Empty Apartments: A Curatorial Project Using Crowdsourced Images and Machine Learning
Q&A With Jeff Thompson and Angeles Cossio
Have you ever searched for an apartment on Craigslist and considered how the space you are thinking of renting has been represented in photographs? Craigslist, along with other apartment/housing rental websites, make public an intriguing, ephemeral archive of images of private interior living spaces. These otherwise utilitarian photographs — used to advertise space — take on different meanings when viewed as single images with unique characteristics or when viewed collectively as an informal record of how people live(d).
“There are photographs that are so perfect and more commercial — so perfect that they are devoid of anything human, even though they are of a human space. They are almost like portraits of capitalism, in a way.” — Jeff Thompson
Artists and curators, Angeles Cossio and Jeff Thompson, enjoy working with crowdsourced material for their experimental curatorial projects and are specifically interested in what images that are often ignored can reveal. The Empty Apartments project was conceived during Cossio and Thompson’s personal search for a new apartment after relocating to New Jersey in 2016. Empty Apartments is an online project comprised of 125,000 photographs extracted from apartment listings on Craigslist on May 20th, 2016. The photographs were scraped from the site using code, sorted manually, and then further organized by visual similarity using an algorithm.
At a time when data is extracted and mined like a precious raw material and machine learning and algorithms are working surreptitiously either against us or in our favour, Cossio and Thompson are using this technology to defy convention. The algorithm, used to sort and organize the data, acts as a third-party contributor to the project, challenging traditional curatorial practice while simultaneously recognizing and revealing visual patterns in a way that a human might not. The work is presented as an interactive, web-based piece, where the viewer creates their own pathways through the exhibition.
After their presentation of Empty Apartments at the RIT Photo History/Photo Future Conference, I had the opportunity to sit and chat with them about their work.
Ingrid: What brought you two together to work on this project?
Angeles: We started a collaborative artist run center in 2010 in Lincoln Nebraska called Drift Station Gallery. With Drift Station, we were really trying to find ways to push the boundaries of curation. We didn’t have a big budget but wanted to try and find ways to exhibit art that was interesting to us, but with limited financial means. So that’s really what pushed us to create these, what I would see as, innovative curatorial projects. We really experimented with format and new media.
Jeff: We are also both artists who have independent practices and were interested in using Drift Station as a means to talk about artwork beyond the things that we were doing in the studio but couldn’t address directly as artists. We used Drift Station to experiment with these larger concepts that we wanted to address somehow.
Ingrid: Are you still working with Drift Station?
Angeles: We’re working *as* Drift Station, but not in that physical space. We went from having that physical space in Nebraska to having no space when we moved to the Northern New Jersey area. We found that we couldn’t have a space anymore. Empty Apartments, evolved from us personally looking for an apartment at that time. We both became very intrigued by the images we were finding, and I was really intrigued by them individually as well. Jeff was more interested in the totality of the images, but we were both interested in them as an archive.
Jeff: We had also both done projects where we had looked online for things. For example, I worked on a project where I was taking screenshots from Law & Order and I gathered 11,000 images from the entirety of the show and Angeles did this webcam project.
Angeles: I did a weather webcam piece where I would take screenshots from weather webcams. There are tens of thousands of webcams that are publicly available. And people put these webcams in their window and they actually act like a mirror. What I did over the course of two years, was I pulled stills from those where you could see the interiors of places as well. We were both already working with these images that are normally not paid attention to and are both interested in what they can reveal.
Ingrid: When you say images that aren’t really paid attention to, what do you mean by that? You also made a very interesting distinction between the more commercial or professionally shot images, and the ones that were just amateur or vernacular ones.
Angeles: I was very interested in the unintentional beauty of them.
Jeff: I think maybe all of those things and also none of those things because they are so not intended to be viewed this way. I think you can take them as individual images and they become something more. There are some that are beautiful and others that are just weird.
Angeles: They are really intriguing to look at.
Jeff: There are also ones that are so perfect and more commercial — so perfect that they are devoid of anything human, even though they are of a human space. They are almost like portraits of capitalism, in a way. Especially here in the New York area where real estate is such an intense thing. And there are those spaces that are really run down and represent transient spaces as well. Craigslist is a weird place because it has all of those different images in one place.
Angeles: It’s also very much about the relationship. Not just the individual photographs, but the relationship that they have with one another.
Ingrid: I’m curious about how the images were selected initially. Was it the algorithm that selected the images? Can you talk a bit about that process?
Angeles: No, that was actually me, manually.
Jeff: It began with a bulk download of everything (scraping) all the images. And then we ran a couple of automated things to try and make the process a bit easier. I wrote some code that would try to look for specific images like floor plans and exteriors, but it didn’t work perfectly. There are a lot of listings that appear twice or multiple times…and this is a total tangent, but it’s really fascinating…when they are re-listed the images get compressed differently and so it is not literally the same image, it appears that way, but is actually slightly different.
Angeles: He tried to narrow it down for me, but then someone actually had to go through and look at the images and take out things like the exteriors. There are ones that are just ads. We also took out things like lobbies and gyms or ones that were more like shared spaces, and just focused on the private interior spaces.
Ingrid: So you culled from your initial bulk download and then how many images did you narrow it down to in the end?
Angeles: 125,000
Ingrid: That’s amazing, so you looked at all those images individually?
Angeles: Yes, three times actually! It took me a long time.
Jeff: Months, it took months!
Angeles: It really suited our different creative practices though, for me it was almost this meditative practice. I mean some days were hard and I was like, am I really going to do this for 8 hours today? But most times it was actually quite pleasurable.
Ingrid: So where did the code or algorithm come into play?
Jeff: In the beginning, we didn’t really know how we wanted to do it, we talked about it being random or having a separate page for each city. We retained all that data, initially, because we didn’t know what we would ultimately want. What is interesting about using an algorithm is you don’t really know what it is going to see, but then you get things like specific colours or sorting by composition. Angeles went through them manually, so you would maybe recognize certain patterns like the bright windows and dark rooms, but to know that there were so many of that kind and have them grouped together, that would only result from the algorithm organizing it. Which to me is delightful and surprising in a way. It organizing it…it’s a collaborator in some ways.
Ingrid: So you are viewing the algorithm or code as sort of a third-party collaborator?
Jeff: Yes, I mean because it’s unpredictable and it actually takes a long time. I was running most of the project on my laptop, so it would sometimes take hours to see the results and you wouldn’t always know what was going to happen. If you tweaked one parameter, then you might have a very similar result, but you could also have something that is wildly different, which I found really enjoyable. You don’t really know what it’s thinking about, I guess that’s what makes it interesting. It can also be frustrating, because my background is in art, so I really don’t know enough of the math to be able to rewrite the entire algorithm to do different things each time.
Ingrid: So how did you get into that side of things, do you have a background in tech or coding at all?
Jeff: Not formally no, my B.F.A. is in painting and my M.F.A. is in sculpture, and now I write code. But to me those tools are exciting because they allow you to do things you can’t do otherwise. I mean for curating things…like for this project we could have gone through them all, made a hundred selections, printed the images and made them into a nice linear show, like in a gallery or museum, but I think what’s more interesting about this is that we can present a different scale of an archive and show different pathways and it’s not linear. In a gallery or museum show, you are often being shown how to view things. You can move through it visually in a different way and I think that’s super interesting.
Ingrid: What you are describing is in contrast to how archival material is usually organized, managed and viewed, as artists what benefit do you see in doing this differently?
Jeff: There are challenges and limitations to both, but linearity is so reductive. Even presenting a number of items that are showable in a physical space is kind of reductive too. Our project is limited in a way too, because it was captured over the course of a month, so it’s not capturing things in a longitudinal way, but it is a bigger picture and you can then find your own narratives through that. Or it could be used as a tool to find narratives. It could be an amazing way of sifting through an entire archive or an entire collection and then thinking about how to curate a more traditional experience from that selection.
Ingrid: I am curious as to whether this process is something you would repeat or use again for another project?
Jeff: Not with Craigslist as a source, no. It would demand a really different format. I wouldn’t want to take a very different data set and then just plug it into the same process.
Angeles: It just happened that I have a background in photography and Jeff has a background in new media, so we had those tools. I think our next project will be something different, maybe we’ll make samurai swords… I think it kind of just evolves. We might work with something algorithmic or use another photo data set. We are both really interested in anonymous, crowdsourced photographs.
Ingrid: I was just thinking about the editorial eye, especially when it comes to photographs and how your project really challenges that tradition. In a sense, you are removing the human, editorial eye when it comes to image sorting or grouping. What do you think about that?
Jeff: Well, I just showed my students that ACM, the Association for Computing and Machinery, is now drafting a new set of ethical guidelines for their members for the first time since 1992. One of the things they are considering is auditability, the idea that you can go back and show the trail of how something happened, which seems like an attempt to deal with this idea of black box algorithms. In a way though, it’s easier to audit an algorithm than a human decision or a human saying I like this piece, it’s important.
Ingrid: What do you think about crowdsourcing in general as a means of democratizing certain practices?
Jeff: We have done a lot of shows that use crowdsourcing. We did a show called mailto: where we created an email address for the show and then anything sent to that email address over the period of a month had to be exhibited and we printed and hung it in the gallery, including throughout the opening. We had a printer there just continually printing things we received.
Angeles: It was like a portal into the internet and I was actually afraid of what might come through.
Jeff: We literally had thousands of pages. I mean entire novels were sent, weird photos, spam, and then during the opening people realized that it was printing in real time, so they would take a picture and then email the address, watch it come out of the printer and then hang it on the wall.
Angeles: At that point we were really trying to experiment with curation or lack thereof. Just the idea of complete openness and we were just curious what would happen if we just opened it up. We had no real control over it, we just created the portal and the space.
Jeff: We then made a PDF catalogue of it, so you could download that whole archive. So someday if someone ever wants to publish a 3000-page book…we also did a show called Bookstore with a similar concept. Anyone could send us a book or zine they had published, and we would have it in the space.
Angeles: I think as far as crowdsourcing goes, the thing that’s most interesting about it, in the way I have approached it, is that I don’t really get to set the parameters, the parameters are already set for me. And I like that giving over of control.
Jeff: Well it really breaks that power of a curator as being this like ivory tower selector. That’s why Bookstore was really cool, because it was like fine art books and then these punk kid zines. It was clearly people’s work that wasn’t being shown in a gallery, so that made it more interesting.
Angeles: There are situations where it (crowdsourcing) can be problematic. I mean with Empty Apartments we are using other people’s material and they are anonymous, but we are not. So, I have been thinking about that, and there was/is some discomfort there because I am sort of taking over ownership of this material now. I mean, I certainly put in the labour. But it is more about the reframing of these photographs and also the information we are extracting from that.
Jeff: It’s not like we are hiding their names, so it’s not like they should have had attribution and we failed to give it. It’s not like Cambridge Analytica where we are going behind some veil of privacy, like anyone can surf Craigslist, there is no login required. It is really just like classifieds in a newspaper. I can see why people could be spooked by the automation of it. I mean we are definitely violating the terms of service of that website. But in a way that I hope is respectful.
Ingrid: In a way, you created a type of record by archiving those images at that time. What do you think about the otherwise ephemeral quality of those images as they get overwritten or forgotten?
Jeff: To capture that moment in time too. I hope this isn’t speaking too much of our own project, but if in 30 years you were to look at it, it would form a different kind of record that no one else is keeping of where people lived in these transient spaces.
Ingrid: I am also curious about some of the visual patterns you identified. How did the use of digital technology allow you to recognize some of those patterns?
Jeff: Yes, I think the three things — similarity, difference, and distance — that we talk about, are the things that really stand out. But when you look at them individually there are those other qualities that come out too. Another interesting pattern is how similar these places really are within American architecture. It was surprising to see these similar motifs appear over and over.
Angeles: I feel though that when you pull back and look at it as a whole, it becomes this interesting, visually gorgeous image, comprised of 125,000 single images.
Jeff: The grid that we have is a bit of a compromise. The grid is not as perfect as the image cloud representation because of how the images are clustered. If you look at the cloud format it’s not only grouped by colour, but because it captures composition and lighting you will get things like darkly lit basement staircases together or wood grain closet doors. It’s putting those together and those are characteristics that you might not really think of. It’s sort of a unique way to explore motifs in décor and architecture too.
Angeles: I think the major theme was those images with the over exposed windows and underexposed rooms, that was really visually dominant.
Jeff: If you wanted to find artistic images on Craigslist our system might not be the best way, but the thing that’s really interesting to me about these tools is that they can actually identify lots of different things that you as a human might not be able to, because it would take forever. The algorithm becomes this intermediary tool where it’s able to do something you can’t do, which is what makes it interesting.
Related Links:
