Seeing brainwaves

Last week I was in London, where I visited Pierre Huyghe’s exhibition Uumwelt at the Serpentine Gallery. You walk in, and there are flies in the air, flies and a large screen showing images flickering past, fast. The images are generated by a neural network and are reconstructions of images humans have looked at, but that the neural network hasn’t had direct access to – they are generated based on brainwave activity in the human subjects.

The images flicker past in bursts, fast fast fast fast fast slow fast fast fast, again and again, never resting. Jason Farago describes the rhythm as the machine’s “endless frantic attempts to render human thoughts into visual form”, and frantic describes it well, but it’s a nonhuman frantic, a mechanical frantic that doesn’t seem harried. It’s systematic, mechanical, but never resting, never quite sure of itself but trying again and again. I think (though I’m not sure) that this is an artefact of the fMRI scanning or the processing of the neural network  that Huyghe has chosen to retain, rather than something Huyghe has introduced.

Huyghe uses technology from Yukiyasu Kamitani’s lab at Kyoto University. A gif Kamitani posted to Twitter gives a glimpse into how the system uses existing photographs as starting points for figuring out what the fMRI data might mean – the images that flicker by on the right hand side sometimes have background features like grass or a horizon line that is not present in the left image (the image shown to the human). Here is a YouTube version of the gif he tweeted:

The images and even the flickering rhythms of the Kamitani Lab video are really quite close to Huyghe’s Uumwelt. At the exhibition I thought perhaps the artist had added a lot to the images, used filters or altered colours or something, but I think he actually just left the images pretty much as the neural network generated them. Here’s a short video from one of the other large screens in Uumwelt – there were several rooms in the exhibition, each with a large screen and flies. Sections of paint on the walls of the gallery were sanded down to show layers of old paint, leaving large patterns that at first glance looked like mould.

The neural network Kamitani’s lab uses has a training set of images (photographs of owls and tigers and beaches and so on) which have been viewed by humans who were hooked up to fMRI, so the system knows the patterns of brain activity that are associated with each of the training images. Then a human is shown a new image that the system doesn’t already know, and the system tries to figure out what that image looks like by combining features of the images it knows produce similar brain activity. Or to be more precise, “The reconstruction algorithm starts from a random image and iteratively optimize the pixel values so that the DNN [DNN=deep neural network] features of the input image become similar to those decoded from brain activity across multiple DNN layers” (Shen et.al. 2017) Looking at the lab’s video and at Uumwelt, I suspect the neural network has seen a lot of photos of puppy dogs.

I’ve read a few of the Kamitani Lab’s papers, and as far as I’ve seen, they don’t really discuss how they conceive of vision in their research. I mean, what exactly does the brain activity correspond to? Yes, when we look at an image, our brain reacts in ways that deep neural networks can use as data to reconstruct an image that has some similarities with the image we looked at. But when we look at an image, is our brain really reacting to the pixels? Or are we instead imagining a puppy dog or an owl or whatever? I would imagine that if I look at an image of somebody I love my brain activity will be rather different than if I look at an image of somebody I hate. How would Kamitani’s team deal with that? Is that data even visual?

Kamitani’s lab also tried just asking people to imagine an image they had previously been shown. To help them remember the image, they were “asked to relate words and visual images so that they can remember visual images from word cues” (Shen et.al. 2017). As you can see below, it’s pretty hard to tell the difference between a subject’s remembered swan or aeroplane and their remembered swan or aeroplane. I wonder if they were really remembering the image at all, or just thinking of the concept or thing itself.

Figure from a scientific paper.

Figure 4 in Shen, Horikawa, Majima and Kamitani’s pre-print Deep image reconstruction from human brain activity (2017), showing the reconstruction of images that humans imagined.

Uumwelt means “environment” or “world around us” in German, though Huyghe has given it an extra u at the start, in what Farago calls a “stutter” that matches the rhythms of the videos, though I had thought of it as more of a negator, an “un-environment”. Huyghe is known for his environmental art, where elements of the installation work together in an ecosystem, and of course the introduction of flies to Uumwelt is a way of combining the organic with the machine. Sensors detect the movements of the flies, as well as temperature and other data that relates to the movement of humans and flies through the gallery, and this influences the display of images. The docent I spoke with said she hadn’t noticed any difference in the speed or kinds of images displayed, but that the videos seemed to move from screen to screen, or a new set of videos that hadn’t been shown for a while would pop up from time to time. The exact nature of the interaction wasn’t clear. Perhaps the concept is more important than the actuality of it.

The flies apparently are born and die within the gallery, living their short lives entirely within the artwork. They are fed by the people working at the gallery, and appear as happy as flies usually appear, clearly attracted to the light of the videos.

Dead flies are scattered on the floors. They have no agency in this Uumwelt. At least none that affects the machines.

28. November 2018 by Jill
Categories: Digital Art, Machine Vision | Tags: , , , , , , | Leave a comment

Updates on algorithms and society talks

I’ve given a few more versions of the “algorithms and society” talks from this spring. You can still see the videos of those talks, but here are a few links to new material I’ve woven into them:

Social credit in China – this story by the Australian Broadcasting Company paints a vivid picture of what it might be like to live with this system. It’s hard to know exactly what is currently fact and what is conjecture.

Ray Serrato’s Twitter thread about YouTube recommending fake news about Chemnitz,and the New York Times article detailing the issue.

19. September 2018 by Jill
Categories: Algorithmic bias | 2 comments

Generating portraits from DNA: Heather Dewey-Hagborg’s Becoming Chelsea

Did you know you can generate a portrait of a person’s face based on a sample of their DNA? The thing is, despite companies selling this service to the police to help them identify suspects, it’s not really that accurate. That lack of precision is at the heart of Heather Dewey-Hagborg’s work Probably Chelsea, a display of 30 masks showing 30 possible portraits of Chelsea Manning based on a sample of her DNA that she mailed to the artist from prison. The work is showing at Kunsthall 3.14 here in Bergen until the end of September.

Many masks resembling human faces hang from the ceiling in an art gallery.

Continue Reading →

11. September 2018 by Jill
Categories: Digital Art, Machine Vision, Visualise me | 1 comment

My ERC interview: the full story

It seems more and more research funding is awarded in a two-step process, where applicants who make it to the second round are interviewed by the panel before the final decisions are made. I had never done this kind of interview before I went to Brussels last October, and was quite nervous. I must have done OK, because I was awarded the grant, and my ERC Consolidator project, Machine Vision in Everyday Life: Playful Interactions with Visual Technologies in Digital Art, Games, Narratives and Social Media, officially started on August 1! Hooray!  Continue Reading →

22. August 2018 by Jill
Categories: Academia | Leave a comment

The god trick and the idea of infinite, technological vision

When I was at the INDVIL workshop about data visualisation on Lesbos a couple of weeks ago, everybody kept citing Donna Haraway. “It’s the ‘god trick’ again,” they’d say, referring to Haraway’s 1988 paper on Situated Knowledges. In it, she uses vision as a metaphor for the way science has tended to imagine knowledge about the world. Continue Reading →

21. June 2018 by Jill
Categories: Machine Vision | Leave a comment

Skal samfunnet styres av algoritmer? To foredrag og syv bøker

[English summary: info about two recent talks I gave about algorithmic bias in society]

Algoritmer, stordata og maskinlæring får mer og mer å si for samfunnet vårt, og brukes snart i alle samfunnsområder: i skolen, rettsstaten, politiet, helsevesenet og mer. Vi trenger mer kunnskap og offentlig debatt om dette temaet, og jeg har vært glad for å kunne holde to foredrag om det den siste måneden, en lang og en kort – og her kan du se videoene om du vil! Continue Reading →

23. April 2018 by Jill
Categories: Algorithmic bias | Leave a comment

Best Guess for this Image: Brassiere ( The sexist, commercialised gaze of image recognition algorithms.)

Did you know the iPhone will search your photos for brassieres and breasts, but not for shoulders, knees and toes? Or boxers and underpants either for that matter. “Brassiere” seems to be a codeword for cleavage and tits. Continue Reading →

28. March 2018 by Jill
Categories: Machine Vision, Visualise me | 1 comment

My project on machine vision will be funded by the ERC!

Amazing news today: my ERC Consolidator project is going to be funded! This is huge news: it’s a €2 million grant that will allow me to build a research team to work for five years to understand how machine vision affects our everyday understanding of ourselves and our world.

Three images showing examples of machine vision: Vertov's kinoeye, a game that simulates surveillance, Spectacles for Snapchat.

Here is the short summary of what the project will do:

In the last decade, machine vision has become part of the everyday life of ordinary people. Smartphones have advanced image manipulation capabilities, social media use image recognition algorithms to sort and filter visual content, and games, narratives and art increasingly represent and use machine vision techniques such as facial recognition algorithms, eye-tracking and virtual reality.

The ubiquity of machine vision in ordinary peoples’ lives marks a qualitative shift where once theoretical questions are now immediately relevant to the lived experience of ordinary people.

MACHINE VISION will develop a theory of how everyday machine vision affects the way ordinary people understand themselves and their world through 1) analyses of digital art, games and narratives that use machine vision as theme or interface, and 2) ethnographic studies of users of consumer-grade machine vision apps in social media and personal communication. Three main research questions address 1) new kinds of agency and subjectivity; 2) visual data as malleable; 3) values and biases.

MACHINE VISION fills a research gap on the cultural, aesthetic and ethical effects of machine vision. Current research on machine vision is skewed, with extensive computer science research and rapid development and adaptation of new technologies. Cultural research primarily focuses on systemic issues (e.g. surveillance) and professional use (e.g. scientific imaging). Aesthetic theories (e.g. in cinema theory) are valuable but mostly address 20th century technologies. Analyses of current technologies are fragmented and lack a cohesive theory or model.

MACHINE VISION challenges existing research and develops new empirical analyses and a cohesive theory of everyday machine vision. This project is a needed leap in visual aesthetic research. MACHINE VISION will also impact technical R&D on machine vision, enabling the design of technologies that are ethical, just and democratic.

The project is planned to begin in the second half of 2018, and will run until the middle of 2023. I’ll obviously post more as I find out more! For now, here’s a very succinct overview of the project, or you can take a look at this five-page summary of the project, which was part of what I sent the ERC when I applied for the funding.

28. November 2017 by Jill
Categories: Machine Vision | 2 comments

Hand signs on musical.ly = emoji for video

drawings of a musical.ly user using hand signs

You know how we add emoji to texts?  In a face-to-face conversation, we don’t communicate simply with words, we also use facial expressions, tone of voice, gestures and body language, and sometimes touch. Emojis are pictograms that let us express some of these things in a textual medium. I think that as social media are becoming more video-based, we’re going to be seeing new kinds of pictograms that do the same work as emoji do in text, but that will work for video.

I wrote a paper about this that was just published in Social Media and Society, which is an open access journal that has published some really fabulous papers in social media and internet studies. It’s called Hand Signs for Lip-syncing: The Emergence of a Gestural Language on Musical.ly as a Video-Based Equivalent to Emoji. As you might have guessed, it argues that the hand signs lip-syncs on musical.ly use are doing what emoji do for text – but in video.

Musical.ly is super popular with tweens and teens, but for those of you not in the know, here is an example of how the hand signs work on musical.ly.

Musical.ly has become a pretty diverse video-sharing app, but it started as a lip-syncing app, and lip-syncing is still a major part of musical.ly. You record 15 second videos of yourself singing to a tune that you picked from the app’s library. You can add filters and special effects, but you can’t add text or your own voice.

I think the fact that the modalities are limited – you can have video but no voice or text – leads to the development of a pictogram to make up for that limitation. That’s exactly what happened with text-based communication. Emoticons came early, and were standardised as emoji 🙂 after a while.

Hand signs on musical.ly are pretty well defined. Looking at the videos or the tutorials on YouTube you’ll see that there are many that are quite standard. They’re usually made with just one hand, since the camera is held in the other hand, and often camera movements are important too, but more as a dance beat than as a unit of meaning. Here are the hand signs used by one lip-syncer to perform a 15 second sample from the song “Too Good” by Drake and Rihanna. First, she performs the words “I’m way too good to you,” using individual signs for “too”, “good”, “to” and “you”.

drawings of a musical.ly user using hand signs

The next words are “You take my love for granted/I just don’t understand it.” This is harder to translate into signs word for word, so the lip-syncer interprets it in just three signs, pointing to indicate “you”, shaping her fingers into half of a heart for “love”, and pointing to her head for “understand”.

drawings of a musical.ly user using hand signs

Looking at a lot of tutorials on YouTube (I love Nigeria Blessings’ tutorial) and at a lot of individual lip-syncing videos, I came up with a very incomplete list of some common signs used on musical.ly:

In my paper I talk about how these hand signs are similar to the codified gestures used in early oratory and in theatre. These are called chironomics, and there are 17th and 18th century books explaining them in detail. The drawings are fascinating:

I think it’s important to think of the hand signs as performance, and in the theatrical or musical sense, not in the more generalised sense that Goffman used for a metaphor, where all social interaction is “performative”. No, these are literal performances, interpretations of a script for an audience. That’s important, because without realising that, we might think the hand signs are just redundant. After all, they’re just repeating the same things that are said in the lyrics of the song, but using signs. When we think of the signs as part of a performance, though, we realise that they’re an interpretation, not simply a repetition. Each muser uses hand signs slightly differently.

And those hand signs aren’t easy. Just look at Baby Ariel, who is very popular on musical.ly, trying to teach her mother to  lip-sync. Or look at me in my Snapchat Research story trying to explain hand gestures on musical.ly just as I was starting to write the paper that was published this week:

The full paper, which is finally published after two rounds of Revise & Resubmit (it’s way better now) is open access, so free to read for anyone.

Oh, and sweethearts, if you feel like tweeting a link to the paper, it ups my Altmetrics. That makes the paper more visible. How about we all tweet each others papers and we’ll all be famous? ?

27. October 2017 by Jill
Categories: social media | Tags: , , , | Leave a comment

I’m a visiting scholar at MIT this semester

I’m on sabbatical from teaching at the University of Bergen this semester, and am spending the autumn here at MIT. Hooray!

It’s a dream opportunity to get to hang out with so many fascinating scholars. I’m at Comparative Media Studies/Writing, where William Uricchio has done work in algorithmic images that meshes beautifully with my machine vision project plans, and where a lot of the other research is also very relevant to my interests. I love being able to see old friends like Nick Montfort, look forwards to making new friends and catching up with old conference buddies. And just looking at the various event calendars makes me dizzy to think of all the ideas I’ll get to learn about.

Nancy Baym and Tarleton Gillespie at Microsoft Research’s Social Media Collective have also invited me to attend their weekly meetings, and the couple of meetings I’ve been at so far have been really inspiring. On Tuesday I got to hear Ysabel Gerrad speaking about her summer project, where she used Tumblr, Pinterest and Instagram’s recommendation engines to find content about eating disorders that the platforms have ostensibly banned. You can’t search for eating disorder-related hashtags, but there are other ways to find it, and if you look at that kind of content, the platforms offer you more, in quite jarring ways. Nancy tweeted this screenshot from one of Ysabel’s slides – “Ideas you might love” is maybe not the best introduction to the themes listed…

Thinking about ways people work around censorship could clearly be applied to many other groups, both countercultures that we (and I know we is a slippery term) may want to protect and criminals we may want to stop. There are some ethical issues to work out here – but certainly the methodology of using the platform’ recommendation systems to find content is powerful.

Yesterday I dropped by the 4S conference: Society for Social Studies of Science. It’s my first time at one of these conferences, but it’s big, with lots of parallel sessions and lots of people. I could only attend one day, but it’s great to get a taste of it. I snapchatted bits of the sessions I attended if you’re interested.

Going abroad on a sabbatical means dealing with a lot of practical details, and we’ve spent a lot of time just getting things organised. We’re actually living in Providence, which is an hour’s train ride away. Scott is affiliated with Brown, and we thought Providence might be a more livable place to be. It was pretty complicated just getting the kids registered for school – they needed extra vaccinations, since Norway has a different schedule, and they had to have a language test and then they weren’t assigned to the school three blocks from our house but will be bussed to a school across town. School doesn’t even start until September 5, so Scott and I are still taking turns spending time with the kids and doing work. We’re also trying to figure out how to organize child care for the late afternoon and early evening seminars and talks that seem to be standard in the US. Why does so little happen during normal work hours? Or, to be more precise, during the hours of the day when kids are in school? I’m very happy that Microsoft Research at least seems to schedule their meetings for the day time, and a few events at MIT are during the day. I suppose it allows people who are working elsewhere to attend, which is good, but it makes it hard for parents.

I’ll share more of my sabbatical experiences as I get more into the groove here. Do let me know if there are events or people around here that I should know about!

02. September 2017 by Jill
Categories: Uncategorized | 3 comments

← Older posts