James Elkins Images without Sense A microscope image (top), with some of the results obtained by Google’s ›Search by image‹ function, February 2012 The two images I list as ›unidentified‹ were missing when I compiled these identifications In the two weeks between my initial image search and my writing of these notes, either Google had changed its search algorithms, or some websites had been taken down Those ›unidentified‹ images could have been searched, but I wanted to remain faithful to the initial image array Here’s my image to answer the editor’s wonderful but impossible request for one image that can emblematize the most interesting properties of images My image is in three parts The originating image is at the top I then dragged it, on the desktop of my computer, into Google’s Image Search box, and Google performed a search for similar images The other images on the page are the second part of the image: they are some of the irst results Google returned: images it thinks are similar to the one I dragged into the search box The third part of this image is invisible: it is the hundreds of other images Google also proposed I put one of those into the bottom row, sticking out, as an emblem of that larger set of images Because I have to be brief, I’ll divide my remarks into four parts First, what the image isn’t It isn’t painting; it isn’t romantic, modernist, or postmodernist; it isn’t scientiic; it isn’t ine art, or popular art, or commercial art; it isn’t a single thing; it isn’t a series of sequence of things (because it changes each time you search); it isn’t beautiful (at least, I don’t think so); it isn’t sublime; it isn’t picturesque; it doesn’t seem to exemplify any theories of semiotics, phenomenology, psychoanalysis, or gender theory; it doesn’t speak to art history, art criticism, or Bildwissenschaft (except in the trivial sense that Google images are samples of mass culture); it isn’t politically or socially signiicant; it isn’t clearly meaningful or expressive or affective or pathos-laden or full of tokens of love, memory, culture, loss, time, or space Briely, these are the images: the initial image is a relection interference contrast image (RICM) image of a slime mold, with an interpretive caption (Why I was searching for this is another question, for another essay.) First row, left to right: spread of an invasive species called the tropical soda apple in a test site in Florida; a four-storey icicle that from room 419, Baker House, on the MIT campus, c 1960; a US Army Ballistic Research laboratory, in 1949, which used an ENIAC computer to calculate trajectories; a city discovered on Mars, 20 km southeast of the famous ›face‹ and including a pyramid, a fortress, and a city square Second row, left to right: an unidentiied comedian; seismic data interpreted using the free scientiic software NIH Image; unidentiied bicyclists; synthetic aperture radar (SAR) image of internal waves in the Yellow Sea Third row, left to right: carbon nanotubes and nanowires, from a website summarizing the University of Antwerp’s electron microscopy programs; an image from Bildbericht der Woche, end of March, 1939, from the Calvin College website; the lunar crater Jehan, formerly Euler K, from a wikispace about the Moon; funeral in Iran, from the Iranian website hmc.tums.ac.ir (in Farsi) Fourth row, left to right: a Harappan seal; Fab Five Freddy, Deborah Harry, and Lee Quinones, 1981; a ireback in New England; focused ion beam analysis of intertial electrostatic coninement fusion, University of Wisconsin at Madison Bottom row: British trade union banner, from a blog.1 Third, some remarks on the computer science of image search This part doesn’t interest me, but it’s necessary in order to make sense of some things I would like to say under the inal heading As of 2012, Google’s ›search by image‹ function was one of several image search algorithms on the internet, which are supposed not to rely on the texts that accompany the images, links to the images, or clues in the ilenames of the images Google, TinEye, and other image search engines not disclose their code The reason is not so much what they look for in images as it is how they collect and gather that information and deliver results so 117 quickly In the digital world, images are known quantities: they have set properties, and there are well-known ways of analyzing them Some of the analyses that Google performs on images probably include the following: (1) Google probably applies its face recognition algorithms, in case the user wants Google to ind other photos of a speciic person (2) Google probably takes a color histogram, because the search results almost always match the dominant colors in the image (in this case, shades of gray) (3) Google probably searches for separate images inside the image, because if you give it a landscape with the McDonald’s logo in the background, it may return an image of the McDonald’s logo (4) Google may be performing what’s called a Fourier transform on the image, which would allow it to look for repeated forms such as stripes or grids (5) Google’s oficial announcement says they perform multiple Gaussian ilters to determine ›points of interest‹ such as contours, shapes, and bright and dark spots But nothing is certain Roger Reed, who helped give me advice on this, says that Google’s oficial announcement about the ›search by image‹ function »carries as much obfuscation as clariication (they [claim they] are using matrices, but applied to the page rank or the image?), and one gets the sense they’re just covering every patent-able base« These things—high pass ilters, matrix analysis, Fourier transformations, Gaussian ilters—are in any book of image analysis They aren’t secret or especially advanced What’s mysterious is the speed of the service, and the ranking (the choice of which image to put irst) For my purposes, I am content to know that the search is carried out using non-narrative, non-iconographic criteria: the Google computers don’t identify people by name, even famous people; they don’t identify the Eiffel tower by name; they don’t recognize the Nativity scene The computers use only the rudiments of visual perception—and not even human perception, but the crude toolbox of digital image analysis And then they rank the results, mysteriously And this brings me to my point of interest Sequences of images, in art history, are always meaningful They have a telos, or at least a theme, an agenda, a coherence, a direction, a purpose Georges Didi-Huberman’s chains of images are linked through his interest in the fragility and unpredictability of the failures of representation; Aby Warburg’s are linked by the unexpected recurrences of the Pathosformel in different cultural contexts; Erwin Panofsky’s are linked by shared textual citations The sequences of images that comprise elementary education in art history are linked by what E H Gombrich called The Story of Art The apparently disparate images in the textbook Art Since 1900 are linked by the authors’ common interest in modernism and its driving concepts: negativity, ambiguity, complexity, indexicality, and political resistance André Breton’s images were linked by the surrealists’ paradoxical insistence on the unconscious, and the tropes of surprise, juxtaposition, condensation, and transference Oulipo’s image sequences, if any existed, would be bound by ostensively random self-imposed rules, which would in turn produce surrealistic awkwardnesses and surprises Series of images after minimalism and support-surface are linked by interests in iteration and mechanical reproduction The images in Fluxus, installation, post-pop, and process art sequences, such as Gerhard Richter’s Atlas or Zoe Leonard’s photo projects, are linked by aggregative principles that seek to avoid closure and linear meanings, while also proposing themselves as projects that can be thought about as wholes I could go on But I only want to say the sequence of images here is not any kind anyone has seen before It has no immediate human meaning I would like to propose this as a new relation between image and history It produces effects that are not known in art history: in fact what seems surreal (to see Hitler, suddenly, between microcrystals) is not surrealist at all, and what seems 118 logical (to see several microscopic images, perhaps because of their similar contrast ratios) is neither interesting nor suficient as an explanation This is neither a new Mnemosyne (as in Warburg’s project) nor a new taxonomy (like Marion Müller’s project at the Jacobs University) There are supericially similar projects to analyze large collections of images, most notably Lev Manovich’s open source plugins to the software called ImageJ which permit large arrays of images to be sorted and graphed But Manovich and his lab are interested in producing meanings for art history, visual communications, and art criticism: they hope, for example, to be able to point out new features of the development of Mondrian’s work by presenting arrays of his work Here I am interested in the fact that no such meaning is on the horizon Whatever Google’s algorithms are, and however page rank is involved, this constellation of images (I’m taking this word ›constellation‹ from Mallarmé, Un coup de dés) has no immediate meaning, use, affect, instrumentalization, historical precedent, or other nameable source of coherence That is precisely its fascination 119 ... histogram, because the search results almost always match the dominant colors in the image (in this case, shades of gray) (3) Google probably searches for separate images inside the image, because... that Google? ??s oficial announcement about the ? ?search by image? ?? function »carries as much obfuscation as clariication (they [claim they] are using matrices, but applied to the page rank or the image? ),...quickly In the digital world, images are known quantities: they have set properties, and there are well-known ways of analyzing them Some of the analyses that Google performs on images probably