all lead back, or forwards, to the holocaust.
Which might turn out to be a very sophisticated understanding of the human condition. We think we’re searching for something, but what we’re really searching for is something else; something unconscious. For other people’s misfortune perhaps. If Google really is learning, as Daniel Soar suggests, then it is either learning pretty damn quickly or it needs to search a dictionary to find a definition of what learning is.