Content uploaded by Mrunal Gawade
Author content
All content in this area was uploaded by Mrunal Gawade
Content may be subject to copyright.
Exploring employment opportunities through
microtasks via cybercafes
Mrunal Gawade1, Rajan Vaish2, Mercy Nduta Waihumbu, James Davis2
1Department of Computer Science
Centrum Wiskunde & Informatica
Amsterdam, Netherlands
gawade@cwi.nl
2Department of Computer Science
University of California
Santa Cruz, CA, USA
rvaish@ucsc.edu, davis@cs.ucsc.edu
Abstract—Microwork in cybercafés is a promising tool for
poverty alleviation. For those who cannot afford a computer,
cybercafés can serve as a simple payment channel and as a
platform to work. However, there are questions about whether
workers are interested in working in cybercafés, whether
cybercafé owners are willing to host such a set up, and whether
workers are skilled enough to earn an acceptable pay rate? We
designed experiments in internet/cyber cafes in India and Kenya
to investigate these issues. We also investigated whether
computers make workers more productive than mobile
platforms? In surveys, we found that 99% of the users wanted to
continue with the experiment in cybercafé, while 8 of 9 cybercafé
owners showed interest to host this experiment. User typing speed
was adequate to earn a pay rate comparable to their existing
wages, and the fastest workers were approximately twice as
productive using a computer platform.
Keywords- Human Computation, Crowdsourcing, Microwork,
Cybercafés, Amazon Mechanical Turk (MTurk), ICT4D, India,
Kenya.
I. INTRODUCTION
Unemployment and underemployment are one of the most
pressing problems in society. Over the last half century, the
global literacy rate has risen faster than the employment rate.
This implies the availability of huge human resources, if they
can be reached efficiently.
Business Process Outsourcing (BPO) has a global reach
and provides employment, but rarely reaches the poorest
populations, or those in remote villages. There have been
explicit attempts to extend its reach into rural areas [1];
however these initiatives are capital intensive, and lack the
current extent of cyber cafes worldwide.
Crowdsourcing is an industry which might eventually
provide income in even the most remote locations [2]. It is in
its early stage, and is evolving quickly [3] [4]. The market
demand for crowd-sourced work quintupled in 2010 & almost
quadrupled in 2011[5]. One of the first platforms to implement
the concept of paid crowdsourcing was Amazon Mechanical
Turk (MTurk). MTurk is a workplace for offering low skilled
computerized jobs [6] [7]. MTurk offers jobs which pay from
$0.01 up to tens of dollars depending on the complexity of job.
Examples include work such as data entry, audio and video
transcription, and image labeling. Many of these microtasks
are of the sort that humans do easily, but are challenging for
computer algorithms [8] [9].
However, the use of this income channel has not been very
widespread among the poor. A study by Khanna et al.
concluded that lack of a suitable user interface was a barrier
[10]. When the interface was improved, work could be
completed, but still did not result in an acceptable pay rate for
all workers. It is possible that despite its attractiveness, this is
not a feasible channel for development.
Our work investigates additional questions related to
determining why cybercafés have not yet become informal
work centers. Are workers interested in working in
cybercafés? Are cybercafé owners willing to host this model?
Perhaps the cafes are inconveniently located, the owners aren’t
willing, or this style of work is simply not interesting. Are the
workers skilled enough to earn an acceptable pay rate?
Perhaps they lack computer skills, cybercafé rental is too high
a cost, or payments are simply too low?
To find answers to these questions, we developed a test
application and deployed it in cybercafés in India and Kenya.
Workers were paid directly by cybercafé operators, using our
funds. Payment was for words typed correctly, since
transcription is a common task. In addition to surveys and
records kept by café owners, the application logged user
activity so that we could measure statistics of interest such as
repeat visits and typing speed. We found that café owners and
potential workers were overwhelmingly positive about the
idea, and typing logs suggest skills at a level necessary for
adequate pay.
Some organizations and researchers have targeted work to
mobile devices, reasoning that potential workers already have
access to this technology platform, and thus don’t need to
obtain access to a computer at all [11]. This raises one
additional question. What is the relative productivity of
workers on mobile devices versus computers?
To answer this question we performed a user study with a
low income population in Kenya to explicitly compare typing
speed on mobile devices and computers. We found that the
most skilled users were substantially more productive using
computers.
The primary contributions of this paper are the results from
our studies, suggesting that workers and owners are interested,
that low income workers can be skilled enough to earn
acceptable wages, and that computers provide a more efficient
platform for work, even among the poor.
II. RELATED WORK
The primary existing labor channel for development
oriented digital work is through employee-employer
relationships similar to traditional BPO operations. For
example, Samasource [12] assigns microwork to people at the
bottom of the pyramid through partnerships with NGOs, where
NGOs train the workers and provide the necessary resources to
work. Similarly, CloudFactory trains its own workforce, and
village BPO operators explicitly follow that model [1]. This
paper explores whether the poor can potentially be reached
using cybercafés as informally organized work centers.
Many organizations have attempted to deliver microwork
through mobile devices, since the penetration of mobile far
exceeds that of computers. Examples include MobileWorks,
mClerk, and txtEagle [11] [13] [6]. This paper compares the
relative efficiency of mobile and computer based data entry.
Attempts to understand the impact and dynamics of
cybercafés as work centers have been studied [14]. In 2003,
Mutula studied the origin, challenges and growth of the
cybercafé industry in Africa [15]. Cybercafés were used to
train low-income minority seniors in Los Angeles to cross the
digital divide [16]. Overnight use of community centers for
employment was investigated in Nigeria [17]. This paper
investigates the use of cybercafés specifically as potential
centers for informal microwork.
Research comparing input technologies and relative typing
speed between devices is common [18] [19] [20]. In the context
of development, voice versus typed input has been studied for
gathering information from low literacy users [21] [22] [23]
[24]. This paper compares typing speed on mobile devices and
computers in a low-skill population.
III. THE EXPERIMENTAL SETUP
Experiments in cybercafés in India and Kenya were
conducted on an application specifically designed to test the
typing speed, and thus earning potential, of workers. Figure 1
is a screen shot of the application where users type the falling
words. Money earned for correctly typed words is displayed.
Words can be typed in any order and are no longer available
when they reached the bottom of the screen.
The application was developed in Adobe Flash 10, and
saved user logs locally in XML since the workers performed
their work unsupervised. An example log file is shown in
figure 2. The dictionary of words was in English.
In India, the payment rate was set at $0.0011/word or $0.01
per 9 words. The words were generated every 1.5 seconds,
totaling 40 words per minute. At this rate a perfect user would
earn a maximum of $2.64 an hour. In Kenya we experimented
with a lower skill population and changed payment rate to
$0.01 per 3 words. Monetary values were always discussed in
local currency, but are reported here in US Dollars. We use an
exchange rate of 50 INR to 1 USD, and 100 KSH to 1 USD
throughout.
Figure 1. A screen shot of the application developed to test typing skills of
low income users in cybercafés.
We also collected surveys from workers and owners, and
had cybercafé staff report all earnings in a log book. Money
was provided to café owners and distributed to workers at the
time work was completed. The paper and digital logs were
cross checked, and we did not observe any unexplained
discrepancies.
Figure 2. Sample data collected from an auto generated log file.
IV. EXPERIMENTS IN INDIA AND KENYA
A. India
We deployed our application in three cybercafés of Pune,
Maharashtra, over a period of six months. The locations of the
cybercafés were chosen after informally surveying locality,
size of cybercafés, number of users visiting cybercafés, hourly
fees, and owner interest.
Locality was one of the primary factors in decision making.
The first net cafe was chosen in a relatively affluent and
residential locality of Aundh in Pune, pictured in figure 3. The
second cybercafé was chosen in the midst of one of the busiest
roads surrounded by a number of educational institutes on the
J.M. Road in Pune. The third net cafe was chosen next to the
University of Pune, with a mix of college going and job seeker
crowd. The users were trained via a “walkthrough power point
presentation” and/or cybercafé owners. After that, the users
were welcome to work as long and often as they wished,
subject to a maximum daily earning of $2.
Table I shows the demographic data for users in India. The
income and education levels were higher than we intended,
including primarily college students. In order to validate that
our findings generalize, we repeated the experiment in a lower
income settlement in a second country (Kibera, Kenya).
We also conducted a second survey of café owners in
Bhubaneshwar, Orissa, aimed at judging both the interest of
owners and their financial ability to participate.
Figure 3. Interior of a cybercafé in Aundh, Pune. We placed an advertisement
poster to attract cybercafé customers to our experimental employment
opportunity.
TABLE I. DEMOGRAPHICS OF SUBJECTS
INDIA AND KENYA
B. Kenya
The experiments in Kenya were conducted during a period
of 15 days. The application was deployed in two cybercafés in
Kibera, a slum in Nairobi. The café owners handled payment,
just as in Pune, but were not asked to keep records of users or
payments. Figure 4 shows the external view while figure 5
shows the interior of one of the cybercafes, with a worker
using our application.
In addition to the test application, a user study was
performed to compare typing speed on basic mobile devices,
mobiles with mini-keyboards, and computers. Twenty three
users were given 3 minutes on each device to transcribe as
many words as possible from a sheet of paper. These users
were also surveyed for income, education, and computer
knowledge.
Table I shows the demographics of our user study
population in Kenya. While most people had a high-school
education, far fewer had college education, and income was
substantially lower. Informally, we observed that the youth
who hang out at cybercafes had good english language skills
and moderate computer skills. We drew participants for the
user study from both the cybercafés, and the wider
community. We obseved that the older population in Kibera
frequently did not speak english, and had frequently never
used a computer. All participants were clearly very familiar
with their own basic mobile phone.
Figure 4. Exterior of a cybercafé in Kibera, Kenya.
.
Figure 5. Interior of a cybercafé in Kibera, Kenya. A worker is using our test
application on a laptop owned by the cybercafé.
V. RESULTS
A. Are workers interested in working in cybercafés?
Of 105 potential workers surveyed in Pune and 30 in
Nairobi, 99% stated that they would like to do microwork in
cybercafés. The survey result is backed up by informal
anecdotal observation. In Kibera, competition was observed
for the one machine which had the software installed. In
addition, the café owners reported reserving a portion of the
work for themselves, indicating they thought a meaningful
income was available. Lastly, after terminating the experiment
some café owners and workers in both locations asked
repeatedly to continue the program.
In order to control for participant response bias [25], we
also tracked the actual behavior of users when no
experimenters were present. Figure 6 shows the frequency of
returning users in the residential area of Aundh, Pune. In this
location, the experiment spanned the time frame from June to
August, long enough for novelty effects to diminish. Blue dots
represent returning users, while red dots represent users who
tried the application only once. Approximately 50% of the
participants returned to use the application, with the maximum
being 11 times. Workers sometimes spent up to 4 hours
working, with up to 1 hour being quite common.
Figure 6. Frequency of returning users at a cybercafé in a residential area of
Pune, over a two month period. Blue dots are workers who returned for
employment multiple times, while red dots are workers who tried our test
application only once. Notice that many workers did return.
We observed very few return users at the Pune cafes
located in city center and near the university, although one of
those cafes attracted the most total participants (60 users). The
Nairobi experiment was too short to measure returning users.
We hypothesize that a residential location allowed the café to
attracted regular customers who could conveniently return to
earn extra money.
Based on the combined evidence of surveys, informal
discussions, and the fact that some workers returned multiple
times and worked long hours, the authors believe that at least
some potential workers actually do consider this a desirable
income source.
B. Are Cybercafé owners willing to participate in this
initiative?
The three upscale café owners in Pune were directly
compensated, to incentivize them to participate. The two café
owners in Kibera enthusiastically participated with no extra
compensation. To explore whether other cybercafés would be
interested, we surveyed 9 owners in Bhubaneshwar, Orissa.
Orissa is one of the least developed states of India. We
gathered data on interest, as well as café income, capital costs,
and customer base. Some of this data is presented in Table II.
We found that 8 of 9 cybercafé owners showed interest in
implementing this model. The cafes had a suitably sized
customer base, ranging from 20 to 300, mostly in the 15-30
year old age range. All the owners had a diploma or a graduate
degree, which suggests capacity to train workers and manage
payment if needed. The internet tariff ranged from $0.20 to
$0.40 per hour, while the bandwidth ranged from 512Kbps to
5Mbps.
We also asked whether owners would be willing to
consider night operations or expanding their café with
additional capital outlay. The responses to these questions
were less enthusiastic, with slightly less than half agreeing that
they could consider those options.
TABLE II. SURVEY RESULTS FROM THE CYBERCAFES OF
BHUBANESHWAR, INDIA
In order to at least informally check for participant
response bias among owners, we also surveyed business
owners in an unrelated sector. We asked six owners of water
retail points in Kibera to consider our innovative business
models such as delivering water to customers, or purifying it
prior to sell. All of these owners declined these changes,
always providing a reason why it did not make business sense.
Given the combined evidence that some owners
participated in our microwork experiment for months, that
additional surveyed owners we positive, and that owners in
other business sectors were willing to state non-interest, the
authors believe that at least some cybercafés would be
interested to serve as work and payment distribution points.
C. Are the workers skilled enough to earn an acceptable pay
rate?
To determine the efficiency of workers, we analyzed the
data collected by our test application. Figure 7 plots the actual
time spent and money earned by our participants. There were
both fast and slow workers. Some stopped after only a few
minutes and some worked for over an hour. A few continued
until the reached our maximum daily income of $2. Notice that
most workers earned in the $0.50-$1.75 per hour range.
Figure 7. Analysis of time spent and money earned by actual users. Most
workers earned between $0.50-$1.75 per hour. Notice that some workers sat
and used our application for more than 2 hours, even when earning relatively
little.
The measured actual earnings are dependent on our
arbitrarily set pay rate of $0.0011/word in India and
$0.0033/word in Kenya. We used these rates because the
going rate in summer 2011 for transcribing 1000 CAPTCHAs
was between $1 to $10. Our pay rate in India was near the low
end of this range, while in Kenya we paid near the median.
Gupta et al. reported commercial transcription costs in India in
the $0.004-$0.01/word range, so we believe we paid
appropriately [13].
Since typing speed is directly correlated with earnings and
independent of our arbitrary payment choices, we plot this
directly in figure 8. This report is for total words over total
time, including distracted workers who used another
application while leaving our test application open in the
background. Thus burst typing speed was higher than reported
here. The relatively more educated workers in India averaged
18.9 words per minute, while those in Kibera averaged 7.4
words per minute. In both cases there was a range of typing
speeds.
Assuming the measured typing speeds and a pay rate of
$0.002/word, at 8hrs/day we obtain pay rates in the $6-10/day
range in Kibera, and $10-20/day in Pune. Equivalently, about
$150-$300/month in Kibera, and $200-$600/month in Pune.
This pay rate is comparable to the existing reported incomes of
our workers. The fastest workers earned more than this, and
we believe that with daily practice all the workers would
improve their skills, and thus their earnings.
The authors believe that despite relatively slow typing
speeds, especially in Kibera, computer skills are good enough
that digital microwork provides a plausible income source.
Figure 8. Typing speed of workers using our test application in cybercafés.
The relatively more educated workers in Pune typed faster than the workers in
Kibera. Notice that there is a large variation in typing skill in both locations.
D. What is the relative productivity of workers on mobile
devices versus computers?
Since many organizations are targeting microwork to
mobile devices, we conducted a separate user study in Kibera
measuring typing speed. Twenty three users were asked to
transcribe as much as possible in 3 minutes from a sheet of
paper, using each of: a basic mobile phone, a Nokia N900 with
mini-QWERTY keyboard, and a netbook computer, shown in
figure 9.
Results are plotted in figure 10, sorted by speed of
participant. The user IDs are sorted independently per device.
Some users were faster on basic mobile than computer.
However, on average, the least technologically literate half of
the users were approximately the same speed on mobile and
computer. In contrast the fastest user was almost twice as fast
on a computer. The average typing speed for N95, N900 and
Net book was 6.8, 7.3 and 9.4 words per minute respectively.
Figure 9. Nokia N900, Netbook, and Nokia N95 that were used for a user
study of typing speed on different platforms.
We brought a Nokia N95 with us, but most users owned
and all knew how to use a basic mobile already. We tested
basic mobile text entry on the users own phone when possible.
Some users had to be taught how to use the mini-QWERTY
and computer keyboards. Thus we expect that practice would
increase these speeds more than the basic mobile speed.
Figure 10. Comparative analysis of transcription speed in a 3-minute test,
using basic mobile input (N95), mini-keyboard (N900) and netbook. The
slower half of the users were approximately the same speed on all devices,
while the fastest users were noticeably faster on a computer.
The additional average 2.6 words/minute possible on a
computer translates to about $2.50 per work day, assuming
$0.002/word. Assuming that cybercafé owners kept 50% of
this extra income and were open two shifts a day, the capital
cost of a low end $300 computer would be repaid in 4 months.
Based on the measured typing speeds, the capital cost of
computers and the expectation that practice would improve the
interface advantage of computers further, the authors conclude
that cybercafés offer a cost effective alternative to mobile data
entry for populations that cannot afford personal computers.
VI. CONCLUSIONS
Digital microwork using cybercafés as informal work
centers is a promising income source for the poor. Since this
practice is not yet widespread, there must be some factor
preventing adoption. This paper has explored several
possibilities: worker interest, owner interest, skill levels, and
platform alternatives.
We deployed a test application in five cafes in two
countries. In addition to usage statistics we surveyed both
workers and café owners about attitudes and income
demographics.
We conclude that the workers are interested, the café
owners are interested, and the workers are skilled enough to
earn acceptable wages.
In addition we compare worker efficiency on basic mobile
devices and computers, and conclude that computers provide
efficiency gains, even among low skill, low income subjects.
REFERENCES
[1] S. Malhotra, P. Rathi, T. A. Gonsalves, A. Jhunjhunwala, T. Giri,
“Untapped Resources! Amazing Potential! TeNeT’s Rural BPO
Initiative”, in IPIP International Federation for Information Processing,
Volume 241, Home Informatics and Telematics: ICT for the Next
Billion, Springer 2007, pp. 21–33.
[2] W. Thies, A. Ratan. J. Davis. “Paid Crowdsourcing as a Vehicle for
Global Development”, ACM CHI 2011 Workshop on Crowdsourcing
and Human Computation.
[3] B. Frei, “Paid Crowdsourcing: Current State & Progress towards
Mainstream Business Use”. Smartsheet White Paper, Sep 2009.
[4] I. Medhi, G. Menon, and K. Toyama, “Challenges of computerized job-
search in the developing world”. CHI, 2008.
[5] BUSINESSGROW, 2011, Retrieved April 2012.
http://www.businessesgrow.com/2011/12/13/enterprise-crowdsourcing-
blasts-off-as-social-media-growth-industry/
[6] N. Eagle, “txteagle: Mobile Crowdsourcing”, In Internationalization,
Design and Global Development, 2009.
[7] S. Chen, M. Ravallion, “The Developing World Is Poorer Than We
Thought, But No Less Successful in the Fight against Poverty”, Policy
Research Working Paper 4703, The World Bank, Development Research
Group, Aug 2008.
[8] J. Ross, L. Irani, M. S. Silberman, A. Zaldivar, and B. Tomlinson, “Who
are the Crowdworkers? Shifting Demographics in Amazon Mechanical
Turk”, In alt.CHI, 2010
[9] J. Zittrain, “Ubiquitous human computing”, Philosophical Transactions
of the Royal Society, 366, 2008.
[10] S. Khanna, A. Ratan, J. Davis. W. Thies, “Evaluating and Improving the
Usability of Mechanical Turk for Low-Income Workers in India”, ACM
DEV 2010.
[11] P. Narula, P. Gutheim, D. Rolnitzky, A. Kulkarni, B. Hartmann, 2011,
“MobileWorks: A Mobile Crowdsourcing Platform for Workers at the
Bottom of the Pyramid”, 3rd Human Computation Workshop (HCOMP
2011).
[12] SAMASOURCE, 2012, Retrieved April 2012. http://samasource.org/
[13] A. Gupta, W. Thies, E. Cutrell, R. Balakrishnan, “mClerk: Enabling
mobile crowdsourcing in developing regions”, In Proceedings of CHI
2012, Human Factors in Computing Systems, (Austin), ACM press.
[14] S. Liff, F. Steward, “Shaping e-Access in the Cybercafé: Networks,
Boundaries and Heterotopian Innovation”, New Media & Society
September 2003 vol. 5 no. 3 313-334.
[15] S. M. Mutula, “Cyber café industry in Africa”, Journal of Information
Science November 2003 vol. 29 no. 6 489-497
[16] Y. Jung, W. Peng, M. Moran, S. A. Jin, M. McLaughlin, M. Cody, M. J.
Marsh, J. Albright, M. Silverstein, “Low-Income Minority Seniors'
Enrollment in a Cybercafé: Psychological Barriers to Crossing the
Digital Divide”, Taylor & Francis, Educational Gerontology, Volume
36, Issue 3, 2010, pp.193-212.
[17] E.E. Adomi, “Overnight Internet Browsing Among Cyber Café users in
Abraka, Nigeria”, The Journal of Community Informatics, Vol 3, No 2
(2007).
[18] K. Lyons, T. starner, D. Plaisted, J. Fusia, A. Lyons, A. Drew, E. W.
Looney, “Twiddler typing: one-handed chording text entry for mobile
phones”, In Proceedings of the SIGCHI conference on Human factors in
computing systems, CHI 2004.
[19] N. Green, J. Kruger, C. Faldu, R. S. Amant, “A reduced QWERTY
keyboard for mobile text entry.”, In CHI 2004 extended abstracts on
Human factors in computing systems.
[20] E. Matias, S. MacKenzie, W. Buxton, “One-handed touch typing on a
QWERTY keyboard.” Hum.-Comput. Interact. 11, 1 (March 1996), 1-
27.
[21] N. Patel, S. Agarwal, N. Rajput, A. Nanavati, P. Dave, T. Parikh, “A
Comparative Study of Speech and Dialed Input Voice Interfaces in Rural
India”, In Proceedings of ACM Conference on Human Factors in
Computing Systems, CHI 2009.
[22] J. Sherwani, S. Palijo, S. Mirza, T. Ahmed, N. Ali, R. Rosenfeld,
“Speech vs. Touch-tone: Telephony Interfaces for Information Access
by Low Literate Users.” In Proc. Information & Communications
Technologies and Development, Doha, Qatar, April 2009.
[23] S. Patnaik, E. Brunskill, W. Thies, “Evaluating the accuracy of data
collection on mobile phones: a study of forms, sms, and voice”, In the
Proceedings of the 3rd international conference on Information and
communication technologies and development, ICTD 2009.
[24] I. Medhi, S. Patnaik, E. Brunskill, S. N. N. Gautama, W. Thies, K.
Toyama, “Designing mobile interfaces for novice and low-literacy
users”, ACM Transactions on Computer-Human Interaction, Volume 18
Issue 1, April 2011.
[25] N. Dell, V. Vaidyanathan, I. Medhi, E. Cutrell, W. Thies, “"Yours is
Better!" Participant Response Bias in HCI”, the ACM SIGCHI
Conference on Human Factors in Computing Systems 2012.
