Please flip your device.

AI Stories · When an Algorithm is Your Boss

Clocking into Worker Data

Dan Calacci is a PhD student with the Human Dynamics Group at MIT Media Lab researching how data, its stewardship, and analysis can impact community governance. They’re currently working on building frameworks and tools that allow informal collectives of workers and others to pool information in order to make better decisions and build power.

This is an extended cut of the interview from When an Algorithm is Your Boss that has been edited for ease of reading.

You’ve helped create an app called WeClock that helps workers collect their own data. How would you describe it?

WeClock allows workers to collect information about their daily life and use it to measure their working conditions. It takes the sensors that are sitting on your smartphone, the apps you use, your location, how your battery is being used, and gives you the opportunity to collect and share that with other workers and organizers. You can also see things like your daily movements and step count, which are collected through basic mobile APIs but are usually sent to some server and used for apps that you have no real control over.

How does it work, technically?

It’s an open source project built mainly in collaboration with the Guardian Project, a nonprofit group that develops a lot of privacy tech. We think of WeClock as a technology of resistance. The technical aspect is not so complicated, because it gives people ownership of data that’s already being collected by other apps. We use the core APIs available through Android and iOS. We also have simple custom surveys and worklogs. We just added a really simple tracker to the Android version, where you can mark when you’ve started and ended a work shift. Then, when we analyze the data, we can look back and compare it with things like the amount of time you spent standing or your pay. We also have an Apple Watch app for folks who can’t bring a phone into the workplace. Like, if their workplace is surveilled or if they’re a warehouse worker, for example.

How is this data collection empowering?

WeClock is an open-ended tool for use in specific worker-led campaigns. We’ve done a few pilots with labor unions across the US, and we’re doing a workshop right now with union leaders in the Global South, where they’re collecting information from workers about their locations and hours. We’ll use that to help them analyze data to gather evidence of wage theft. So it is much broader use case than just for gig workers. In fact, WeClock is more oriented towards traditional working contexts. This is a secure way of collecting information automatically and sharing that with organizers.

It’s one of an array of tools I’ve been working on for the past couple of years, like the Shipt Calculator or Gig Box.

Could you tell us about these too?

Gig Box is aimed at delivery drivers, especially drivers who work with multiple apps. It allows you to track shifts and collect location data during work. It’ll show you things like average pay per mile or pay after expenses. The goal is to build a work tracker and a pay tracker for delivery workers that is open source, independent, and that allows delivery workers to share their work experience data with organizers and researchers.

The Shipt Calculator was a texting bot that helped organizers and researchers audit an algorithmic change in the shopping delivery app, Shipt, in the summer of 2020. It allowed workers to text screenshots of their pay to a number, and it would automatically infer if workers were being paid by an old algorithm which was transparent, or a new algorithm which was not. We aggregated pay information from hundreds of workers to show that the new algorithm resulted in a significant pay cut for 41% of workers in our study.

What happens after people get control of their data?

Yeah, that’s a great question. What do people actually do with their data once they’re able to grab hold of it. This is both a really simple and complicated question, because analyzing data as a worker or even as an organizer is really hard. It takes a specialized set of skills. Like how to do data analysis or data visualization and come out of that process with meaningful information. My goal is to build tools that empower workers to tell stories, backed by evidence, to provide more ammunition to worker campaigns. It could be auditing an algorithmic change, or showing that warehouse workers are standing up for too long. This just augments the broader work of organizing.

Worker data is really only valuable for building power in the aggregate. In order to make meaning from this aggregated data, you need to do analyses guided by questions that workers themselves are asking. The end goal is not just to give you ownership of your data, it is to answer questions that are unanswerable now, or to provide stronger evidence of workers’ experiences.

So working closely with workers is important?

Absolutely. When you are building technology to facilitate resistance and to enable organizing, especially with groups that are experiencing oppressive conditions, like gig workers, it needs to be driven by the people who are living that experience. I worked as a bike delivery person for a few months, so I have a window on what it’s like. But that’s not enough. You don’t get to a point where your design paradigm is really shifted by just trying something once, or even by doing a “user inquiry” where you talk to a lot of workers.

Something WeClock is trying to do, is serve as a blank canvas that can be used by folks who are organizing themselves, so it’s actually co-created with the folks it’s aiming to serve.

What advice would you offer to others builders?

The most important thing is data access, making sure you provide people with a way to delete their data, or to share it with other organizers and to use it in these different, modular ways.

Another thing is usability. One of the things I found really frustrating in building tools for platform workers is that folks are used to using apps that are built by billion-dollar companies with enormous engineering and design teams. It’s so hard to compete with that as an independent developer trying to build something scrappy.

Putting aside the technology, organizing is hard. It’s about maintaining interpersonal relationships over time, building a network.

What else do we need to balance power?

I think the world needs two things desperately when it comes to algorithmic management. The first, is more tools to audit existing algorithms, and the second is participatory algorithm design.

Algorithmic management is spreading rapidly to every industry. And as it spreads, we’re going to need more tools of resistance. Right now, the most popular way to design algorithms is top-down, using feature engineering developed by data scientists, training an algorithm on a prior data set. We need other ways of making these algorithms that incorporate people’s values, their constraints and their goals.

There’s a small and growing body of work that questions how to make algorithms that are democratically derived. Are the interests of managers and workers so misaligned that there’s no way of coming to a solution? How do you ensure that people’s voices are part of systems that govern them, including at work? We need way more research on that.

Portrait photo of Dan Calacci is by Hannah Yoon (CC-BY) 2022

Mozilla has taken reasonable steps to ensure the accuracy of the statements made during the interview, but the words and opinions presented here are ascribed entirely to the interviewee.