I'm. Jenna Burrell I'm an associate professor here at the School of Information at, UC Berkeley and I. Wanted to welcome you all to our second. Speaker. In, our brand. New speaker series. Which. Is part of the I schools, algorithmic. Fairness, and opacity working. Group, which. Is funded by Google and so far we've had guests. From Microsoft, and Intel. Thank. You. And also, I wanted to just say hello to the people on the livestream because we're also live streaming this, talk. Today as we did our previous, speaker. And. The, person we have speaking, to us today is Don Navis dr., Don APHIS who. Is a I have, this I have notes here but I also am just gonna add some things that I know about you because I've known Don for many years she's. A senior research scientist at Intel labs she's. An anthropologist by, training she. Has her PhD from the University of Cambridge right, am i right yes good okay and, she. Has been, she's. An ethnographer, she's been looking at the. Self tracking movement the quantified, self movement and. She. Has a great book which I will I will. Promote here. Called self tracking, which she published. In 2016. And co-authored, with Gina Neff. With, MIT press should read that book it's really good she also has been. The editor of a couple of edited collections, which are also excellent, and. She, had told me that she is Co, chairing the ethnographic praxis. An industry, conference that. For this when is that within 2018. 2018. 2018, and. Yeah. We will be data and ethnography. I think. That will be of great interest, to many students. And other people in the room here and maybe. Some of the people listening in on the live stream so promote. That for you as well. So. Please welcome dr.. Don, nice thank, you. Thank. You so much for coming, out today, it's. It's just really a pleasure to be here at the iSchool. You. Guys just do some really, interesting work and it's it's, just a real pleasure to to. Speak. With you all today. Now. I'm. Gonna begin with a small confession which, is that I did, grow. Up in an anthropology. Department, and the, worst thing about anthropology. Departments is that, they teach people to give talks by having written talks, that. They read and I've been trying for. Many years, to pretend, that I'm not doing this but, I am in fact doing this. I'm. Sorry, but. I'll. Try to kind of get off the texts as much as possible. So. As. Jenna, mentioned, I've been interested in self tracking and. And. Really specifically, the the quantified, self community. For. A few, years now which. Some, of you probably know is a group. Of people who keep track of themselves in some way right, so either, and. Spreadsheets. Or through electronic. Sensing, or, through blood and saliva tests. But, the, thing to know about qsr is that it's. Really not anybody, who's worn a Fitbit right, it's, an actual community, where. People get together in. Person in, person to discuss, what. They decided to collect and, what they've learned from that. And. So. In this talk I want, to use what. I've learned about, those practices. To. To. Think through what data, aggregation, might, mean right, if we think about data, aggregation. As, something that happens beyond kind, of large stockpiles. Of, data that are sort, of controlled by some. Very few actors. Right. So so, there's been lots of work. That. Really. Talks about just how terrifying those, stockpiles, are. And. You. Know the sort of the very real power asymmetry. Is that they create and, we should in fact be too terrified by a lot of what goes on. But. My claim here is going to be that there's more. Than one way to do data and that's, not, the only way. So. So. In the way I'm going to do this is I'm gonna think through. If. We look closer, at data aggregation, not as a stockpile, but as a kind of a process. That. Sort of unfold. Slowly. We. Can think about sort. Of maybe there are multiple, kinds of politics, right there's maybe multiple. Ways of knowing about the world or, in fact inhabiting. The world if you want to go a bit further, and, some. Of those actually bear a relationship. To. What. I think of as ethnography, so. I'll point out where that is. So. So. What I'm working, with here is this idea that the practice, of self tracking. Really calls attention to the ways that data aggregates. That. Are very different from the. Kinds of things you might see in clinical, medical. Research, right which really. Privilege, a kind of bird's-eye. View across. A population. So. You, know for those of us in science and technology studies we'd. Really recognize, those large stockpiles, as a, form, of God trickery, right they make, this view. From everywhere, and nowhere at the same time right so you're sort of responsible. To absolutely. Nobody. You're. Right so here, qsr opens. It has this ethos of participation. And that. Ethos, I think opens up a space for contestation, about, who.
Gets To fight to define problems of the body and, how you would do that. Now. You'll. Also see for some of you who are involved, in, the STS, world, that. Some of my emphasis, on participation, also comes from the social life of methods literature, so. That's work from Evelyn Rupert North. Shamar's. Mike. Savage, mostly. Based in the UK, and. Their. Work suggests, that, methods. Really can't be separated, so easily from, the. Social, lives that they're designed to comprehend, right. That's. You. Know they participate. As well as. Comprehend. And and you, know surveys, are really a good example of this right surveys, where these things that these techniques that were designed to you. Know sort of assess the world from afar and lo. And behold. Lots. Of us do them right they're not just research, instruments. Per se right they participate, and in fact actively, reshape. That. World right, so. So, one, lesson I'm really taking from this literature that if that's the case if methods, moved between. Universities. And practice, and then back again, is, that. Really. We can't it's not a reasonable thing to sort of assume ahead of time that, we know who has a research, method, and who, doesn't. That's. You, know one person's, method, becomes. Another person's everyday practice, and and vice versa right so you see there's you, know lots of movement. Around. So. Here you know in a sense data science methods are being used in. A way that maybe data scientists, themselves might not recognize and maybe clinical, researchers. Also might. Not expect. Right. So my. Title then. Is. This. Notion of n of many ones and. That, really. Greenfields. Work in, a volume i. We. Or she talks about west. As a kind of a para clinical, practice, and. So what she means by that is, she. Says. You know by taking up the tools of medicine, but not necessarily, its claims to expertise, this. Is medicine turned inside-out where. Members can reformulate, the. Epistemic cultures, of. But. With. Some complications. So. Some examples of what this might look like is, you. Know we see a Q s in. Okay. So. Some examples of what a pair of clinical practice looks, like is you. Might take blood sugar data and. Find. The. Presence or absence of diabetes, but the presence or absence of stress right. You might take. Steps. Data and find, not, weight management, but. The. Onset, of autoimmune. Diseases. Right these are real examples of stuff that people have done by, by, taking data in a new direction and.
So. In, qsr folks sometimes talk about para. Clinical practice as a form, of n of one research and. What. They mean is you know the one is the self here, and the. Self is really bounding. The context, for what is to be known. So. You know there's there's also a lot of literature on what is the self and self tracking, and, here, I'm gonna use it as a kind. Of an. Epistemological. Unit. Right is this is the frame that, situates. And mobilizes. Knowledge. So. If I were to be stirring it about it which is a kind of a thread. Within, contemporary. Anthropology. I might. Say that self trackers, cut, the network at, the self. So. You, know strands idea about cutting, the network. Really. Comes out of this problem of how, you handle. The, infinite. Entered set of interdependencies, that make us a certain situation a certain way. So. For example if you think about a patented. Invention. There, might be five people on the patent, there. Might be ten. People on the subsequent paper and, there might be 50 people involved in that reference system and. 50. People beyond that and so on and so forth and then you have the invisible labor usually, done by women who actually sustain this stuff and then you have policymaking. Right you see this is kind of like this biblical. Kind of beginning. Of. Social. Relationships, and. And. So. You, know it's at a certain point you got to cut the network and. Sort, of frame what counts or what doesn't count and. And. What she's saying here and cutting the network is that there are really two implications. Of that right, there's one implication. That's, just about how we know what we know right so you can think about a researcher, who's trying, to understand, what creativity is or. Invention. Right, they're gonna cut the network in a certain way, but, it also has live, effects. In social relations, right, these, are people are making ownership claims as well right. So the, argument here is that we. Can't be too quick to separate. Out those, two things that they're actually really tightly. Coupled. And. And. And. Sort of tied together um. You. Know in that same volume there's. A piece by Judith, Gregory and Jeff. Bowser on. It's. Really argues that you, know there's this interesting. Emphasis. On the self right now in medicine, particularly, in the context of personalized, medicine and. They. Say that it's notable, that that's happening, at precisely, the same time, as. The distinction, between the, self and its environment is perhaps at its least obvious. So, you, know we have now more data about microbiomes. Right the bacteria, in the gut we have more data about expose. Ohm's right the toxins, in the environment that, we live with on a day-to-day basis, and of course in genome. All. Of which connect, us to different to other people other species. Other. Physical, things in new, ways. Right, so so really cutting the network between self and not self it's actually, not, the easiest thing to do. Right. In fact we see all of those things right species. Bacterias. You, know substances. Of various kinds actually, in that self now. So. So, why aren't you that let that at least in western context, and talk a little bit later about other, formulations. Of the self but, at least in Western context, trade the self I think actually still has utility. As a give as a not, as a given, right not as a sort of unit. That we just sort of assume away on our way to Co hearing a thing we call a population. But. Kind of worldly good like a heuristic. Right. A heuristic, that does it to some illogical, work and, also moral work in this sense that it's. The place from which, Europeans. And Americans are, making claims and, rights and data right we've one of the ways we think about the self is that it has a boundary at the skin right and, data. Is made from bodies right and that gives us some claim, on it um. Right. So so then returning back to this end of one thing. You. Know Greenfield is arguing, that that n of one char you know challenges. Both traditional. Evidence-based, medicine which, has a real emphasis on. Sort. Of large randomized. Controlled, trials. But. It also has another relationship. With personalized. Medicine, right which which is really trying to, customize. Treatments, according to microbiome, and genome and all the rest of it, and. And, it says. That nf1 is actually really different from that stuff right.
That. Both, of them in a way traffic in really rich aggregations. About the individual, but precision. Medicine is really, knows what it knows always through this generality. Right it's always making, an end of a billion before. It gets to a satisfying, notion, of what, a one is, so. Here's, here's one example um this, is a recent. NIH, project. That. You. Know is aiming to get biomarkers. And. Data because, of course, shows. Up everywhere. Anyway. They're aiming to get a bunch of data from, about a million people and. Right. So we no longer have a sample, here right we have you know the rhetoric is all of us not. Not. A sample but a whole universe. And. And. Each, person, in a way sort of emerges, after I, was different right from that kind of like mass commit. You have a mass commiseration. Before you have difference. And. So you know returning to Q s then you know a Greenfield. Says that. Almost. In passing actually she says that and if anyone experiments, might actually be better thought about not, as n of a billion but or. If sorry if we're gonna crowdsource, and of what experiments, now. That. That's not n of a billion but actually, n of a billion ones right, where each of those ones each of those people, has. A coherence. Of a kind but only partially related to others right so we're no longer, assuming. A kind of a commensurate, right, atom at a mass scale. So. This was kind of a shorter, line and I just thought, this was a super, useful idea, and, I wanted to take it further. In, the sense that I've been my own work has been all about making those and want, n of many ones. Sort. Of aggregations. Right so I don't work in billions because, I don't, know how to do that. But. I handle, many and. I've, been doing that with. A technical, team through, a software, project called. Datasets where. We've been really, trying to figure out kind of what kinds of aggregations. Either. Aggregation. Within one person right because there's still a lot of data there, how. Did you that. And. Slash. Across many, people, right like what is that relationship, between the, one and the many here. If you're doing n of one or para clinical, research. And. I've also learned a little bit about the issue just by participating, in and. Observing. Some of the discussions, that have been taking place between the. Self-tracking, community, and in public health research world and there been a sort of a series of symposia that. Kind of trying to bring those two worlds together. And. Sort of looking at the various projects going on you know I've kind of come to to suspect that there's a logic, right between two, about. How like which projects, kind of wit data, aggregations, efforts kind of really. Kind of you, know make some progress which, ones stall out which. Ones are kind of hard to do but still worthwhile to do I mean, which ones are kind of like flat-out you know kind of a dead end. So. I've, been doing some software development. With. A team personally. But. You know my method for doing the software development is not straightforward, user, research, it's. It's really trying to situate myself in, data's dimensions. Right so. Thinking you know really. Being inside, it's temporalities. It's it's spaciality, Zitz chain of associations. And. Doing that really like quite literally with the people to whom it refers. So. You. Know these these, pathways then these you know sitting through these pathways as you, know people themselves are kind of walking them through is. Really what's getting my me my sense of many here right so it so I kind of scoped many as a, what. You might get your head around ethnographically. Which. You. Is, contested. In and of itself for those of you participated. In the Anthropology. Debates right, you know how hard that is.
But. You, know the idea here, is is kind of walking. Walking. Through data with people, as opposed to trying, to build for myself a whole universe, of the stuff and then sort of querying, it. So. You, know the phrase I write the, term I used to talk about this is kind of lingering, and data sort, of hanging, out there in a way and. I want to share some contrasting, images for, for what that might mean so. The first one is the image that I want us to get away from but it's the one that always haunts. Quantified. Self which. Is something like this right, so. You've got kind of a you. Know white, rich. Very. Thin woman. Who. Apparently, has a heart rate of a hundred and forty and we're, supposed, to take that to mean that she's working really hard right, never. Mind that you know somebody with a heart rate of 140, sitting on a yoga mat is like not that. Works. But. That's, you're not supposed to think hard about this number right. There's. Nothing here, to kind of linger in it's, sort of this like it's epistemological, dead, zone right. And. So we think there's a better image and. I'm gonna pull. That from, a sculpture, actually, a data. Artist, named, Steven Cartwright. Who. Carved, into this cube. A. Topographical. Map in. A way so this is a map it's it's sort of you know carved into acrylic. And. And. So it looks like a landmass, but it doesn't correspond, to a landmass, and the highs and lows are, defined. By. How, often he or how much he's driving. So. When. I Saint linger, in data right what I'm really asking is, like what, is it like to sit at the base of that hill, just like just after that dip right I'm, just before the rise, you. Know what does it like to live with the memories, that sit, in those places -, it's a sort of borrow from Keith basses work. How, does being there, connect. Someone, like Steven. Cartwright to other bodies, to other environments. And. So forth. So. The rest of the talk is sort of roughly like this I linger. And some data for a little bit and. Then talk about those efforts at aggregating, across people and kind of where, that got hard and and where we made some headway. So, the first kind. Of lingering I'm gonna do is with a. Woman named and right oh. She's. A Q of Ember she's also a former, NASA, robot, assist she actually worked on the Mars rovers, right so there's certain, amount of privilege that does come with Q s even, though the. Argument, here is that it it, also articulates. A view from below in a particular, way. And. So. This is a from, a video of them that she did with the community, and.
And. Her, so she had this quite serious medical issue, she. Found. Herself in the position, of having a disease that, you have to fight to get which, is a term, I'm lifting from Joe doom it's work to. You know to sort of you know there's just these classes, of diseases like chronic fatigue for, example, where. There are no clinical categories, that are gonna help you anything. They just don't exist we. Just don't have categories, for everything in life. And, so. She, received, the kind of diagnosis, that's you, know the doctor said yeah you do have a real problem like we do actually believe you but. We have no idea what's. Going on or how to solve it that's what she's alright well um. You. Know so she turns to data and. Tries. All, these different things right, so she tries a combination, of tracking what she's eating, how. She's sleeping, right a whole lineup different activities. Like you name it she's. Tried it and and she what she's trying to do here is to debug, the problem, as, she puts it right so, so, here we got our first flavour of aggregation, where you. Know that, it turns out that like sensing, and recording. Across multiple modalities is, actually, really important, right it's not good enough to, just put the Fitbit on and be. Done right and so that's what she's doing here. And. Roughly how she works is she she, you, know chooses. What data to collect, just. Makes a guess at first and, then and that of and, that evolves um as, she's develops, a kind of a loose sense for what's going on right, and so then she changes some stuff up and adjust, the data as well as some practices, just to see what's happening right. You. Know in particular to the body, you know the body sensation, says she's feeling right, so you see this iteration between, what data exists. And what's, going on in the body right which is not, that different photography. Actually and, for those of you who do it um, but. But she thinks about that in terms of debugging and I and I actually do think that that's significant. In. That you, know as she's talking about you know when. She was working on the Rovers. You. Know there's really no instructions, for how to fix a rover on Mars. How. To do that yet and, so you, know she says you've got to develop an instinct, for you. Know how to how to fix stuff and, that's, what she's doing here, right so so, she's interestingly not. Using, her, computer science, skills or she isn't a particular, way right what she's not doing is. Probability. Stuff, right, or, a sophisticated, machine learning. Which. She's using are these you know techniques, of observation. That I think actually a lot of us are really capable of doing, but. We don't think to do it because we're not thinking in terms of, debugging, right so that's where you, know again the privilege comes in here. So. You know one of her techniques which i think is pretty telling of this is that um, she. Actually heard, the diet tracking, she took a picture of everything she ate instead. Of writing. Down all the macro and micronutrients, I mean, for the reason she did that was. Because. Even though you know writing all the numbers down leaves. It available for math, and, commiseration. All that business but, the pictures can actually be interpreted, in more flexible, ways, so. You, don't know what you're looking for right you do the coarsest possible, thing first and and. So that's kind of what, the debugging instinct, kind of gets you. And. She could do that because she didn't have to be commensurate with anybody else you know like her problem was exactly because she wasn't commensurate. Right so she has this you. Know peculiar. Relationship but not peculiar sexually a lot of people do end up having to do this more than you would think. And. So and, so she works stuff out right so. The the the. Punchline. Here is that it. Turns out that, vegetables. In the nightshade, family we're, giving, her this debilitating, thing. I mean it was actually it was properly. Debilitating. Right but. The solution, was, like totally, not medical, writers, just don't don't, eat potatoes, and. Eggplants. And stuff like that. So. Afraid. So what she's doing here she's collecting stuff about symptoms and and. Potential. Causes you, could see sort of elements, of. You. Know what we fantasize, about us the scientific method right but. A lot of the time you know the data isn't even that explicit, right so in another story, um comes, from somebody I met through. Her. Name. John Wright. Where. So. John was having difficulty, sleeping and.
So. He turned to this consumer-grade sleep device it was one of the super. Early ones, but. Sort of did sleep states, and. And. So. They went through this really kind of elaborate process of churning through. What. That data could be telling right, so first it. Was looking at the you know the start time of sleep like when you lay down in bed or, maybe there's something to do with with, sleep duration. Or. You know maybe, there's. You. Know some. Other thing happening and then finally. He comes to this point where, they. Kind of literally. Stacked the night, grafts on one another and found a recurrence, we, are at, precisely, 3:00 a.m. every day he, was partially, waking up, which. Is a weird thing and so. So. He started thinking about okay well what's in the room and. It turns out you. Know machines, are actually that regular even though bodies aren't and, his. His. Computer, was in the room and that computer was set to back itself up at 3:00 a.m. and so, it started whirring and fly-in flashing lights and so forth and that's was. Contributing, pretty, significantly, to, the sleep issue, um. Rates. Are for people like John. Right. What's going on is he's, he's looking for clues and things that lie just beyond. The data itself, so. Not sort. Of he's, not believing, that that, the whole set of answers is in is explicit. In the data right, and. And, the way I'm going to think about that is is to think with this concept of a hinge. Which. Can. Be a temporal, aggregation, like he's doing here or a spatial, aggregation. But. Something that sparks a memory, of something, right. So a place. In the data where there's a connection, to the world. From. Which the data came. And. That can actually can happen a lot of ways you can get fancy anthropologically. About hat like is that semiotics is, it you, know materiality, and visuals, and all the rest of it I'm not making a claim about what specifically, is happening here right but just just, working, that there's something there and it's not in every single part of every single data set that. You're. You know it might be the range of highs and lows it. Might be the you know location. Or. An average of this month versus that, but. What's happening here is is there something that's that's bringing the person out of that, sensory, world of a visualization. And. You, know number. And. Into, memory, and. And. Churning, through each of those right which is really what he's doing is a kind of a churning. Is. Really I think about following those hinges back. Into. A context. Where you know that that that is being evoked right always partially. And. One way that Ann talks about this is she says you know what makes a good self trackers, not actually, the ability to do math it's the ability to think of multiple, time scales. So. I think actually temporalities. Are really, rich with, hinges, right, you, had in this case a kind of an interaction. Between circadian, rhythms, and machine rhythms, right. You. Know you have a you. Know if you're if you're looking for. You know a pattern to do with whether you're, not gonna see it come. Up every Thursday, right because that's a human workweek right segment you can see social patterns and how social patterns intersect with bodily patterns, if, something does come up every Thursday, right so you've got all this like richness, here and you've. Got that widths you can do the same move with space as well right you can do a whole kind, of rhythm, analysis, on this stuff. And. You, know anthropologically, I kind of get excited about doing. So. So that's kind of what, I think it's. A little bit like to kind of be in those places in those data sets with people um, and, now, we're gonna sort of think a little bit about cross, person, aggregation. Right what does that actually mean. So. In this in, this kind of public health, self-tracking. Intersection. World. We. Have I, think right now the kind of the dominant, notion is really one of data donation. So. Right so this is this is the idea that medical, discoveries, are really more likely to be made, if, somebody, you know donates, their data to medical research right so no longer tissues but data, and. And, so you've got this that's, also social relationship, right a relationship. Between the one and the many that's. Really one of stranger, hood where. I give you your my data and you might make some knowledge from it and maybe or maybe not there's some karmic, returns. And. You. Know I I you, know that style a bit of aggregation, absolutely, has its uses in life, but. Stranger, hood here you know it cuts the network in particular, ways, and. In, a way that I think necessarily. Erases. From meaning, making, the. The cycles. Of computer. Backup that John had right, that. There's no epidemiologist. In the world I, was, going to look for them because. That's not an epidemic. But. Actually. There could kind of be an epidemic here but of, sort of sleep induced, you know machine and do sleep disturbance, but, you're not gonna see it as a series of 3 a.m. spikes, what.
You Really got to do is you got to know what John knows and. And other people like him where you got to you got to follow the hinges with people. So, you, know so we have been seeing these you know a few, technical infrastructure, for sort of supporting. A data donation, model right and that sort of moves data around in very particular ways. But. We started in the lab started asking okay well you, know what. What, what kind of a technical infrastructure, actually, would take that 3 a.m. backup, seriously, right what would that actually do. And. You. Know is there even a role for a mutual learning there, right if you've got someone like John with. His data alongside other, people. So. You. Know. Roughly. You know it to. Do that and actually I am, looking at my slides and realizing I had missed some apologize. I. Missed. Um you know this was the idea that, finding. A hinge in this in a spreadsheet stinks. You're. Probably not gonna do it but, that's a kind of a visualization, that's. Not John's but kind of might as well be John's where. You can kind of spot a recurrence pretty, easily, the, other thing to mention here is that I think this is like following, the hinges is like the inverse, of Tufte, visualizations, right, like where Tufte visualizations are really trying to like tell a story, in a very carefully, crafted way, but. If you're finding this story that. Crafting, isn't gonna help you is only gonna narrow what you can follow and, what you can't so there's kind. Of this tension here about like what supports, a hinge and what doesn't. Which. Is not to say that you know we started data sense on the grounds that we thought okay, like because this is where people are learning there were learning in and of one and themselves. That's, we really need to think about intra, personal aggregation. First. That. You. Know people are learning in that way like. Yeah right so if you think back to what Ann has done she, polled. Heterogeneous. Data that weren't just from apps so. You know she's marking up things by hand um you. Know things that actually you don't just build an API for which, is how most of the tech world works. And, there's also of course there's a tricky time, stamp details, to deal with it um few. People really have an appetite to deal with so, we figured okay let's just handle a bunch of that stuff and, so this. Is a gonna. Give you a couple screenshots from data sense now this is our kind of our, you, know file import, where we do expect, a certain layout and, we can only do some things and not other things but you know it's the idea that, you. Know we can take that stuff alongside you. Know stuff that comes from where standard apps. Alright, so we said that okay like if is if exploring, data is actually really about churning. Through the hinges. Then. We. Try to figure out which, intrapersonal. Aggregations. And self trackers, tended. To use right, like where do you tend to find some luck and where, do you not. Knowing. Of course it you're not gonna get it right for everybody but the idea is to help that happen faster, so if it's not in our lineup you, can kind of go on to something else. And. That. Um. So there's a set of visual tools here there's a set of kind of driver trying to drop filtering, tools that I won't bore you with. But. The thing that I learned that was kind of surprising, to me or not surprising, really but it, turns out that that. Too. If you're not a nasa roboticist. That. Churning, through these hinges sometimes requires explanation. Right, so. You know for example if you're transforming, data from, you. Know once, a minute to once a day aggregation. You know the idea that you need an aggregation, function, to do that right and then you have to make choices about, what that is that account is that an average, so. All of this sort of stuff, is. Not, obvious, and. So we we did a series. Of, tools. To kind of just. Show that um so this here is, the. Kind of the drag-and-drop filter, area, so this is the idea that if you want to do some data transformations. Um before. You try to visualize it um here's, where you would do it um I. Just wanted to give you a sense for what that is. These. Explanations. Are we actually went a video route, um and, I'm blatantly, just showing you a video to like not talk for two seconds.
But. Here so the, video I'm about to show you is. For. This tile so you put your mood your data in and then you put this tile in if what you want to do is assign. Some. Values, for your missing data so if you want to create data in the, places where you have holes in data which. Again sounds obvious but, it. Turns out you got to make choices about like well what sample rate do you want to assume and you, want to take the average or some random number and sometimes you do want to take some random number because if you're, trying to find you, know the the patterns and that you don't want it to look like the numbers that you already have right. Anyway. So here, is our video. Okay. There's no, iterate. I think. It's I think, that's all we have to say. Yeah. I had. Fun making them which is. All. Right so so we've got these, tools in, place, for, the most part with bags and all the rest of it. But. You. Know as that part started to solidify we. Started to ask about this question again of what is interpersonal, aggregation. What does that do here. And. One of the things we had noticed in the meanings that people were going to was, that. It, was actually pretty useful to, have other people's date examples, of other people's data as kind of context, to think with and. And. And, an, example of that comes. From my own group in Portland, Oregon we're. We're. Having this discussion discussion about sleep quality sensors. Or. Quality, scores sorry so there's like you know bunch of apps that will give you, sometimes. I have sensors usually, have sensors or. Use. The phone as a sensing device and, they'll say okay you know we're gonna have we have our magical algorithm, that we're not gonna tell you about and, we'll. Give you a sleep quality score, and. So. Some people want to know like what is that what. Does that mean. And so somebody in that group is complaining, that he. Can't he, thinks there's something wrong with the algorithm because, he. Can't possibly get below 98%. Sleep, quality so clearly it's the things broken and. Another. Gentleman kind, of raised his hand in that room and said you, know I use, that same exact one and I'm getting 70, all the time so. Clearly. It's you know the, algorithm is not that not as bad as it looks. And. Ok, that's, interesting right that's that's interesting context. And. So what we thought was well. Maybe it might be useful that. To, kind, of you. Know try to do some of that in an online way, so. Their. Strategy was to make. It possible to opt in to anonymous. Aggregations. For. Whatever data you've uploaded so. The, simple, version this is a this. Isn't working anymore by the way but.
The You know this is like the simplest, possible version. Of this right so you have like a tiny. Little data, pipeline, that says take, the average of everybody steps and then. Spit it out and, it kind of puts it on a histogram, that histogram looks weird because it's test data but, the idea is you've got a histogram and then you've got you, know wherever you. Are. And. And. And so you know we, had you know a set of interesting rules around this right so you can't query the thing if you yourself don't have data right you've got to participate in, order to come in, and. You. Know in the spirit of kind. Of trying to generate algorithmic, transparency, because that discussion, had just started to heat up we, said ok you know a you. Can get a heck of a lot fancier than that and so, the thing to do is to, if. You. Know somebody's querying a pool of data, to. Not only send the the. Results back in the data but also how you got there so. That other people could, also build on that and tweak it and make their own kind. Of baby, algorithms, race isn't that machine learning okay, but, instead of tools but you know that people you know hopefully could sort of learn, from each other about how to ask questions about a pool of data. And. Of course to do that you need like a ton of different privacy. Safeguards and, all the rest of it but. The point is we're trying to mimic those those triangulations. And. That. Ended. Up having two problems that I think are pretty telling. So. First, we. Really. Quickly ran into the issue of commencer, ability. Which. Is you. Know if you already. Know like NIH already, knows what data it's it's getting if you already are just working with Fitbit data. That's. Close. Enough. That's, almost, the same thing it turns out that there are nuances in, Fitbit that means you're not actually looking at the same thing even if you're looking at Fitbit data but, it's the, same enough where you can start to query that algorithmically. If. What, you've got is a data pool about sleep. Quality, you've. Got ten people you, know measuring. That along ten different lines if you've got a sleep quality score, and then you've got sleep duration, and you've got sleep start times and, or. Qualitative. Like I felt better after I you, know I felt rested or not right there's a meaningful signal to, and. There's one which is my favorite example this is a real thing, someone. Had counted, the words that, they. Remembered, about their, dreams when, they woke up this, like a sleep quality measurement. It's. Great so if you're trying to fuse all that together you're not gonna have a whole lot of luck. So. We. Thought you know okay part, of that ethos of really taking the you know the 3m backup you, know in a way I'm, part of taking that seriously is, recognizing. That this variety actually. Isn't noise. That. That actually what's happening here is they're they're emerging, from very different social worlds, um it, really shouldn't, be imagined, to be commensurate. In the first place right. So you could imagine a clinical approach would be oh okay well just do some transformations on it and then make an approximation and, then we're all the same right I mean people do this.
But. We're trying not to do that and so, when. We hit that wall we said okay, maybe. It's just better to kind of pass data, visualizations. Along, you. Know individually. With people you actually know. You. Know in this sense that you know Anne and John were working together people, do work together right. So we built that mechanism this is a how, you manage, your a, data, source as we call it and data sense and you sort of have you, know who can use who cannot use, can. You put it in aggregation, or not. Right. So so. That was kind of one, solution, to that the. Second problem was kind of more meaningful to me in the sense that, you. Know we, had this question of who, in the world is actually going to query data like, this, we. Had at, this one beta tester, who told me you, know he basically said, look I don't know who these people are and why. I should care about them okay. Fair enough. Good. Point. And. That again cut gets back to this issue of like how you're cutting the network it's. A one response I think a clinical, response might be okay like let's build in you. Know a set of demographic. Information and. You, know location, information so you could start to make these cuts by weather like do people take a lot of steps during rainy, times or whatever it is. But. That kind of felt to me like we're getting back to stranger, hood again and. That we were true that would be an attempt to try to make relations, that didn't, actually exist. And. And. So that. Was a little too close to date or donation. And. So we kind of hit a little. Wall. There. But. I wasn't fully, deterred, in the sense that you. Know we know from STS that there's no such thing as data. Outside social. Relations, right, that. You know we don't have cells, that are these sealed off voids you. Know we know you. Know for example from Nick Merrill's work who's in the room that you know people have ideas about what sensors, actually, mean and an act in response to them right we know. From the quantified self stuff, that people are learning about like, what aggregations, you think to make right. Art is an entirely cultural. As well. So, so all of that stuff too was really weighing on my mind and. We. Had this major redesign, kind of when we were thinking through this I'm just, sort of at the front end and and so it was my job to do the user testing, and so you, know I said look you, know the thing to, do here is not like. Just sort of straight up you know one on one user testing but actually collaborate, with a group who, had, a reason to know about, the. Interpersonal, aggregation. Right that they know what they're looking at with that pool and have an agenda for it. So. That ended, up becoming in a way it's its own user test is a research, project it, was called the real-time health monitoring, project. Which, is a collaboration, with. A group called the fair tech collect collective. So. That's run from an STS scholar I mean when Ottinger at. Drexel, and we had some Carnegie Mellon folks and. We also had, a group. Of environmental, justice people, so. These are people who are their, community. Members alright so they live locally. Or right next to a. Very. It's, in the US and they live right next to some very serious. Facilities. That that emit on very serious air toxins right some. Carcinogens. Specifically. And it gives you asthma and it's it's some enough. To have inspired on some. High-grade. Air, quality monitoring, right so data kind of once, every five minutes for a good you. Know 10 to 20 pollutants. Right. So. So. We got together and. You. Know in the participatory, you, know design slash, can be based participatory health. Tradition, you know we sort of set a research agenda together. And, that was you, know both about how what kind of data we collect and. You. Know how we are, you gonna sort of you know think about the results and. We set up just a really small pilot it was nine volunteers. And. We kind of kitted them up with a bunch of wearables. A bunch, of apps and, the. Idea was to try to correlate that health, stuff with what's, going on in the air. Now. It very much was a pilot, in the sense that you know you know we know you know from qss we sort of know that apps work and devices work about half the time and then half the time after that you, get meaning, he's. So figuring out where that where that meaning was gonna come from was really the project.
And. And. So what we did was I, which. Is a technique I've I've been working with for a while is I just I just literally sat down with people and walked. Through. Data with them both first, individuals, and then we all came together, as a group so. This is a kind of a screenshot from that so I just trained the camera on. The. Data as we were working with it and, in. This interview in particular, you know so we find all this interesting stuff like this woman. You. Know, we. Were looking, at her blood oxygen, data and you could almost tell, a little bit that it kind of goes it spikes up towards the end of the sensing period and she, said oh yeah that's because I you, know we're traveling for a week so I left the area. It. Kind of shot. Up and. You. Know so she had this sense that that was going on but this kind of framed it in in, a different way you know another example from that work was. You. Know we had this I had this discussion with another participant about. Heart rate and it's spiking at a certain time of day, and, we were trying to work out you know is that because the pollution also does that and and and factories, do actually, have their, temporal, cycles or. Is it because of her that's this television, show that she really likes it's like infuriates, her, she's. Like, you. Know but, you can have these really like, critical. You, know these, critical conversations, about what's actually going on. But. The real interest, here, isn't. Individual. Data for them right they want to know like, collectively. Are we do, we are, we suffering from physically. From air pollution um. And as you just saw a data sense is really not optimized, for that, and, so, I did. A little hack. Which. Was we have this feature where. You can sort of throw in. Lots. Of like, a lots. Of different data types right, so you know that you, know it doesn't matter what the data source is just put it in and we'll just sort of churn a bunch of correlations, for you then put them on a grid so it's just a little bit like um you, know those old Atlas distance. Maps we, kind of go up one, and then over and then that's your correlation, but. We had that data exchange, so what we did was. Kind. Of just put it all in one account, and. Then put. Everybody's, in one account and then so, this is just taking heart rates and, correlating. It with one substance, right so we put one pollutant, along the side and then you can kind of just sort of look down the side to see all right are those correlations, kind of matching up or not right this is totally, eyeballing. The real data scientists, who no doubt hate this. But. Just kind of eyeballing it right. But. What, that accident. Did was something actually really interesting, it's. It. Enabled, a conversation. About a really complex data set.
In. The room right so we could because there anybody who's kind of spread out you. Could start asking questions like, well you, know that. Person lives closer to the factory, so, do you think is it about distance, from the factory, or, somebody, said look you know we live in a toxic soup. We're. There's lots of chemicals in the air all at once so, really maybe let's, not do substance, a with one pollutant, at a time but you know actually try to fuse those together, and. See what you come up with that, right it's like okay, like we're having a conversation. Here. And anything, there's something important about that right I think there's something important, to, the process. We're participants. Themselves are asking, these questions. Right. After. Just having seen themselves as, one of those boxes. Rather, they're sort of in a way becoming visible, as a. One in, the many and. I think that was meaningful for the discussion, right there's. A million different question, you could ask of this data set but. These were the ones that were emerging, not. Other questions, on. This, day right through this engagement um, and. That conversation kind of made sense too because there was a shared purpose because. There's people who know each other's stories well right they know who's, traveling, who's not who's getting sick and who's, not but. Also because it creates obligations. Right there was a researcher, in the room and you, know a spying researcher, here and. You. Know I had to look them in the eye if. There's something about looking people in the eye that. Really makes a difference to, the kinds of research questions you. Ask. So. Just to kind of really wrap things up cuz I'm probably running a little late, now. It's. Just a pilot. But. I think it's a pilot that speaks to this broader argument that I've been making. About. There, being more than one way to do, data. Right. And some, of those ways do. In fact make room for situated. Knowledge making, but. Making, room for that also means, making the data flow a little bit differently, both. At a technical, level but also at a social, level. And. And part of that also means, that the social life of one's methods, have to be acknowledged. Right, I'm not pretending to. Not be in the room or not care about what these guys find right. All. Right so there's. You know coming, back coming back to Maryland's two-thirds work then. Right. She oh, do. I have to stop now. Right, Maryland sends that that social relations, are not simply, the things that we people have they're. The things that people think with, their. The vehicle, through which where we make new knowledge, and and I think that's what we were doing in the room that, day on the, the the volunteers, me, the device is the accidental design, they were all participants. In this knowledge making, not. So much as a research sampling, but, really kind of an unfolding. We're. The knowledge and the, question sort of unfolded, through these, relations, right so the relations, were a research vehicle in effect, and. We also did cut the network right we didn't chase any and all potential. Connection.
You. Know only some mattered, to us but. That cutting. Was really made possible, by a shared set of assumptions, about what a self is of. What a community is right, and and and the assumptions here at work in the room where selves. It had interior, T's right, they watched some, TV shows that other people didn't watch when. Traveling, and all the rest of it, but. They also, had, obligations. To one another and. And that was important, there and. Those were set of beliefs that I think existed, alongside also. Notions, of population, right there's kind of like coeval, together and. Notions, of stranger, hoods and and extractive, economies, as of data donation. You. Know we could even speculate. Just a little bit more radically, right we could say that you know maybe. Like it'd be interesting to think about what a data science, would look like that, assume wildly, different, notions, of personhood, and environment, right such as those of you find in the Melanesian literature. Which is where, stirrings where it comes from. All, right these are places where knowledge is never about something, kind of in the ether abstractly. But, actually where knowledge is really. Embedded, in physical, objects, which, people own right, so you can think about those dots then, has something, you own. But. Also because. You own it you. Take a very deep responsibility. For its consequences. Right for the sort, of the set, of relationships, that follow from moving, it over to another person. But. We're not dealing with Melanesians, hate, we're dealing with Americans, and. I'm. Making that that pretty radical, comparison. To. Remind us that the notions. Of populations, notions, of samplings, inexpert. Driven knowledge making right. Those aren't just epistemologies. Those are actually existing social relationships. That. The NIH as notions of data, of all of us really. Does have a particular, time and place to it right. But. To donate, at scale relies. On notions, of alien ability, that's. You know we can speculate more pointedly about where those come from. But, I think with this this this little foray into, environmental, health actually shows that n of a billion doesn't, just cut the network in a way that returns you, know things to its donors in a kind of a vague way, that. It actually works. Works. Through a set of relations, that place real limitations. On what can, and can't be known right that that, n of a billion is is big and it's useful but it actually doesn't have the capacity to learn the things that we, learned in that room that day. I think. The top weddings over ticket, sociology. My. Question, was, sort, of got started on this end. Of one with, some pretty interesting examples, where these, people. Had. A loss to find the, group that they were in diagnosis. Couldn't, you couldn't come up with the diagnosis, and. The. Person's computer, was in the room and so they were very clearly, in. Any sort of unique, to them situation, where this would've been one thing about a sense. Though. You conclude, on sort of this. More. Collective things. I'm. Curious. If instead of sort of starting with problems, and therefore looking, to see how this one works. Yes if we can sort of flip that and start, to think about some problems, where nf1 may, work better or worse and, how. To think about sort of when, we can't aggregate how much we should aggregate. Sort, of. We should, be forgiven, issue, where problem yeah I mean the other way to think about it would be to rephrase the question of, the. First, person, NASA engineer wasn't, actually in an end of one situation it was just that she didn't know that she was in the community, of nightshade. Your. Personal. Data to find the communities, in which. You. Know I think my talk in many ways was the elaborate version of what I learned the hard way which is you gotta find that through the social relationships. That's. You. Know no. Amount of you, know like, let's let's, try to be smart about that and you know, you. Know it's certainly you know where my mind goes are, the, you know the rare disease communities, and, you know the folks. Who have found their people. You. Know but in it it's like this interesting, problem of you if you don't know where yet then where do you go to learn to, know how to go where. And. I'm not sure and actually don't think data sense solves that I think it it helps, but I think the thing we found is that you can't you can't just do that in software like you've got to have people in some, room and maybe that's not environmental, help it's something else but. I just I think that's really I mean, that's. Just a thing I've just been struggling with is how do you get from that. I'm, an outlier to oh here. You. Know that's. You, know I think us stays in outlier hood right because everybody it's so fragmentary.
Was Doing their own thing. So. What's you, know what's the delta between that and you. Know the, disease communities, and you know the, environmental health folks we have a very clear understanding what, that is and, I wish I knew. Oh come. On. So. Maybe. I'll ask both questions okay. I'm. Really interested in this data sense tool this thing, you're creating a team of software engineers, and. What. What's the story of that tool where is it you know where is it where's. It going yeah. And. Then, I also thought I. You. Know I spent, some time myself trying to think about where as a small. End person, referred, where I fit in this data, science. Emerging. Data science world, and. I think in kind of early my early thinking about this I thought well we're, all kind of interested. In kind of the. Lived reality right, like a lot of data science is kind of trapped in the moment and ethnographers. Like to observe people, doing, things when they do that not just having them self-report, right yeah, but, you brought up this issue of convinced durability which I thought oh that that is another really interesting look. And. You. Know as an ethnographer we. Don't have to burden ourselves with commencer ability of our data we have, like 30 people that we've talked, to and lived with and understood, very well and we write a very long epigraphic. Monograph. Narrative. In a narrative format, and we. Have, that luxury. To sort of unpack. The. Very. Diverse, divergent. Stories that people tell us I have. No idea, how that all of us project could possibly, work if you have a like a million, n of ones, is. This, you know I'm thinking about your book where you talk about people, collecting. Data without. Any idea, what they're gonna do with it there's, not really a plan it's just we can collect it and we can stop Pilate and there's, value, and the market you, know reflects, that there's value in all this data which dock piling but I have no idea what to do with it or what we can get from it right you, think this NIH, project is maybe an example of that I think. What your is. A story, of kind, of emergent, communities, finding. Their own way to kind, of gain meaning, and find patterns and data through a lot of labor-intensive, but personally. Motivated. Efforts. And building. Connections how, does that how. Is an N of, one a million. Figure a valuable, proposition, exactly, well I mean I think the thing about the NIH project is this actually it's not end of a billion wines its end of a billion just, straight. Up. No. It is a million yeah no sorry it's yeah it. Kind of doesn't mean but, then with this data set sizes off, of a million people it's like okay. Yeah. I mean I don't think they're thinking these terms I suspect, not I mean they might be which would be great if. I were running that project the thing I would do is I would send like, 10, 20 teams out to do this stuff.
First A, positive. Assets they're like thinking about and, then. Maybe kind of have a way of thinking, about how you can get those questions right. I mean I actually I juice you, know and the live discussion in the environmental, health world is, you. Know that's not bad like you know we want you. Know like, large. Scale sensing, systems, to, catch the stuff that's I mean that's a huge area of undone science, it's just wofully. Inadequate, space so there's a role for supers. Like, large-scale stuff there but I would if. We're running it I would start with those smaller. Seemingly. Smaller things to get better at figuring out what the good questions are and. I wouldn't assume that, clinical. Research had all the answers to that although they certainly have some good ones. You. Know I think the commencer ability thing is, you. Know it's the thing I almost like about data is it makes your incommensurability. Z-- really, explicit, really, soon. In, ways that you have to think harder about when you're thinking of new graphically, but I think we do actually work on commencer ability I mean that's that's our whole thing in a way right that that the whole process of writing is, to this thinking through what. Is or isn't commensurate. And a lot, of the anthropology, right now is trying to make a much. Much. More radical claims, about, you. Know they're asking questions, about the impossibility, of, translating, between, social, worlds I mean they're kind of running. Towards, very. Radical. Incommensurability. Okay, well, it's. A big debate I won't, register an opinion about. It. Stop it and they have some really valid points in there so they're for you you know where I said but. Yeah I mean I think that, intersection, between the social komissarov, Allah T and the numerical. Commence. Our ability or not is this really. Interesting site. That, is. Worth attending. To ethnographically. And. Since you asked about data sense um we. Are we're. Winding down active. Sort. Of new feature development right, now there's. A live discussion about it's not a commercial product because it's just it's. It's not that way and, so we're asking questions about well should we be hosting it should we be you, know inviting, other partners, you, know should there be a role for open source like how does it all work so for. Those of you are interested in the software side. That's, of interest to you please supposed to talk, with me. Thank, You Don for actually. Bring. Out the nuance that became, one-in-a-million. There, are these interesting, you, know. But, it's fascinating to me that you've chosen, in. This pilot in. A sense it, is a group, of volunteers. Bottom-up. Grassroots you. Know sort of sharing, a common question. I'm. Wondering I, suspect, that there would be other types of social relationships. Where. Maybe is more top-down I'm, thinking for instance in the, workplace, where the manager, says all right everyone. In the world will that start collecting. Sunday. Question. Or. Some other context, so so do you think that. There. May be different. Parameters, that may make, these. End. Of, many. Work. Better or, not work at all you know in, addition to maybe, the social dynamics, dynamics. Yeah. Exactly. Yeah I mean there could be there, are so many you, know flavors. Of what men many could be and. You, know and because you know because. They are real social relationships, as well as a sampling. You, know there's always that. Like. Where does power sit.
How. Does that all work you. Know in an, odd way I can imagine if. What. You're doing is workplace, sensing. You. Know you can imagine where places where the hierarchies, are so strong they don't bother like, I just do it and, so you're, commencer. Ability thing. Isn't a problem anymore, you can imagine a more progressive work workplace, where that. Where. Commencer ability does, come to it an issue, does. Come to a head and you can also imagine one but that was actually employee driven, right, that was all about. Where. Is, you. Know what's happening at the worksite, that's. Making our jobs harder right so then the sensing is of us but the focus, is on. The physical infrastructure, or whatever else it is so. It's, a exerts, playing with what the social relationships, are then, gets get, you into the different, resource questions but yeah I think there's a million of them and they're worth exploring. Research. Book we had the, PhD, students decide that we're all going to put on biosensors. On ourselves and watch an episode of. What type of social dynamics peer pressure. I thing, but I want to maybe pick up on this thread a little bit on. Commencer. Ability and kind of maybe centralization. Or power and these little relations, and how sort. Of what you've seen. We're, sort of relates to like who who, is in charge of deciding what is and is not commensurable. Because I think I really, like the debates here just referring, to you around like anthropologists. Debating like to what extent you know radical. Thinking, it's really colonial. Empires didn't care they get it anyway, right by, generating over munch we're gonna get it whether we like. It or that. See. Those tensions at play in terms of who, gets to decide like what is a good, but. Commiserating. With care versus, something gets more, top-down. Or sort of has, concerning. Power. Yeah. The. You. Know I mean I sit in two places you know in the, the. The series of symposia that. Qsr labs put on between, public health researchers, and people, in the self tracking community, that. Was really at stake and and people were very much you know I mean there was this great quote that I was trying to weasel in but couldn't about, api's, as, you, know the data you get through api's is, an opinion, about the data worth having. And. You. Know there. I think. There were tensions in the room about. You. Know there were the you know let's let's scale up and donate folks and then there were the let's let's, hold on here, and I may, think they got to a place where there were there where there was even a discussion right so the first set of meetings it's. Like it was kind of like but. By the end you, know there were, clinicians. Talking, about doing, they're even. Doing their own self tracking, to come to understand, what it is they put patients through. Something. Happened. Yeah. I mean the other thing I found through the pilot. Was. You know the qsr folks are like they can get pipey about these things cuz they kind of know how to do it they you know they know they, know it's a choice. And, the. Thing I really struggled, with in the environmental, health pilot, was, that these were people who do they really understand, it or quality, data and how you would create aggregations.
And Good aggregation, method and bad aggregation, methods and sensor. Quality, I mean they're really into the technical details on that, but. For this stuff I was doing that was new and so. Having. A. Top-down. In the sense that somebody was a little bit more directed, about what. Is this well how about this direction, how about that direction that you need that kind of. Someone. To provide that little that, push, um. But. Then I mean, I had to draw on all, of the anthropological. Instincts. About how you deal with you, know a little bit of power you have. To. Do, that in a reasonable way and you, know so but sometimes a little bit of to
2017-12-07