I stumbled onto an interview with a researcher at Microsoft, Jaime Teevan. Researchers in her division are given broad latitude to investigate questions they find interesting; as she explains, “It’s just a bunch of people that the company hires with an interest in ensuring the long-term viability of the company, but we don’t have anything we’re supposed to do.” Teevan researches work, and how “we” can make it more efficient, easier, and better.
This interest I have in tasks has expanded beyond search to think about, how do we support complex tasks? I’m particularly interested in these complex tasks that are really hard to do.
One of the things I spend a lot of time doing is writing academic papers for publication. And sitting down and starting to write a paper is hard. How can we make that easier?
So one of the things we’re thinking about is how our time is so fragmented. A couple of people in my lab found that it takes a person 25 minutes to reach full productivity when you’re working — but we’re actually interrupted every 11 minutes. So people are never working at full productivity, because you have all this ramp-up time. A lot of research has been done to think about how to help prevent interruptions. But it’s also interesting to ask, Can we time their interruptions to be at a better time?
For example, I’ll block off an hour on my calendar, and be like, This is my work time. I’ll turn off Facebook or take email vacations. But rather than trying to change our environment to create long blocks, what about trying to fragment the work that we’re doing so that it fits in our fragmented work style?
So what would it look like to write a paper in 30-second bursts from my phone, so that I could get it done in these small little bursts? I’ve been working a lot with people who do attention management, thinking about, What kind of a context do you need to get a small task done? What is it like to be able to take a large, complex task and break it into these small, little pieces?
Teevan is a computer scientist by training, and I think that that’s apparent on close reading of this excerpt. One classical problem in mathematics is constrained optimization. The goal is to optimize an “objective function”, a single metric that captures the goodness of an outcome. But, of course, there are constraints or barriers to achieving maximum goodness; we don’t have infinite money, or infinite time, or any number of other resources. And thus the optimization becomes constrained; only some of the “solution space” is available for us to search.
Reductionism lies at the heart of applying mathematics to the real world. In the case of constrained optimization, there are two types of reductionism. One is in the objective function, in the idea that we know and can quantify our goal — in this case, making work “better”, whatever that means. Teevan, I think, takes it to mean that we get more accomplished. That, instead of always working, Sisyphus-like, in a cycle of ramping up our attention to attain maximum productivity, and getting distracted before we get there, we reside in that state on a permanent basis. The second type of reductionism is in the constraints. Teevan would like to assume, for simplicity, that interruptions are an immutable fact of work, but that our work habits themselves are not. We cannot avoid checking Facebook every 5 minutes, or seeing Slack’s red dot in our peripheral vision, but we can structure our tasks so that they lend themselves to short bursts of effort. The way Teevan structures the problem involves assuming that the fault lies in ourselves instead of our technologies. It is not surprising, then, that the solutions she proposes are technological instead of social or political: changing ourselves to fit our environment instead of our environment to fit ourselves.
I was struck by Teevan’s description of herself as an “idealistic” person: “I’m trying to change the world [laughs] — for the better.” I think many people in our field — of, broadly speaking, corporate-sponsored research — would describe ourselves in the same way, and with the same nervous laugh. She dreams of being able to
[S]tart breaking down tasks into these small, little micro-tasks, and surfacing them in a way, we can start automating a bunch of it, getting human intelligence in a really thoughtful way, allowing people to focus on the creative and interesting aspects of their job more efficiently.
The reason Teevan felt compelled to laugh is because we can never be sure that our actual work tracks our stated intentions. We would like to relieve white collar workers of tedium; the drudgery of data entry and spreadsheet fiddling and construction of Powerpoint slide decks. We would also like to make sure that, in the process, we aren’t putting anyone out of work. We tell ourselves that there are companies using data science and statistics and artificial intelligence for evil, but we aren’t employed by one of them. We are not the people Cathy O’Neill calls out in her book, Weapons of Math Destruction: we are not developing algorithms to identify criminals based on their pictures, or predicting which ones are more likely to recidivate based on their race or socioeconomic status. Instead, our work involves the far more mundane project of selling people shit they think they want and possibly don’t need, or of taking white-collar labor and improving its productivity by 10%. (The head of data at Stitch Fix, a multi-billion dollar company that relies heavily on data science, was asked about bias and AI, and how it related to his work. He responded, not entirely convincingly, “Well, I think there’s a difference in severity between maybe using AI in the criminal justice system and we’re about systematic bias there and being a retailer, a personal styling service.”)
Is any of that true, or is it simply self-serving? As you might expect, I’m not entirely sanguine on this issue. One of the areas of data science within my purview is warehouse “optimization”. Once again, the term “optimization” is loaded. In practice, optimization means reducing the amount of money spent by the company on the warehouse, while still maintaining the same level of output (or, perhaps, spending the same amount of money and increasing the level of output). One primary component of this cost is the cost of labor. If we can track what warehouse workers are doing — how many widgets they move from point A to point B per hour, how many orders they process — then we can identify those workers who are over- and under-performing, and reward the former and fire the latter. And my job, therefore, becomes collecting and organizing and aggregating the data to support those efforts. Our surveillance of the warehouse workplace is not quite as totalitarian as Amazon’s, but those are technical limitations, and do not necessarily reflect any lack of desire on the part of our management.
One thought I keep turning to, these days, is that there is less and less quantification, and thus less and less accountability, the higher up the “ladder” you are. With warehouse workers, there are clear metrics: units per hour, hours worked, downtime, overtime pay, cost per order, and so on. This impetus to track and quantify work, in order to reward and punish, has infected even white collar occupations.
This past winter, research firm Forrester released a report that linked overall U.S. economic productivity, which has been stagnant since 2011, with individual workplace productivity and argued that “insights-driven businesses” will be the companies that ultimately drive growth. This means companies that have data showing how their employees work are the ones that will thrive. And this past Wednesday, Microsoft made available to its enterprise Office 365 customers a new tool called Workplace Analytics. This tool, which relies on artificial intelligence for data analysis of worker behavior, joins WorkSnaps and BetterWorks as performance-monitoring software designed to log people’s computer activities to make sure every moment on the clock is a profitable one for their employers.
However, when management talks about productivity in a white-collar context, they’re usually talking about it in the “How do I get my workers to spend less time on Reddit and more time churning out finished projects?” context of personal productivity. This is the kind of productivity that presumes one can achieve a state of distraction-free industry, whipping through to-do lists without the encumbrance of busywork and optimizing every moment for greater workplace benefit. Software has been selling the promise of that kind of productivity since VisiCalc laid out the electronic spreadsheet that streamlined accounting. Analytical tools that promise to improve how you use software are a logical next step.
But my line of work, much like Jaime Teevan’s, has, so far, remained blissfully unaware of these developments. God knows I’d be horrified if someone collected data on what I spend the workday doing: the hours surfing Twitter, the dozens of times I needlessly refresh my email inbox, the meetings that go nowhere, the periods when I try to do work and end up staring blankly at my screen because I am unable to focus. I wonder if the reason we, meaning corporate researchers and data scientists, are so “idealistic” about our work is that we cannot envision the terminal state, in which it is deployed by the capitalists against us.
A Humble Politics Blog 