the field

The Algorithm and The Field

In his 2011 TEDTalk, director Amit Sood introduces Google Art & Culture (GAC) as an online project born of the earnest “desire to make art more accessible…and to supplement the museum experience” for anyone with an internet connection. With a mind already on art publications as a kind of democratising supplement to art viewing, my curiosity about the potential of GAC was instantly piqued. By 2016 the project was updated to include the virtual exploration of spaces, both of institutions and cultural sites around the world. Using the technology that has become simply pedestrian in Google Street View, an interior recording of New York’s Guggenheim Museum means that Sood’s idealistic vision can “imagine…accessibility for a kid in Bombay who’s studying architecture, who hasn’t had a chance to go to The Guggenheim” (“Every Piece”). I am imagining the opening and democratisation of objects in space. I am excited.


I choose to view the GAC Project as a hybrid between a virtual art museum and a vast web-based publication: one which amplifies digitized pieces of material culture through a public interactive portal. Excited as I am, however, GAC raises some important and perplexing questions about the future of sharing, displaying, and looking at art – or indeed, the meaning- and value-making of art entirely. “The Field of Cultural Production,” as coined by sociologist Pierre Bourdieu, consists not only of the artists or artefacts in relation to each other, but also of the players in the field: “the producers of the meaning and value of the work – critics, publishers, gallery directors and the whole set of agents whose combined efforts produce consumers capable of knowing and recognizing the work of art” (qtd. in Squires 55). These agents work to build either market (economic) or cultural (symbolic) capital, depending on their “position” and motivation in the field (56). In traditional publishing, there is a clear and a unified gatekeeper in production: the publishing house, with its acquisitions body of editors and marketers. In the case of GAC, what is the impact of the agents who play a part in the framing and filtering of online content: the GAC editorial team, the partner institutions… the algorithm itself? What happens to the field of cultural production and the symbolic capital of art if a “producer of meaning” is an algorithm? Exploring the complexity of framing and filtering cultural artifacts in GAC may stand as a kind of case study, or early map, of the vast and uncertain field of what digital and online cultural production might become.



In the first place, let’s consider the partnering institution – the museums, galleries, foundations and archives – who must choose to partake in the project and share some or all of their collection for scanning and display. By becoming a part of the “Google Cultural Institute,” a partner is provided access to the patented gigapixel Art Camera to digitize paintings, the Museum View for producing 360º tours, and the Collection Management System to be plugged into an institution’s homepage for creating and showcasing art on a customizable interactive platform. It’s a formidable trade: access and display of hundreds of pieces from a collection in exchange for a world-class digitization, presentation and archiving package. Currently, Google has digitized and collected approximately 6 million art objects in partnership with about a thousand institutions (“Every Piece”). And yet, out of 6 million images, a simple search does not produce Da Vinci’s original Mona Lisa, a substantive body of Cindy Sherman’s oeuvre, or a shred of Bill Reid, to name a random few.


At the time of GAC’s launch in 2011, Tate Media creative director Jane Burton points out that “Google Art risks giving a very skewed image of creative output through time and around the world. At minimum…a large tranche of twentieth-century modernism will be absent because of high reproduction fees or other obstacles” (Proctor 216). As I may have noticed gaps in a North American perspective on what “essential art” should be included but is missing from a virtual world museum, I can only imagine what canonical artists or works might be missing from the contingent representing art coming from China, for example. Furthermore, Sood makes explicit the fact that all of the art “belongs to the amazing museums, archives and foundations that [Google] partners with. None of [it] belongs to Google” (“Every Piece”). This means that those who choose not to participate have chosen to filter their collection out of the publication.


We’re then faced with the issue of completeness: only those artworks which belong to the partner institution’s collection may be documented and shared; if it does not consist of a wide body of an artist’s work (which is often the case for contemporary art, for instance), the rest of an artist or movement database ends up sparse. Additionally, small institutions like artist-run centres don’t own the rights to the artworks they gather for exhibitions, so can’t bring the same weight to Google’s bargaining table. The filtering that happens on the part of member institutions isn’t so much curatorial as it is a question of that institution’s means and model – the breadth or control of an institution’s own collection becomes a measure of its value as a cultural producer.



How about the publishers of GAC itself? Perusing the main page, it is obvious that the wizards behind the Google curtain exercise editorial direction over the content. The front page is decorated with disparate magazine-style digest and stories such as “Surprising Facts: The Last Works of Vincent Van Gogh” or “Fashion in Focus: How to Make a Kimono” (Google Arts & Culture). There are also a variety of “Theme” channels, which group individual story posts by curatorial contributors. Topics within the Themes range from popular Western artistic mediums, trends or icons to lesser-known or international movements. The impression is that the display of well-produced “articles” are desperately trying not to present content from only one particular era or aspect of art: “Tribal Chinese embroidery” sits next to “Inspiring British Women Musicians.” It isn’t difficult to see that GAC is trying to appeal to the broadest possible audience: the entire world. One might even think the project as trying to reframe Bourdieu’s field. The mainstream and the niche are presented on the same platform with the same level of disinterestedness (that is, perceived as being unconcerned with consumption or economic capital) as a way of enticing the viewer with pure cultural capital or taste (Squires 55; Curation 294). Taste can be exercised by the viewer as well. Like so many online pinboards, GAC allows viewers to save works they like and organize them as they please (fig. 1). Suddenly, the field includes the public-as-curator, not public-as-consumer; the dream of accessibility is taken to a new level.


Fig. 1: Selected works from the Google Arts & Culture search results “the field”. Curated by Emma Walter. Screenshot Nov. 25 2017  Source:


As I do not personally possess machine learning (just human learning), my grasp of how the main GAC algorithm works is still forming and rudimentary, but this is what I’ve gathered.  For simplicity’s sake, I will refer to the coding substructure of the GAC as “the algorithm,” but I acknowledge that this is a minimizing view of GAC’s engineering. Regardless, when each artwork is initially added to GAC, it is tagged with “creator”, “medium” and “topic” tags, whereupon it is grouped with other images with matching tags (“Tag Items”). This can be considered a “semi-supervised” learning environment for a training algorithm, as “association rules” (code based on the patterns of relationships between things) will develop on their own and the algorithm will take over organising the body of GAC (Flach 14). The various filters which can aggregate channel content by popularity or by colour, for example, are patterns found through user interaction (click-rankings of images tagged with “Leonardo DaVinci”) or visual recognition (the images tagged in “oil painting” with the highest proportion of red hues) (18).  In action, this grouping seems like an automated and reductive way of filtering and framing a collection of art.  It’s easy to disregard an algorithm’s potential as an “agent in the production of meaning” – algorithms don’t have feelings or taste.


This dismissal is a human, self-aggrandizing idea. Taste is merely the way certain choices define an entity in relation to others: “it’s about separation by making statements that distinguish us from those ‘above’ or ‘below’ us in the perceived hierarchy…Having internalized various elements of taste, we then ‘self-select’ into a certain class” (Curation 294).  With association rules, algorithms study hierarchies taught by user clicks or visual information and produce classifications, just as artefacts in Bourdieu’s field are defined by their relationship to other cultural artefacts via the agent (Squires 55). While the algorithm that controls the GAC database is relatively simple, the technological possibilities linked to it, the Experiments, signal ways in which algorithms can be compelling agents in meaning-making.


Fig.2: The Algorithm and the Field. t-SNE Map Screenshot, Nov. 25 2017. Source:


By 2016 GAC had hosted a series of “Experiments”, collaborations between programmers and artists using machine learning and the GAC database. For instance, digital interactive artists Cyril Diagne, Nicolas Barradeau and Simon Doury used the t-SNE algorithm (originally developed to recognize, pick out, and debug static images) to create t-SNE Map, which clustered the images in the GAC collection by visual recognition only. Clustering is a way in which an algorithm groups data or digital objects without any attached metadata; by assessing similarities between different data sets, it creates association rules on its own, unsupervised (Flach 14). The result is a literal – albeit virtual – cultural field (fig. 2). Zooming in, we can find an Alaskan Mask clustered with a kerosene lamp and an Egyptian Djed-Pillar, and as viewers we can’t help but imagine the dialectic between these things (fig.3). Is this choosing and arranging of art objects any different than what a publisher or curator does? Facetiousness aside, virtual space and algorithmic processing are now inextricable parts of our contemporary culture. As Nancy Proctor writes in support of the GAC, “We need to move beyond false binaries and futile contests between the ‘real thing’ and its online representation” (221).  This holds true in the ways we envision art publishing moving forward – that the digital framing, filtering and amplification of artworks will not only enrich the field of cultural production, but grow it into a vast and beautiful plain, open to all.  


Fig.3: A carved Alaskan mask, kerosene street post lamp, and Egyptian pillar: the t-SNE Dialectic. Screenshot, Nov. 25 2017. Source:





Bhaskar, Michael. The Content Machine. Anthem Press, 2013.


Bhaskar, Michael. Curation: The Power of Selection in a World of Excess. Piatkus, 2016.


Diagne, Cyril, Nicolas Barradeau & Simon Doury. t-SNE Map. Arts & Culture Experiments. Accessed 27 Nov. 2017.


Flach, Peter. Machine Learning: The Art and Science of Algorithims that Make Sense of Data. Cambride Unviersity Press Textbooks, 2012.


Google Cultural Institute. Google. 2015, Accessed 23 Nov. 2017.


Proctor, Nancy. “The Google Art Project: A New Generation of Museums on the Web?” Curator: The Museum Journal, vol. 54, no. 2, Apr. 2011, pp. 215–21. Wiley Online Library, doi:10.1111/j.2151-6952.2011.00083.x.


Sood, Amit. Building a Museum of Museums on the Web. TED Conferences 2011, Accessed 20 Nov. 2017.


Sood, Amit. Every piece of art you’ve ever wanted to see — up close and searchable. TED Conferences 2016, Accessed 20 Nov. 2017.


Squires, Clare. Marketing Literature. Palgrave MacMillan, 2007.


“Tag Items”. Cultural Institute Platform Help. Google. 2017, Accessed 23 Nov. 2017.  


Thompson, Hilary H. “International Law and Its Vision of the Ideal Museum.” Curator: The Museum Journal, vol. 51, no. 1, Jan. 2008, pp. 5–10. Wiley Online Library, doi:10.1111/j.2151-6952.2008.tb00289.x.