Conference PaperPDF Available

Working with Machines: The Impact of Algorithmic and Data-Driven Management on Human Workers



Software algorithms are changing how people work in an ever-growing number of fields, managing distributed human workers at a large scale. In these work settings, human jobs are assigned, optimized, and evaluated through algorithms and tracked data. We explored the impact of this algorithmic, data-driven management on human workers and work practices in the context of Uber and Lyft, new ridesharing services. Our findings from a qualitative study describe how drivers responded when algorithms assigned work, provided informational support, and evaluated their performance, and how drivers used online forums to socially make sense of the algorithm features. Implications and future work are discussed.
Working with Machines: The Impact of Algorithmic
and Data-Driven Management on Human Workers
Min Kyung Lee1, Daniel Kusbit1, Evan Metsky1, Laura Dabbish1, 2
1Human-Computer Interaction Institute, 2Heinz College
Carnegie Mellon University
{mklee, dkusbit, emetsky, dabbish}
Software algorithms are changing how people work in an
ever-growing number of fields, managing distributed
human workers at a large scale. In these work settings,
human jobs are assigned, optimized, and evaluated through
algorithms and tracked data. We explored the impact of this
algorithmic, data-driven management on human workers
and work practices in the context of Uber and Lyft, new
ridesharing services. Our findings from a qualitative study
describe how drivers responded when algorithms assigned
work, provided informational support, and evaluated their
performance, and how drivers used online forums to
socially make sense of the algorithm features. Implications
and future work are discussed.
Author Keywords
Algorithm; algorithmic management; human-centered
algorithms; intelligent systems; CSCW; on-demand work;
sharing economies; data-driven metrics; work assignment;
performance evaluation; dynamic pricing; sensemaking.
ACM Classification Keywords
H.5.m. Information interfaces and presentation (e.g., HCI):
Increasingly, software algorithms allocate, optimize, and
evaluate work of diverse populations ranging from
traditional workers such as subway engineers [16],
warehouse workers [28], Starbucks baristas [19], and UPS
deliverymen [7] to new crowd-sourced workers in
platforms like Uber, TaskRabbit, and Amazon mTurk [13].
How do human workers respond to these algorithms taking
roles that human managers used to play?
We call software algorithms that assume managerial
functions and surrounding institutional devices that support
algorithms in practice algorithmic management.
Algorithmic management allows companies to oversee
myriads of workers in an optimized manner at a large scale,
but its impact on human workers and work practices has
been largely unexplored. In recent years, the press and
many scholars have brought attention to the importance of
studying the sociotechnical aspects of algorithms [2, 10,
37], yet to our knowledge, there has been little empirical
work in this area.
We explored the impact of algorithmic management in the
context of new ridesharing services Uber and Lyft.
Algorithmic management is one of the core innovations that
enables these services. Independent, distributed drivers with
their own cars are algorithmically matched with passengers
within seconds or minutes, and the fare dynamically
changes based on where passenger demand surges, all
through the app on their mobile phones. Drivers’
performance is evaluated by passengers’ rating of their
service quality and drivers level of cooperation with
algorithmic assignment. Algorithmic management allows a
few human managers in each city to oversee hundreds and
thousands of drivers on a global scale. Drivers have little
direct contact with company representatives, but can
interact with each other through online forums to gain
social knowledge of the rideshare systems. This setting
allowed us to explore the practices that emerged when
algorithms assigned work, optimized work behavior
through information processing and evaluated job
performance. Do human workers cooperate with
algorithmically-assigned work? How much are people
motivated or demotivated by algorithmic optimization?
How effective is algorithmic, data-driven evaluation and
how do workers feel about it?
As a first step toward answering these questions, we
interviewed 21 drivers with Uber and Lyft and triangulated
their experiences by interviewing 12 passengers and
conducting archival analysis of online driver forums and
official company communication materials. The findings
highlight opportunities and challenges in designing human-
centered algorithmic work assignment, information, and
evaluation as well as the importance of supporting social
sensemaking around algorithmic systems. We use the
findings to discuss how algorithms and data-driven
management should be designed to create a better
workplace with intelligent machines, offering implications
for future work.
Our study makes the following two contributions to human-
computer interaction (HCI): 1) we describe the upside and
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. Copyrights for
components of this work owned by others than ACM must be honored.
Abstracting with credit is permitted. To copy otherwise, or republish, to
post on servers or to redistribute to lists, requires prior specific permission
and/or a fee. Request permissions from
CHI 2015, April 18 - 23, 2015, Seoul, Republic of Korea
Copyright is held by the owner/author(s). Publication rights licensed to
ACM 978-1-4503-3145-6/15/04...$15.00
downside of algorithmic and data-driven management, its
impact on human workers, and sensemaking strategies that
workers developed; 2) our results highlight new areas of
research for HCI, computer-supported cooperative work
(CSCW) and intelligent systems.
Two threads of research are closely relevant to the topic of
this paper: research on the impact of technology in the
workplace and interaction with intelligent machines.
The impact of technology in the workplace
A long and rich stream of research in HCI and CSCW has
investigated the impact of technology in the workplace:
email [15], instant messaging [18], organizational
information repositories [1], groupware [31], video
conferencing [9], awareness technologies [12], desktop
office software [8, 39], GPS for taxies [11], and robots in
hospitals [30]. Collectively, this research shows how
psychological, social, and organizational factors shape the
adoption of new technologies and how new work practices
and norms emerge in the workplace. To our knowledge,
very little empirical research has investigated the impact of
algorithmic management in the workplace, with the
exception of emerging studies of new work-monitoring
devices such as GPS for bus drivers [34].
Interaction with intelligent machines
Much research has investigated how people respond to
intelligent machines such as automated manufacturing
technologies [23, 36, 44], recommender systems [6],
context-aware systems [3, 43], agents [35], and robots [30].
As intelligent machines are only recently being integrated
into workplaces outside of factories, there is relatively little
work that examines intelligent machines in work contexts
with exceptions being systems in hospitals [30] and offices
[22]. To our knowledge, our paper is one of the first studies
that investigates how people respond when intelligent
machines take on managerial roles in a workplace.
Research on intelligent systems with different roles in other
contexts such as home, entertainment, or education has
identified important theoretical concepts and design
principles for successful human interaction with automated
and intelligent machines: establishing trust and cooperation
[33], creating accurate mental models [21, 35], providing
transparency and explanations [3, 6, 20, 24], and designing
shared control between humans and intelligent machines
[23, 36]. Our study significantly extends this work by
probing the consequences of these design dimensions of
intelligent systems in a new context, and by identifying new
problems and implications arising from novel managerial
roles of intelligent systems.
We conducted a qualitative study on algorithmic
management in Uber and Lyft. To understand drivers’
experiences and perspectives, we interviewed 21 drivers
and triangulated our findings by interviewing 12 passengers
and analyzing 128 posts by drivers on online forums and
132 official blog posts and communication materials from
both of the ridesharing companies.
Research context: Uber & Lyft ridesharing services
Uber and Lyft are currently the largest peer-to-peer
ridesharing companies. Founded in 2009 and 2012
respectively, Uber and Lyft operate in over 100 cities in 37
countries. Lyft attempts to create a social culture among
customers by encouraging passengers to sit in the front seat
and greet the driver with a friendly “fistbump. Uber
creates a more professional chauffeur environment where
social experience with the driver is not stressed. Anyone
over 21 years of age with a valid drivers license and a
personal vehicle in good condition can apply to be a driver.
Companies screen applicants with a background check, and
new drivers go through brief video-based online
orientations. Once accepted, new drivers become
independent contractors,” not employees, and are in
complete control over where, when, and how often to drive.
Algorithmic management in the ridesharing platforms
Three algorithmic features of Uber and Lyft passenger-
driver assignment, the dynamic display of surge-priced
areas, and the data-driven evaluation that uses acceptance
rates and ratings respectively correspond to decisional,
informational, and evaluation roles of human managers in
organizations [29].
Work assignment: Driver-passenger assignment
algorithms. Drivers need to turn on their ridesharing app to
be able to receive and execute their work. According to
Uber, after a rider requests a ride through the mobile
application, “the closest driver to that rider automatically
receives the trip request with a 15 second window to accept
it [42].” (Uber and Lyft will not reveal details of proximity-
based assignment algorithm, but in our study, we learned
that other things could be factored into the algorithm.) The
request includes information about the passenger’s location,
rating, picture, and name. If the driver accepts the request,
the passenger is notified, and the driver drives to the
passenger location to start the ride. With both Uber and
Lyft, drivers cannot choose or set preferences for specific
passengers or rides they wish to receive on their app. Lyft,
however, does allow drivers to block assignments from
passengers that do not pay the full suggested donation in
areas where pricing is still donation based. Uber and Lyft
allow only passive rejection of assigned passengers in that,
if the driver does not wish to accept the request, they must
wait out the allotted 15-second window. After this, the app
goes into stand-by mode until there is a new request.
Informational support: Dynamic in-app display of surge-
priced areas. Pricing is determined by a standard fare and
fluctuates according to a dynamic pricing algorithm. The
companies explain this feature to drivers and the public in
broad terms. For example, “when demand outstrips supply,
dynamic pricing algorithms increase prices to help the
market reach equilibrium [41].” In this paper we will refer
to this as “surge pricing,” adopting the term that Uber uses
(Lyft uses the term “Prime Time”). Surge pricing can play a
major role in shaping driver income, as their eighty percent
commission remains constant through these periods of peak
pricing. As of September 2014, both companies show
surge-priced areas in-app with map areas shaded in
different colors based on the price in real time. This is
designed to motivate drivers to move to areas where
demand (and price) is surging, in order to meet passenger
demand and maximize the total number of transactions.
Performance evaluation: Rating systems and acceptance
rates that track driver performance. After the ride both the
passengers and drivers evaluate each other through a 5-star
rating system. Lyft instructs that when rating a driver,
passengers should “consider whether your driver was
friendly, safe, a good navigator, and made you want to use
Lyft again [25].” Drivers also have an acceptance rate that
is calculated by the number of accepted rides divided by the
total number of requests sent to the driver. Drivers are
encouraged to keep a high ride acceptance rate through
occasional promotions that offer a guaranteed hourly pay if
the driver’s acceptance rate is above a certain threshold.
Drivers with a low average passenger rating and acceptance
rate may be subject to review or even immediate
deactivation on the ridesharing platform. Likewise
passengers who fall below a rating threshold risk rejection
by drivers, as drivers have the ability to ignore incoming
requests from passengers with ratings below their liking.
Long-standing drivers who maintain a high passenger rating
and acceptance rate are occasionally promoted to become
mentors or recruiters. In addition to driving for the service,
mentors and recruiters recruit new drivers and oversee the
application process, while earning extra income for these
We conducted semi-structured interviews with drivers to
understand their experience with algorithmic management.
We also interviewed passengers to confirm drivers’
opinions and perceptions about passenger behaviors
expressed in their interviews.
Participant recruitment
Online postings and paper flyers were used to solicit current
or former Uber or Lyft drivers and passengers. We posted
ads on Facebook driver groups, volunteer sections of
Craigslist, and relevant Reddit webpages. Paper flyers were
posted in three major cities in the US. A $10 gift card was
offered for participating in a 30-45 minute interview.
Driver interviews
We interviewed 21 drivers for ridesharing platforms who
operated in 13 cities across the United States (15 Males;
average age of 35 (SD=8.86)). Of the drivers interviewed, 5
drove for Uber only, 5 drove for Lyft only and 11 drove for
both Uber and Lyft. The drivers worked an average of 23.5
hours (SD=21.4) per week and had a range of experience
driving for a ridesharing platform (3 weeks to a year) with
an average tenure of 149 days (SD=107). 19 of the drivers
drove part time and 2 drove full time. Four of the 21 drivers
were also working for, or had previous experience with
similar driving services including taxis, trucks, and personal
chauffeurs, and car services. We conducted additional 30-
45 minute long interviews with these drivers to compare
their experience with more traditional driving jobs to Uber
and/or Lyft.
Interviews were done through video chat, phone, or in
person, depending on the interviewee’s location. The
interviewer began with questions about the driver’s last
ride, best or worst assignment and ride experiences. Follow-
up questions probed drivers’ understanding of three
algorithmic features and how their understanding
influenced their work strategies. For the ridesharing drivers
who also worked as taxi drivers, personal car service
drivers, or chauffeurs, we asked them to compare work
assignment and evaluation models in two driving jobs.
Passenger interviews
We interviewed 12 passengers who had used or currently
use Uber and/or Lyft in 4 cities across US (8 Females,
average age 24.2 years (SD=7.1)). On average, the
passengers had used the services 2-3 times per month for
about 5 months (SD=4.46). 8 of the passengers had used
both Uber and Lyft, 3 only used Lyft, and 1 only used Uber.
Interview questions focused on confirming or disconfirming
drivers perceptions of passengers’ use of services, in
particular, how they rate drivers and their attitudes toward
and behaviors around surge pricing.
Archival data: online driver forums & company websites
Online driver forums. We analyzed postings on online
driver forums as all drivers in our interviews mentioned that
the forums were primary knowledge sources and places for
socialization. We looked at two categories of online
forums: groups unmoderated by Uber and Lyft, including
various Facebook groups and Reddit pages, and official
company Facebook “lounges.” One author signed up to
become a Lyft driver, and was given access to a “new
driver” Facebook forum, hosted by Lyft, in which
information was disseminated with direct relation to the
company. We accessed other unmoderated private driver
forums on Facebook by requesting to join as researchers, to
avoid deceptively posing as drivers, and maintained an
observation-only status. Following the approach used in
[27], we sampled 128 online forum posts and comments
mentioning the algorithmic features selected out of
thousands made in a five-month period.
Company websites. We also looked at how companies
officially educate and communicate with drivers in order to
understand how much information they share about the
underlying mechanisms of the platforms’ algorithmic
features. We analyzed information on the Uber and Lyft
company websites and 132 official blog posts posted
between 2012 and 2014.
We triangulated our findings from interviews and archival
data. We qualitatively analyzed [32, 38] interview
transcripts, excerpts of online forum postings and
comments, and company communication materials using
Dedoose, a qualitative data coding software. We used three
algorithmic features of the ridesharing services to divide the
data set, and then open-coded the data about each feature at
the sentence or paragraph level. We analyzed the rest of the
data to identify important themes including social
sensemaking and socialization. This resulted in a total of
372 concepts. We then categorized the concepts into 18
themes explaining emerging phenomena. In addition to the
ones reported in the paper, themes such as employee
(de)identification with company culture emerged but were
excluded in further analysis. We focused on 8 categories
relevant to our research questions around algorithmic
management, and used modeling techniques and affinity
diagrams in order to explain the relationship between the
categories. The final coding scheme had good reliability
across two coders when tested with 10% of the transcripts
(Kappa=.71). Conflicts between coders were resolved
through discussion.
We describe how drivers responded when algorithms
assigned work, provided informational support, and
evaluated job performance, as well as how drivers used
online forums to socially make sense of the algorithmic
features of the system.
Background: driver motivation
According to drivers, one main advantage of working for a
rideshare platform was the flexibility that the system
affords in terms of where and when to work, and the low
level of commitment that is required by signing up. Some
individuals drove full-time, but many also drove for fun,
out of curiosity, or on a part-time basis. Many drivers used
the ridesharing app in collaboration with their own daily
routine to earn extra income, turning the driver app on for
the daily commute for example, or doing chores around the
house while waiting for a ride request to come in. In
addition to the added financial flexibility that rideshare
work affords, many drivers we interviewed mentioned
social motivations for rideshare driving. Several drivers, for
example, weighed the fun of meeting and having
conversations with new people and the desire to help out
the community as greater than or equal to their motivation
to earn extra income.
Algorithmic work assignment: proximity-based driver-
passenger assignment
Our findings highlight how transparency of algorithmic
assignment influences worker cooperation, work strategy
and workaround creation, and describe the potential impacts
of automating choices that workers used to have in similar
work settings.
Accepting and cooperating with algorithmic assignments
Previous research suggests that people may cooperate less
with work assignments made by machines rather than
people [33]. In Uber and Lyft, the way that assignments
were presented to drivers on their app and the regulation of
acceptance rate cut-off influenced drivers to accept as many
assignments made by the assignment algorithms as
possible. I mean you can always decline to pick up a
passenger if you can make that decision within 12 seconds.
(Uber/Lyft) make it sort of difficult to say no for a couple of
reasons. […] when they show the spot on the map where
you're going to pick someone up its very zoomed in so if
you're not immediately familiar with the area you probably
wouldn't be able to discern within 12 seconds if its
somewhere you want to go or not. They just tell you how far
away it is in driving time (P4).”
Interestingly, one factor that influenced driver cooperation
was whether the assignment made sense to them. While the
assignment was generally based on proximity, there were
other factors that influenced assignment such as passenger-
driver mutual rating and driver login time. This sometimes
caused drivers to receive requests from distant passengers
to which they were not the closest driver. When this
happened, many drivers reported rejecting the unfavorable
ride assignment given that they would have to drive a great
distance (such as 15 minutes) to the pick-up location. For
example, P23 stated: I’m one that keeps a close watch of
where [other drivers] are when I’m not with a passenger.
So if I’ll see three [other drivers] over on the [area A] and
I get a request from [area B] to the [area A] knowing that
there should have been three [drivers] sitting right there
ready to go, tells me one of two things happened. Either all
three of them passed on the ride which is highly unlikely
that they’re sitting here and they pass on a ride that’s right
in front of them. Or the system didn’t coordinate the GPS
correctly and sent it to me over ten minutes away instead of
somebody that was 30 seconds away. In this quote, it is
unclear whether the assignment was due to errors in the
system, or for other legitimate reasons, because no
explanation was given about how the assignment was
decided on the drivers’ app. P23 assumed that the
assignment was made by mistake and rejected the
assignment. On the other hand, even with distant and
inconvenient requests, drivers accepted rides when the
assignment made sense to them. For example, P13 stated:
Distance wise, sometimes I’ve gone like 17 miles, but
that’s not really the [passenger’s] fault; that’s because
there’s just not that many drivers out right now and I just
really am the closest.This suggests that an explanation of
why certain assignments were made might be an important,
but currently missing feature.
Creating work strategies and workarounds for algorithmic
Drivers used their understanding that assignments are based
on proximity to create their work and workaround strategies
that helped them maintain control that the automated
assignment did not support as part of the existing system
Drivers strategically controlled when and where to work
and when to turn on the driver mode of the app to get the
types of requests and clienteles that they preferred: limiting
the area that they worked in by turning off the driver mode
when returning from a long-ride, avoiding bad
neighborhoods to avoid dangerous situations, going
downtown for successive short-rides during the lunch time,
not going to bar areas to avoid drunk passengers, and
instead, staying in residential areas to drive people to bars.
Drivers attracted repeat passengers by arranging rides via
phone, asking passengers to request a ride once they were
in the driver’s car to get matched. Drivers used online
forums to post about bad passengers so that other drivers
could avoid them, similar to self-regulation strategies of
mTurkers [17].
Drivers also distanced themselves from one another by
checking other drivers’ locations on the map so that they
did not compete with each other for passenger requests.
When drivers desired a break but did not want to turn off
their driver applications to benefit from an hourly payment
promotion, they parked in between the other ridesharing
cars in order not to get any requests.
The companies communicated only general rules of
assignment, e.g., “the closest drivers get assigned,” and this
general understanding helped drivers create their work
strategies. The lack of details of the assignment algorithm,
however, seemed to foster drivers ambivalent, sometimes
negative feelings toward the companies: “Uber is very
close lipped about what actually happens right I mean they
say oh we route it to the closest driveror whatever but
who really knows what’s going on behind the scene its up
to whoever engineers their iPhone app (P4).
More knowledge more advantage
Our findings suggest drivers benefited from deeper
knowledge of the assignment algorithm. Drivers with more
knowledge created workarounds to avoid undesirable
assignments, whereas those with less knowledge rejected
undesirable assignments, lowering their acceptance rating,
or unwillingly fulfilled the uneconomical rides. For
example, P2 had knowledge that Lyft’s assignment
algorithms take into consideration how long drivers have
been online and that a drivers radius for pickups will
increase as they wait for passenger assignments. He used
his knowledge to periodically turn on and off his driver
application while at traffic signals, so that he did not get
distant requests. However, this information was not
publicly made available to all drivers, and in our interviews,
Lyft drivers who did not have this knowledge attributed the
distant assignment to the error of the assignment system, or
drivers with higher ratings getting priority. These drivers
could not create workaround strategies to avoid distant
Getting assigned versus choosing whom to pick up
Drivers were generally satisfied with their level of control
over assignment algorithms, except for a few drivers who
desired to have control over the radius that the assignment
algorithm searches to assign the passenger. Interestingly,
one driver P17 who was also a Yellow Cab driver preferred
his Taxi dispatching system where he could see all the
incoming requests and choose freely from among them. He
explained that he could strategically choose the location of
ride requests in the taxi system, and he had developed
knowledge of how to best use them: At certain times of
certain days, you know that they’re usually a lot of really
good trips happening in those areas. Like, Thursdays
around four o’clock in, like, [area names] you know that
there’s gonna be a lot of airport trips, for instance. So you
can focus on those. Another thing that’s good about it is
[…] if it’s […] a busy Friday night. You’re just growing
tired of mining a certain area. You can totally shift. And a
good way to do that is to take something that’s not in a
close area that you think maybe going from far, but then
coming back in. […] It gives us the option for a change of
pace. Uber and Lyft’s automated assignment got rid of
this fine level of control and predictability. He said that
while he could try to be in those locations in Uber and
Lyft’s system, it did not guarantee that he would get
requests in the area. Often, he would get requests outside of
the area, and he did not want to drive to these areas just for
a change of pace.
Algorithmic information support: dynamic in-app
display of surge priced areas
Surge-pricing algorithms are used to optimize pricing in
online, airline, and hotel markets, among others. Our
findings show breakdowns when these algorithms are used
to influence human behavior.
Algorithmic information not accommodating human abilities,
emotion, and motivation
Some drivers in cities where surge pricing was applied
citywide, instead of being neighborhood-based, reported
that they would go out to drive when they received surge-
pricing notifications. Other drivers reported that the times
when they were available to drive were already in line with
surge-priced timing.
More than half of interviewed drivers, however, were not
influenced by surge pricing information as the supply-
demand control algorithms failed to accommodate their
abilities, emotion, and motivation. Surge pricing changed
too rapidly and unexpectedly to utilize the information in a
strategic way to boost their incomes. Surge areas were on
and off, sometimes by the second, and being in the surge
area did not guarantee requests from within the surge area.
Drivers reported getting no requests or requests from
outside of the surge area (which do not qualify for surge
pricing), or the surge price disappearing when they arrived
to the surge area.
The economic and rational assumptions of the supply-
demand control algorithm did not always motivate drivers
behaviors, as it does not account for feelings of unfairness
about dynamic surge pricing [14]. Most passengers reported
that surge pricing was unfair, and they tried to avoid it if
they could. Some drivers, in particular ones that used the
ridesharing services as passengers, expressed that they
thought that surge pricing was unfair and they did not try to
chase surge-priced areas. The appeal of increased incomes
in surge priced areas did not motivate some drivers who
also drove primarily for social, rather than financial reasons.
Trusting their own knowledge more than algorithmic data
A couple of drivers had more trust in their own knowledge
and experience driving in the city than in the unpredictable
surge pricing algorithms. In part, the drivers did not have
knowledge to evaluate how accurate surge-priced areas
were. For example, P19 stated:They probably do have
some kind of algorithm over people who open up the app to
request the ride, and they might have noticed, but they don't
tell us how those [surge-priced areas] work. I ignore them
for the most part, because I'm from here. […] I've lived
here 35 years."
Algorithmic, data-driven evaluation: performance
evaluation through acceptance rate and driver rating
The regulation of the acceptance rate threshold and the
driver-passenger rating system offered many benefits to
overall service functioning. However, these numeric
systems that made drivers accountable for all interactions
were sometimes seen as unfair and ineffective and created
negative psychological feelings in drivers.
Unfairness in treating all assignment rejections equally
The regulation of the acceptance rate threshold encouraged
drivers to accept most requests, enabling more passengers
to get rides. Keeping the assignment acceptance rate high
was important, placing pressure on drivers. For example,
P13 stated in response to why he accepted a particular
request:Because my acceptance rating has to be really
high, and there’s lots of pressure to do that. […] I had no
reason not to accept it, so […] I did. Because if, you know,
you miss those pings, it kind of really affects that rating and
Lyft doesn’t like that.
Assignment algorithms penalized equally all drivers’
rejections of passenger requests, which lowered the drivers’
acceptance rates. Sometimes drivers, however, had
legitimate reasons and circumstances for rejecting
passengers. For example, female drivers did not accept
male passengers without pictures at night because of safety
concerns. Drivers often rejected passengers “blacklisted”
for their misbehavior on online driver forums. Sometimes
technical glitches in the app showed the request with only a
few seconds left to accept. When they felt that they had
legitimate reasons for rejecting the requests, drivers would
sometimes send emails to company representatives, hoping
that they would not get penalized for the legitimate
assignment rejects, but often times they did not hear back
from the companies.
Inaccuracy in using only numeric metrics of service quality
Our findings suggest the passenger-driver rating system
established basic trust and service attitudes in the
ridesharing systems, however they fell short when used for
driver performance metrics.
Drivers used passenger ratings to decide whether to accept
the request or not, trusting 5-star passengers and being
cautious with 3-star passengers. While not paying equal
attention to driver ratings, passengers reported that the
existence of a rating system gave them a sense of security.
The rating system also promoted a service mindset in all
drivers. For example, P16 said: […] I want to get all five's.
So I try to be friendly and engaging with the passengers.
And give them options when they get in, like, you know, ‘Do
you want A/C? Do you want the windows down? What kind
of music do you want to listen to?’ I even have a candy tray,
gum, stuff like that.”
Drivers took their ratings seriously. High ratings such as
4.98 became a source of pride whereas a rating below 4.7
became a source of disappointment, frustration, and fear of
losing their jobs. Being tracked, evaluated, and judged by
each passenger seemed to have a negative psychological
impact on drivers who did not have scores near 5. P11 said:
[the rating system] makes you cautious that what you’re
doing is being judged and rated and if you’re rated poorly
enough over a period time then eventually the platform
could ask you to stop driving for them.
Many drivers felt that the average of the passengers’ ratings
of the drivers was not reflective of their driving
performance and services, as P22 stated: “It's like in
baseball a stat line doesn't always tell the particulars of a
player. A player could hit 35 homeruns and knock in 100
runs but if they're hitting .240 and strike out 150 times, that
doesn't mean they were such a productive player.” Many
reported that various physical and psychological states of
passengers such as being in a hurry and late for a flight or
being drunk, could influence them to give a lower rating
after the ride. Additionally, drivers noticed that passengers
misattributed system faults and negative experiences that
drivers could not control to drivers themselves, which in
turn resulted in lower ratings (e.g., surge pricing, traffic
jams, GPS errors etc.). Drivers also often attributed their
low rating to passengers reviewing them using
inappropriate review rubrics. For example, drivers often
perceived that passengers rated them as if it were a 5-star
rating system for products, movies, or restaurants, where
perfect ratings are rare. This led many drivers to conclude
that passengers needed education for the rating system in
ridesharing services. On the other hand, most passengers we
interviewed reported that they are more lenient and positive
when rating drivers, while they are more critical in their
online reviews of other goods.
Because of their perception that there are many
uncontrollable factors influencing driver ratings, drivers
seemed to develop a detached, indifferent attitude once
their scores were above a certain threshold of deactivation
risk. The rating averages scores from one hundred or more
rides. With this average, the impact of each ride is reduced.
For example, P8 stated: Well I used to micromanage my
rating so to speak. I used to sweat and be oh my gosh my
rating is now going down - it’s a 4.85 that kind of thing.
Now I don’t worry about it. I see there’s a lot of error that
can take place in the rating.”
Online social sensemaking
As drivers worked independently in distributed locations,
online driver forums became a primary avenue for the
driver socialization and system sensemaking. Drivers
discussed the workings of the ridesharing systems
algorithmic management. One of the successful online
sensemaking examples was about improving and
maintaining driver performance in ratings and acceptance
rate. When novice drivers asked for tips to improve their
ratings, other drivers shared strategies and lessons that they
learned over time. For example, one driver posted questions
about how to improve her rating (4.63) after giving 38 rides
in three weeks. About 50 comments were made within two
hours of her posting, empathizing with her feelings,
disclosing common experiences of going through the initial
hurdle, and sharing specific strategies that they developed
creating their own service information brochure for the
backseat, going to downtown during the day for many short
rides, etc. The experienced drivers also explained that the
rating would stabilize over time, and advised that she
should not stress about it too much. Often times, original
question askers followed up on the posting, making
comments that the strategies had worked.
On the other hand, sensemaking activities around
assignment algorithms and surge pricing seemed less
successful in terms of informational usefulness. Common
posts were questions of how assignment algorithms and
surge-pricing work, how to interpret dynamic visualization
of surge pricing areas, and real-time questions of frustrating
events—from getting no requests in surge priced areas or
getting distant requests that required long driving.
When answers to drivers’ questions went beyond the
information that the company officially communicated, the
social discussion on the online forums focused on providing
social and emotional support, rather than informational
support. For example, one driver posted frustration in real
time that he just got an assignment from across the city
from east to west even when he saw other drivers around
the requesting passenger that were closer. Many comments
were made to the posting that provided emotional support.
For instance, another driver from the west chimed in to say
“I should have logged in to save you (from driving from
east to west),” and other drivers said “it sucks when it
happens.” But none of the comments provided an
explanation as to why such an assignment was made.
Postings that made jokes on the surge pricing areas that
seemed wrong according to common sense (e.g. a surge
area extending into a lake), or ones that humorously
interpreted shapes of surge-priced areas as “Surgemon,”
also reflect an attempt to have control over the unknown
through humor and emotional processes instead of rational,
cognitive ones [40]. On rare occasions, company
representatives came to the forums and answered drivers’
questions, but their answers were usually washed away in
the influx of other forum postings and comments.
We discuss how to use our findings to improve the design
of algorithmic and data-driven management.
Designing algorithmic work assignment
Algorithmic passenger assignment in Uber and Lyft
automatically distributes myriad ride requests to drivers in a
matter of seconds. Drivers’ quick and frequent acceptance
of the assignments ensures the efficiency of the service,
maximizing the number of passengers able to get a prompt
ride. Our findings suggest that in algorithmic work
assignment, not only the source of the assignment (i.e.,
human versus algorithm), but also how the assignment was
presented and regulated, influences worker cooperation
with the assignment. Choices of which information to
present on screen, a short time limit to accept the ride, and
the acceptance rate collectively reinforced drivers’
cooperation with the assignments in our study.
Our findings also suggest that transparency of assignment
process could elicit greater cooperation with assignments,
especially undesirable ones. While the company explained
that their assignments are based on proximity, there were
various additional factors that the algorithm took into
account in addition to passenger-driver distance. This
sometimes resulted in assignments where passengers were
not assigned to the closest driver. Providing explanations
for [6], or allowing workers to ask questions about [20, 24]
each assignment could reduce drivers rejection of distant
assignments by reducing their attribution of such
assignments to technical errors. Transparency may also
improve some drivers’ ambivalent feelings toward the
companies. This finding is consistent with previous
research on recommender systems where transparency
improved people’s trust and acceptance of
recommendations [6]. Moreover, the study highlights new
implications of transparency, which have received
relatively little attention in previous research on intelligent
systems: how transparency in algorithmic assignment helps
people create better work strategies and workarounds.
Drivers with more detailed knowledge about the assignment
algorithm could create workarounds to avoid less
economical rides whereas people only with a general
understanding of proximity-based assignment could not.
The stakeholders involved with work platform apps
(companies and workers) complicate providing
transparency. In previous work on transparency of
intelligent systems, explaining a user model usually
sufficed [3, 6, 20, 24]. Algorithmic work assignment offers
new challenges in design transparency where fully
disclosing the algorithm may not be a viable solution.
Companies may be unwilling or unable to share the
underlying mechanisms of their assignment algorithms, as
they might be patented or proprietary assets. Companies
may also desire a degree of user ignorance to prevent the
system from being gamed.
We were surprised that ridesharing drivers desired little (or
did not feel entitled to) direct control over the assignment
algorithm, for example, specifying pick-up locations or
being able to see and choose from all requests. We believe
the organizational context of being independent contractors
played an important role: the flexibility and choices that the
ridesharing drivers have in work compensate for the lack of
control in assignment algorithms. Another explanation for
why drivers did not desire control could be the lack of
experience with other systems. For example P17, who
worked as both a taxi and rideshare driver, preferred the
taxi assignment process where he could directly access and
choose passenger requests. P17 did not like the ridesharing
assignment systems because algorithms made decisions that
he used to make himself, making him feel he lost agency
regarding strategies that he had developed to maximize his
income. This could be interpreted as resistance to change,
but also raises open-ended, ethical questions about the trend
in new technology that sacrifices individual control for the
sake of overall system efficiency, and its implications for
learning and development on the job.
Designing algorithmic information support
Supply-demand control algorithms were originally designed
to solve mathematical optimization problems that involve
non-human entities. In Uber and Lyft, however, they are
used to motivate and control human behaviors. This causes
problems, as the supply-demand control algorithm does not
consider the pace at which drivers work. Consistent with
previous research on a smart agent that tried to encourage
sustainable behaviors [5], the algorithm failed to account
for feelings of inequity people had toward surge pricing,
and ignored the social and altruistic motivations of drivers.
This highlights the importance of making algorithmic
management accommodate: a) the speed and the way
humans work, b) diverse types of motivations rather than
only economic ones, and c) emotions that people feel about
the decisions that algorithms make. In addition, some
drivers did not trust the surge-priced areas as they trusted in
their experiences more. Transparency in how the surge-
priced area was computed in real-time could improve
workers’ trust toward the algorithmic information.
Designing algorithmic, data-driven performance
Using driver ratings and acceptance rate, companies are
able to evaluate drivers at a large scale. Driver ratings in
particular may seem to be a legitimate evaluation metric
because customer satisfaction is an important metric of
service success and human service provider quality. Using
only the tracked performance data in evaluating workers,
however, revealed many complications that can occur when
one relies too heavily on quantified metrics without deeper
consideration of their meanings and nuances. Consistent
with previous research on letter-grading systems or numeric
evaluation of teaching skills [4], many random factors that
are out of drivers’ control influence the way passengers rate
drivers. The efficacy and accuracy of averaged collective
evaluation, rather than an in-depth holistic evaluation done
by a human manager or peer, is also in question. As P18 put
it, “you are at the mercy of random people, in [his other
work], you are evaluated by people that you know.Our
study also shows drawbacks of adopting the 5-star rating
system shared with online products, content, or business
reviews to review human workers. Drivers felt that
passengers rated conservatively as they do in online
reviews; yet interviews with passengers suggest that they
are being more lenient and positive than drivers think. This
misconception suggests that a 5-star rating metaphor and
rubric may have brought up inappropriate associations.
Finally, the long-term motivational effect of the rating is
also in question. As the drivers’ ratings were averaged over
multiple rides, the impact of one positive or negative ride
was minimized, and drivers in our study became less
sensitive to the changes in their ratings once they were
above a minimum threshold.
Successful management provides work protocols and
allows improvisation in response to changes and exceptions
[29]. On the other hand, assignment algorithms penalized
equally all driver rejections of assignments even when
certain drivers had legitimate reasons and circumstances for
doing so. While we did not observe serious problems from
this lack of flexibility in algorithmic assignment in our
study, it brings up an open challenge in creating flexibility
in algorithmic management.
In most online rating systems, review is optional, and many
even skip the process. In the ridesharing services, all
passengers were encouraged to rate their service encounter,
and most of them did. Being held accountable for every
interaction, drivers were very aware of the existence of this
external evaluation. Trying to deliver good services for all
service interactions could pose psychological stress to
workers. Additionally, as extensive research on the impact
of extrinsic rewards on intrinsic motivation suggests, the
external device could weaken the intrinsic motivation that
drivers might have and change the meaning that they
attribute to their behaviors. From the passengers’ point of
view, the ambiguities in providers’ motivation for
friendliness and good services risks rendering provider-
client interaction more superficial and perfunctory.
Supporting social sensemaking online
Our study showed that online forums became a main place
where drivers socialized, asked questions of each other, and
exchanged knowledge and strategies. In most research on
sensemaking and mental models of intelligent systems, the
focus has been on individual sensemaking [20, 21, 24, 35,
43]. Our study shows that social sensemaking is another
important activity that needs to be better understood and
supported for intelligent systems to be successfully adopted.
Social sensemaking activities on the driver forums followed
the form of “fragmented social sensemaking [26] where
there were many active contributors but no central authority
figure to synthesize different ideas and narratives into a
coherent story. This type of sensemaking was effective for
discussing rating improvement strategies where there were
no right or wrong answers, and workers’ experiences and
learned and improvised strategies played critical roles. On
the other hand, fragmented social sensemaking fell short on
subject matters where only an authority figure had the right
information. This highlights opportunities to design
structured online social sensemaking of algorithmic features
where individuals can build on each other’s knowledge.
Like any study, this work has some limitations. Our results
are from interviews with a small sample of drivers,
passengers and archival data analysis. We could not
interview developers or official representatives of the
companies as it was against company policy. Our findings
should be complemented by future research that uses
different research methods such as ethnography, surveys or
experiments. Our study was done in the specific context of
ridesharing (on-demand, independent contractor work) thus
further work needs to be done in different organizational
contexts such as with full-time or co-located employees.
HCI and CSCW have a long history of research on
computational systems that support individual work and
collaboration. Our work suggests that algorithmic
management in the workplace is a new and fruitful ground
for research, where the same effort can be made to establish
theories and design principles for algorithmic management.
The research presented in this paper also offers implications
for research on intelligent systems. Much of the previous
work on intelligent systems was done in the context of
individual user and non-work settings. Our work suggests
that important concepts and theories in the field of
intelligent systems such as transparency and control of
systems, user mental models, and sensemaking need to be
updated to accommodate social and organizational contexts
that involve multiple stakeholders and new roles of
intelligent systems in workflows. Finally, this research
raises the need for new methodological research in HCI and
interaction design on designing human-centered algorithmic
management. HCI and interaction design have established
systematic processes and methods for designing human-
centered interfaces and interactions. Compared to designing
and building traditional user interfaces, designing
algorithmic management will require different ways of
specifying and evaluating requirements, states, and
Increasingly, software algorithms allocate, optimize, and
evaluate work. In this paper, we explored the impact of this
algorithmic, data-driven management in the context of new
ride sharing services, Uber and Lyft. Our findings from a
qualitative study highlight opportunities and challenges in
designing human-centered algorithmic work assignment,
information, and evaluation and the importance of
supporting social sensemaking around the algorithmic
system. The implications for HCI, CSCW, and research on
intelligent systems are discussed. We hope this research
inspires future work, so that we support human workers to
work with intelligent machines not only in an effective, but
also a satisfying and meaningful way.
This research was supported by NSF grants CNS-1205539
and ACI-1322278. We thank Su Baykal for helping us
collect and analyze data, and the anonymous reviewers for
their feedback that improved the paper.
1. Ackerman, M. S. (1998). Augmenting organizational
memory: a field study of answer garden. TOIS, 16(3),
2. Barocas, S., Hood, S., & Ziewitz, M. (2013). Governing
algorithms: A provocation piece. SSRN.
3. Bellotti, V., & Edwards, K. (2001). Intelligibility and
accountability: human considerations in context-aware
systems. HCI, 16(2-4), 193-212.
4. Braga, M., Paccagnella, M., & Pellizzari, M. (2014).
Evaluating students’ evaluations of professors.
Economics of Education Review, 41, 71-88.
5. Costanza, E., Fischer, J. E., Colley, J. A., & Jennings,
N. R. (2014). Doing the laundry with agents: a field trial
of a future smart energy system in the home. In Proc. of
CHI, 813-822.
6. Cramer, H., Evers, V., Ramlal, S., ... & Wielinga, B.
(2008). The effects of transparency on trust in and
acceptance of a content-based art recommender.
UMUAI, 18(5), 455-496.
7. Davidson, A. & Kestenbaum D. (2014). The Future of
Work Looks Like a UPS Truck. NPR.
8. Dourish, P. (2003). The appropriation of interactive
technologies: Some lessons from placeless documents.
CSCW 12(4), 465-490.
9. Egido, C. (1988). Video conferencing as a technology to
support group work: a review of its failures. In Proc. of
CSCW, 13-24.
10. Gillespie, T. (2014). The Relevance of Algorithms.
Media Technologies: Essays on Communication,
Materiality, and Society, 167-193.
11. Girardin, F., & Blat, J. (2010). The co-evolution of taxi
drivers and their in-car navigation systems. PMC, 6(4),
12. Grudin, J. (1988). Why CSCW applications fail:
problems in the design and evaluation of organizational
interfaces. In Proc. of CSCW, 85-93.
13. Hassan, U., O’Riain, S., & Curry, E. (2013). Effects of
expertise assessment on the quality of task routing in
human computation. In Proc. of Workshop on Social
Media for Crowdsourcing and Human Computation.
14. Haws, K. L., & Bearden, W. O. (2006). Dynamic
pricing and consumer fairness perceptions. Journal of
Consumer Research, 33(3), 304-311.
15. Hinds, P., & Kiesler, S. (1995). Communication across
boundaries: work, structure, and use of communication
technologies in a large organization. Org. Sci., 6(4),
16. Hodson, H (2014). The AI Boss that Deploys Hong
Kong’s Subway Engineers. New Scientist
17. Irani, L., & Silberman, M. (2013). Turkopticon:
Interrupting Worker Invisibility in Amazon Mechanical
Turk. In Proc. of CHI, 611-620.
18. Isaacs, E., Walendowski, A., Whittaker, S., …& Kamm,
C. (2002). The character, functions, and styles of instant
messaging in the workplace. In Proc. of CSCW, 11-20.
19. Kantor, J. (2014). Working Anything But 9 to 5. NYT
20. Kay, J. & Kummerfeld, B. (2012). Creating
personalized systems that people can scrutinize and
control: Drivers, principles and experience. TiiS, 24-42.
21. Kulesza, T., Stumpf, S., Burnett, M., & Kwan, I. (2012).
Tell me more: the effects of mental model soundness on
personalizing an intelligent agent. In Proc. of CHI, 1-10.
22. Lee, M. K., Kiesler, S., Forlizzi, J., & Rybski, P. (2012).
Ripple effects of an embedded social agent: a field study
of a social robot in the workplace. In Proc. of CHI, 695-
23. Lee, J., & Moray, N. (1992). Trust, control strategies
and allocation of function in human-machine systems.
Ergonomics, 35(10), 1243-1270.
24. Lim, B. Y., Dey, A. K., & Avrahami, D. (2009). Why
and why not explanations improve the intelligibility of
context-aware intelligent systems. In Proc. of CHI,
25. Lyft (2014) Rating Your Driver. Retrieved from
26. Maitlis, S. (2005). The social processes of
organizational sensemaking. Academy of Management
Journal, 48(1), 21-49.
27. Martin, D., Hanrahan, B. V., O'Neill, J., & Gupta, N.
(2014). Being a turker. In Proc. of CSCW, 224-235.
28. McClelland, M. (2012). I Was a Warehouse Wage
Slave. Mothers Jones.
29. Mintzberg, H. (1973). The Nature of Managerial Work.
Harpercollins College Div.
30. Mutlu, B. & Forlizzi, J. (2008). Robots in organizations:
the role of workflow, social, and environmental factors
in human-robot interaction. In Proc. of HRI, 287-294.
31. Orlikowski, W. J. (1992). Learning from notes:
Organizational issues in groupware implementation. In
Proc. of CSCW, 362-369.
32. Patton, M. Q. (1990). Qualitative evaluation and
research methods. SAGE Publications, inc.
33. Parise, S., Kiesler, S., Sproull, L., & Waters, K. (1999).
Cooperating with life-like interface agents. Computers
in Human Behavior, 15(2), 123-142.
34. Pritchard, G., Vines, J., Briggs, P., & Olivier, P.
(2014). Digitally driven: how location based services
impact the work practices of London bus drivers. In
Proc. of CHI, 3617-3626.
35. Rodden, T. A., Fischer, J. E., Pantidi, N., & Moran,
S. (2013). At home with agents: exploring attitudes
towards future smart energy infrastructures. In Proc. of
CHI, 1173-1182.
36. Sheridan, T. B., & Parasuraman, R. (2005). Human-
automation interaction. Reviews of human factors and
ergonomics, 1(1), 89-129.
37. Steiner, C., & Dixon, W. (2012). Automate this: How
algorithms came to rule our world. New York.
38. Strauss, A., & Corbin, J. M. (1990). Basics of
qualitative research: grounded theory procedures and
39. Suchman, L. A. (1983). Office procedure as practical
action: models of work and system design. TOIS, 1(4),
40. Tracy, S., Myers, K., Scott, C., (2007). Cracking jokes
and crafting selves: Sensemaking and identity
management among human service workers.
Communication Monographs, 283-308.
41. Uber (2014). a Deeper Look at Uber’s Dynamic Pricing
42. Uber (2014). How the Uber System Works [Video file]
43. Yang, R., & Newman, M. W. (2013). Learning from a
learning thermostat: lessons for intelligent systems for
the home. In Proc. of PUC, 93-102.
44. Walker, C. R. (1958). Life in the automatic factory.
Harvard Business Review,36(1), 111-119.
... In their review of algorithmic management, Kellogg et al. [21] highlight six mechanisms through which algorithms assert control over workers: directing workers by restricting and recommending, evaluating workers by recording and rating, and disciplining workers by rewarding and replacing. Further evidence illustrates that the unique features of algorithmic management such as persistent surveillance [22], continuous performance assessment [23], automated decision-making [12], little human-to-human contact [24], and poor transparency of algorithmic decisions [25] facilitate important power asymmetries between workers and management [26]. ...
Full-text available
In many countries, on-demand food delivery platforms (e.g. Deliveroo, Wolt, Uber Eats) have become an inseparable part of the hospitality and tourism ecosystem. A key area of interest in technology research has been how platforms algorithmically manage the interaction between task requesters (e.g. customers, tourists) and task fulfillers (e.g. restaurants and delivery couriers). However, there is a lack of research on how such algorithmic management practices impact workers and what strategies workers adopt to counteract the algorithm. To that end, this qualitative study explores forms of expressing algoactivism in the context of on-demand food delivery platforms by conducting interviews with delivery couriers (n = 5) and restaurant workers and managers (n = 7). It is found that both couriers and hospitality employees adopt specific behaviors to optimize and game the platforms’ algorithms, and that some algorithmic management practices are perceived more negatively than others. Implications for e-tourism management and research are discussed.
... The metrics that platforms collect from and about workers (e.g., their location and speed), gamification, and publication of work statistics are mechanisms of "techno-normative control" that lead to altered behaviors that deviate from the choices platform workers would make if they were working within a system that allowed truly autonomous decisions [14]. For example, the system monitors the rate at which workers accept orders and punishes workers who repeatedly decline work, including if they decline an order due to safety concerns [58]. Requiring workers to communicate with the platform company via the app is also a barrier to providing explanations for decisions and therefore is a mechanism to limit workers' ability to respond to problems [35,45,53]. ...
Full-text available
Algorithms are increasingly used instead of humans to perform core management functions, yet public health research on the implications of this phenomenon for worker health and well-being has not kept pace with these changing work arrangements. Algorithmic management has the potential to influence several dimensions of job quality with known links to worker health, including workload, income security, task significance, schedule stability, socioemotional rewards, interpersonal relations, decision authority, and organizational trust. To describe the ways algorithmic management may influence workers’ health, this review summarizes available literature from public health, sociology, management science, and human-computer interaction studies, highlighting the dimensions of job quality associated with work stress and occupational safety. We focus on the example of work for platform-based food and grocery delivery companies; these businesses are growing rapidly worldwide and their effects on workers and policies to address those effects have received significant attention. We conclude with a discussion of research challenges and needs, with the goal of understanding and addressing the effects of this increasingly used technology on worker health and health equity.
Purpose Food delivery platform work is a relatively new phenomenon in Finland and has not been studied widely hence limited knowledge on its work environment. The aim of this study was to explore and understand its psychosocial work environment in the Helsinki region and how it relates to the mental wellbeing of its couriers. Methods The study draws its findings from 20 semi-structured, in-depth interviews with food delivery platform workers in Helsinki. Data were approached through thematic analysis where the six phases of thematic analysis were meticulously followed. Results Food delivery platform work provided couriers with income and labour market opportunities. However, its work environment was psychosocially burdening, which posed detrimental challenges to the mental wellbeing of its couriers. Conclusion Study findings indicated that food delivery platform workers worked in an onerous work environment, which accentuated their occupational mental health. Thus, this study recommends future longitudinal research that would examine the association between food delivery platform work and mental health of couriers working through such platforms. Also, interventions and policies that aim at improving its psychosocial work environment are required for a more decent and healthier work environment that enhances mental health and wellbeing of its couriers.
This critical review of research on platform-mediated work argues that platform work studies are too focused on gig and remote work platforms. We introduce a framework that identifies three perspectives on how platforms reorganize work: narrow, broad, and systemic. This framework is used to examine the impact of platform-mediated work on four different aspects of work: management power, work processes, social protection and labor rights, and skills and career prospects.
Digital platforms and application software have changed how people work in a range of industries. Empirical studies of the gig economy have raised concerns about new systems of algorithmic management exercised over workers and how these alter the structural conditions of their work. Drawing on the republican literature, we offer a theoretical account of algorithmic domination and a framework for understanding how it can be applied to ride hail and food delivery services in the on-demand economy. We argue that certain algorithms can facilitate new relationships of domination by sustaining a socio-technical system in which the owners and managers of a company dominate workers. This analysis has implications for the growing use of algorithms throughout the gig economy and broader labor market.
Robots are transforming the nature of human work. Although human–robot collaborations can create newjobs and increase productivity, pundits often warn about how robots might replace humans at work andcreate mass unemployment. Despite these warnings, relatively little research has directly assessed howlaypeople react to robots in the workplace. Drawing from cognitive appraisal theory of stress, we suggestthat employees exposed to robots (either physically or psychologically) would report greater job insecurity.Six studies—including two pilot studies, an archival study across 185 U.S. metropolitan areas (Study 1), apreregistered experiment conducted in Singapore (Study 2), an experience-sampling study among engineersconducted in India (Study 3), and an online experiment (Study 4)—find that increased exposure to robotsleads to increased job insecurity. Study 3 also reveals that this robot-related job insecurity is in turnpositively associated with burnout and workplace incivility. Study 4 reveals that self-affirmation is apsychological intervention that might buffer the negative effects of robot-related job insecurity. Our findingshold across different cultures and industries, including industries not threatened by robots.
Purpose The author analyses the strategies developed by workers and unions to obtain representation and the successes and limitations of the strategies, in a context of platform work such as Spanish dominated by labour relations of employee workers. Design/methodology/approach The empirical material is the result of a series of in-depth interviews conducted between August 2020 and September 2021 with 41 workers, 15 of them union delegates, in addition to 4 union members and a labour lawyer. From these interviews, the author obtains a detailed account of the working conditions and the different phases that unionism has gone through in its objective of obtaining representation in a completely new sector. Findings The author found that employment in the relationship does not solve all the problems of platform work, especially those related to algorithmic control, but employment in the relationship provides advantages such as the right to representation. Workers play an important role in union strategies. Originality/value This study is the first in Spain, where platform work in passenger transport includes the employment relationship as a legal contracting mechanism.
Herrschaft bringt immer auch Widerständigkeit hervor. Demnach stellen Regelabweichungen, die sich aus unvollständig determiniertem Arbeitshandeln ergeben, ein strukturelles Merkmal im Arbeitsprozess dar. Die Formierung eines informellen Repertoires widerständiger Praktiken im Kontext betrieblicher Herrschaft ist dabei von der Arbeitssoziologie bisher vernachlässigt worden. Um diese konzeptionelle Leerstelle zu füllen, systematisieren die Beiträger*innen die Vielzahl der Praktiken und stellen verschiedene methodische, theoretische und empirische Perspektiven einer arbeitssoziologischen Widerstandsforschung vor.
Full-text available
Algorithms (particularly those embedded in search engines, social media platforms, recommendation systems, and information databases) play an increasingly important role in selecting what information is considered most relevant to us, a crucial feature of our participation in public life. As we have embraced computational tools as our primary media of expression, we are subjecting human discourse and knowledge to the procedural logics that undergird computation. What we need is an interrogation of algorithms as a key feature of our information ecosystem, and of the cultural forms emerging in their shadows, with a close attention to where and in what ways the introduction of algorithms into human knowledge practices may have political ramifications. This essay is a conceptual map to do just that. It proposes a sociological analysis that does not conceive of algorithms as abstract, technical achievements, but suggests how to unpack the warm human and institutional choices that lie behind them, to see how algorithms are called into being by, enlisted as part of, and negotiated around collective efforts to know and be known.
Conference Paper
Full-text available
Future energy systems that rely on renewable energy may bring about a radical shift in how we use energy in our homes. We developed and prototyped a future scenario with highly variable, real-time electricity prices due to a grid that mainly relies on renewables. We designed and deployed an agent-based interactive system that enables users to effectively operate the washing machine in this scenario. The system is used to book timeslots of washing machine use so that the agent can help to minimize the cost of a wash by charging a battery at times when electricity is cheap. We carried out a deployment in 10 households in order to uncover the socio-technical challenges around integrating new technologies into everyday routines. The findings reveal tensions that arise when deploying a rationalistic system to manage contingently and socially organized domestic practices. We discuss the trade-offs between utility and convenience inherent in smart grid applications; and illustrate how certain design choices position applications along this spectrum.
Full-text available
This paper examines how an occupational group has adapted to the demands of working with a Location Based Service (LBS). Instead of following a rigid timetable, London's bus drivers are now required to maintain an equal distance between the bus in front and the one behind. Our qualitative study employs ethnographic fieldwork and in-depth semi-structured interviews to elicit drivers' perspectives of the new system and show how it has modified their driving and general work conditions. We explore how passengers influence the movement of the bus and how the technology frames bus drivers' relationships to their managers and commuters. This work contributes to our understanding of the impact of LBS in the workplace and shows how technological imperatives can be established that cause unanticipated consequences and gradually undermine human relationships.
A growing concern for organizations and groups has been to augment their knowledge and expertise. One such augmentation is to provide an organizational memory, some record of the organization's knowledge. However, relatively little is known about how computer systems might enhance organizational, group, or community memory. This article presents Answer Garden, a system for growing organizational memory. The article describes the system and its underlying implementation. It then presents findings from a field study of Answer Garden. The article discusses the usage data and qualitative evaluations from the field study, and then draws a set of lessons for next-generation organizational memory systems.
This paper contrasts measures of teacher effectiveness with the students’ evaluations for the same teachers using administrative data from Bocconi University. The effectiveness measures are estimated by comparing the performance in follow-on coursework of students who are randomly assigned to teachers. We find that teacher quality matters substantially and that our measure of effectiveness is negatively correlated with the students’ evaluations of professors. A simple theory rationalizes this result under the assumption that students evaluate professors based on their realized utility, an assumption that is supported by additional evidence that the evaluations respond to meteorological conditions.