It should go without saying that the way we interact with photography has undergone a seismic shift since the days of the album or the shoebox full of snapshots. More and more of our photographs are never printed but stored on our computers and shared through social networking sites like Facebook. We also have far more photographs than ever before, so we need new ways to store and organise these images. Metadata – which the Oxford Dictionary defines as: “A set of data that gives information about other data” [1] – is a critical component of our new photographic ecology, as it exerts a huge influence on both the storage and the visibility of our photographs.
The dictionary definition may make metadata sound fairly innocuous, but it is not simply another layer of information and we should not treat it as such – it is, to quote Daniel Rubinstein and Katrina Sluis, “highly political” with the “potential to contaminate, mutate or change the direction and context of the image at every turn.” [2] While I will be writing about metadata in the context of digital photography, it is worth highlighting that metadata can come attached to any kind of data object. Indeed, the word ‘metadata’ has never been far from the headlines in recent months, following the revelation that security agencies including the NSA and GCHQ have been indiscriminately collecting huge amounts of metadata from Internet users. [3] As this case so clearly illustrates, metadata can have implications that go far beyond helping us to organise our photographs and other digital paraphernalia.
According to Rubinstein and Sluis, the metadata attached to digital photographs falls into two broad categories:
descriptive metadata which is generated mechanically during image creation or added later to the file (containing details such as date, location, camera make, owner and keywords) and is carried within the file; the second type is collected as a valuable by-product of interaction with the image (tags, comments, ratings, number of viewings) and is stored independently of the image. [4]
To that, I would add a greater distinction between ‘mechanically-generated’ and ‘user-generated’ descriptive metadata. Although both describe the image in some way, one is a mechanically recorded quality and the other is (at least partly) a matter of personal interpretation. This is not to say that there is no interpretation or error present in mechanically-recorded metadata, but to clarify its source and the expectations placed on it.
Anyone who has ever used an Internet search engine will know not only something of the importance descriptive tags, but something of their unreliability. Furthermore, these tags highlight metadata’s role as a mediator between humans and computers.
By way of illustration, assume that I want to search Flickr for a photograph of a Siamese cat. The Flickr search engine will try to find me a photograph of a Siamese cat not by looking at the content of photographs for something that looks like a Siamese cat, but by searching the metadata attached to those photographs for the descriptive tags ‘Siamese cat’. The search engine does not ‘know’ what a Siamese cat – or any other breed of cat for that matter – looks like, so it is entirely reliant on user-added tags to identify what is in the photograph. [5]
Carrying out a search on Flickr for ‘Siamese cat’ does return a large number of photographs of that breed, but it also returns a high number of photographs of cats that are very clearly not Siamese cats. [6] The reason for this is simple: there is no set of rules for tagging content on Flickr and so users are free to tag photographs any way they like, as such there is absolutely no way of ensuring that the tag accurately reflects the content of the photograph.
Why is this potentially problematic? I know that a lot of the photographs are not photographs of Siamese cats because I know what a Siamese cat looks like – but what if I did not? If I were presented with a large number of photographs of Abyssinian cats all tagged ‘Siamese cat’, what reason would I have to suspect that I was not looking at a photograph of that breed?
One might argue that this scenario is no different from wrongly captioning a photograph; and perhaps from my point of view attempting to find a photograph of a Siamese cat it is not. The difference is that, as far as the computer is concerned, so long as a photograph is tagged ‘Siamese cat’ it is a photograph of a Siamese cat. That is until it is relabeled ‘Abyssinian cat’, then it is a photograph of an Abyssinian cat. That is until it is relabeled ‘Labrador puppy’; and so on. For the computer, the tag does not contextualise the photographs but “establishes complete identity between image and text” and furthermore assumes that images can be exhausted by description. [7]
Metadata then is not simply a mechanism to help us find an image in a database, it governs the image’s visibility and has the power to impose different meanings on the photograph. Tagging, however, is only one small part of metadata; one element of ‘user-generated’ descriptive metadata. What of ‘mechanically-generated’ descriptive metadata?
As suggested above mechanically-generated descriptive metadata (what I have often referred to as ‘captured data’) is composed of data captured at the moment of exposure – the geo-location recorded by an iPhone’s GPS receiver for instance. Does the fact that it is mechanically (or ‘automatically’) generated mean that we treat it differently from that metadata which is created by humans (metadata that we are accustomed to being subjective or incorrect)? In other words, do we have different expectations of this data? My instinctive answer is ‘yes’.
In my last post, I noted that Susan Sloan describes motion-capture data as a ‘certificate of presence’, acting in much the same way as a photograph as evidence that someone or something was there before the sensor. [8] Can a piece of geo-location data not function in the same way? Attached to a photograph, a geo-location tells us that someone stood in this spot and took a photograph – marking the photographer in a very particular place and time. But how convincing is this data?
Let me answer that question with another question: How many television detectives have disproved a murder suspect’s alibi by producing their mobile phone records (metadata) to place them at the scene of the crime? In this scenario, the implication is that mechanically-captured descriptive metadata does not lie, much in the same way that a photograph does not lie. While these might be fictional examples, metadata is permissible evidence in a court of law. [9] Therefore, like photographs, metadata has a certain evidentiary force.
In Camera Lucida, Barthes describes what he sees as the photograph’s “power of authentication”: a power that he saw as exceeding that of representation. [10] In one passage, he describes receiving a photograph of himself that he had no memory of having sat for; yet for Barthes the very fact of the photograph’s existence proves that he must have done so. [11] Such is the power of the photograph’s “evidential force”. For Barthes, this evidential force is linked to the authority of the camera, which in turn is linked to its automatic, mechanical nature. [12] If the geo-location is recorded automatically as the photograph is taken could it not too serve as evidence? Can it not convince us that the photographer was indeed present at a given location?
Mobile phones are perhaps the most instructive in this sense as we tend to carry them wherever we go and most have built-in GPS that can record our location when we take a photograph, or open any number of other apps that make use of our location – such as maps or running route trackers. They also tend to be very personal devices, without multiple users. I have a photograph on my phone of a nondescript section of pavement, which according to its metadata was taken at 51,30.55N, 0,7.03W on 10th November 2012 at 17:30:51. If I consult a map, I can see that these co-ordinates place the photograph on Waterloo Bridge in London, which I have no immediate reason to disbelieve as at the time I was living in London and regularly walked across Waterloo Bridge. Much like Barthes, it is difficult for me to argue with this data: it is on my phone and since I am rarely without my phone the most logical explanation is that I took this photograph and therefore was on Waterloo Bridge at the time. Until I find evidence to the contrary I am inclined to believe it.
–––
I sense that this post has perhaps taken me into slightly murky waters, as I previously wrote that indexicality has “little to do with visual likeness and perhaps even less to do with truth”. [13] In asking if metadata might be considered a digital index, by attempting to argue its power as an evidentiary force – do I not too then make a link between indexicality and truth? The two are related, but the relationship is far more complex than ‘it is indexical, therefore it must be true’.
Photographs are visually compelling and we widely believe them not to lie because we know they are drawn from life, but, as Lewis Hine wrote, “Photographs don’t lie, but liars may photograph.” [14]
Consider Robert Capa’s photograph Death of a Loyalist Militiaman (1936), the veracity of which has long been a source of debate. [15] What is not in dispute is that the soldier appeared before the lens of the camera, the negative carries the imprint of light from that moment and therefore the photograph is indexical. The dispute arises over whether or not the photograph actually shows what Capa claims it shows. Does it really show a militiaman the second he was hit by a bullet? Or does it show a militiaman slipping on a sandbank? As William J. Mitchell points out, both explanations apparently fit with the visual evidence presented in the photograph; and in the absence of contradictory evidence we are inclined to believe the explanation presented to us. [16]
There is much more to be said about Capa’s photograph and the status of photographs as evidence in general. This post, however, is not the place to discuss them. What I have outlined here may only be a very simplistic example, but hopefully it serves to illustrate the gap between indexicality and truth.
Notes
[1] Oxford Dictionaries, “Metadata” [Online], http://www.oxforddictionaries.com/definition/english/metadata (accessed November 18 2014).
[2] Daniel Rubinstein and Katrina Sluis, “Notes on the Margins of Metadata: Concerning the Undecidability of the Digital Image,” Photographies 6(1) 2013: 151 – 158.
[3] “The NSA Files,” The Guardian [Online], http://www.theguardian.com/us-news/the-nsa-files (accessed 21 November 2014).
[4] Daniel Rubinstein and Katrina Sluis, “Notes on the Margins of Metadata,” 51.
[5] It is true that computer vision has recently taken great strides towards image recognition, but we are still some way from the point where the computer will be accurately able to describe the contents of any photograph. See: Samy Bengio, et al., “Show and Tell: a Neural Image Caption Generator” [Online], http://arxiv.org/pdf/1411.4555v1.pdf (accessed 21 November 2014).
[6] Search carried out on 21st November 2014. I did not see any photographs that did not feature a cat at all, but that does not mean that they do not exist.
[7] Daniel Rubinstein, “Tag, Tagging,” Philosophy of Photography 1(2) 2010: 197 – 200.
[8] Susan Sloan, The Synthetic Photograph [Panel Discussion] (London: The Photographer’s Gallery, 14 March 2013).
[9] Thomas R. McLean, et al., “Electronic medical record metadata: uses and liability,” Journal of he American College of Surgeons 206(3) 2008: 405 – 411.
[10] Roland Barthes, Camera Lucida (London: Vintage, 2000), 89.
[11] Ibid., 85.
[12] For John Tagg, the documentary value of the photograph is also closely linked to standardisation; to strict rules that must be followed in order to produce photographs with little variation that can serve as evidence in a court of law. The requirements of passport photographs are a good example of this standardisation. See: John Tagg, The Burden of Representation: Essays on Photographies and Histories (Basingstoke: Macmillan, 1988).
[13] Catherine M. Weir, “The Index in Digital Photography” [Online], http://www.cmweir.com/the-index-in-digital-photography/ (accessed 21 November 2014).
[14] Quoted in: Mia Fineman, Faking It: Manipulated Photography before Photoshop (New York: Metropolitan Museum of Art, 2012).
[15] Richard Whelan, “Proving that Robert Capa’s “Falling Soldier” is Genuine: A Detective Story” [Online], PBS American Masters, http://www.pbs.org/wnet/americanmasters/episodes/robert-capa/in-love-and-war/47/ (accessed 25 November 2014).
[16] William J. Mitchell, The Reconfigured Eye: Visual Truth in the Post-Photographic Era (Cambridge, MT: MIT Press, 1992), 43.