AI DATA
LABELLING
ROUNDTABLE READBACK
- B U I L D I N G E T H I C A L A R T I F I C I A L I N T E L L I G E N C E T O G E T H E R -
This readback is based on a roundtable on AI Data Labelling, held on the 16th of June 2020,
hosted by Aapti Institute and Dr. Alex Taylor, City, University of London. This is part of a larger
research collaboration that intends to shed light on labelling and annotation processes, and
collaboratively envision best practices for fair and equitable Artificial Intelligence. This exploratory
discussion brought together startups, researchers and civil society leaders to unpack the processes
and structures that are involved in data labelling or annotation work.
The COVID-19 pandemic and lockdown are shifting the structure of markets and nature of available
jobs, well-beyond the present moment. With these immediate changes and longer-term trends, it
would seem timely to revisit and explore the state of technologically-mediated labour, where work is
both found and delivered online. Also presenting pressing questions is the increasing promotion of AI
in emerging and evolving forms of contemporary work. AI and machine learning (ML) are being
touted by multinational technology firms as the backbone to future visions of work, particularly
remote work. Yet many questions remain about the specific role of computational systems in these
new forms of work and the shape such AI and ML systems should take. Furthermore, we are having
to confront longstanding and in some cases new questions about automation in the workplace and
what we imagine to be not just productive but fair and just ideas of labour.
It was with this as a backdrop that the roundtable we report on below was held. Rather than attend
to the oft-hyped, idealised visions of work, however, we opted to pay attention to the work ‘behind
the scenes’, conducted to make AI/ML work. Specifically, we wanted the primary focus of the
roundtable to be on a labour that is rarely discussed and often invisible to end users, data labelling or
annotation. Our interest was (and remains) in how data labelling or annotation plays a critical role in
the success of the future visions of work, and what practices and structures are necessary to achieve
such a success. We thus wanted to promote a discussion to better understand the conditions of this
labour and both the social and technical issues that arise in enabling it.
Annotation and labelling for Artificial Intelligence (AI) offer employment opportunities for many.
Across multiple sectors including e-commerce, autonomous vehicles and healthcare, AI is being
promoted as an innovative solution to provide analytics, automate processes and evolve business
models. However, AI is dependent on the availability of labelled and classified datasets. Without
necessary meanings attached to data, machine learning models cannot be trained or function in the
real-world. This critical role is played by humans who have the skills and know-how to classify new
data sets, and approach problems of uncertainty and ambiguity in data and context in ways that
computer models find notoriously hard to replicate.
To carry out these tasks, crowdsourcing platforms and startups often employ thousands of workers
to label the datasets of text, images, video, etc. Previous work, including from one of our roundtable
participants, Mary Gray, has shown the difficult labour conditions that these workers can face,
surfacing not only the hidden nature of this work, but its potential for low wages, exploitation and
abuse. It is clear that pathways for improving labour conditions and regulation need to be set out and
where possible used to help inform national public and organisational policy programmes. In order for
us to do this, we need a fuller understanding and more systematic representation of the current
business models, the nature of work, and concerns of workers. Building and developing regulation for
more equitable AI depends on the recognition and deeper understanding of this labor. As the future
of work becomes more digital, it is imperative to understand how this work is done, who does it, and
how it can be structured and performed to enhance dignity and protect rights, while enabling
innovation.
A I D A T A L A B E L L I N G
The Roundtable was divided into two sections, the first featured presentations from three startups:
Playment, TaskMonk and iMerit. As limited information exists on the state of AI labelling, startups
provided detailed insight into the industry and current processes. The second half provided time for
researchers to discuss their work and explore areas for future research collaboration. This also
enabled startups to engage in discussion around research, policy and upcoming trends.
M E T H O D O L O G Y
SPEAKERS
MODERATORS
Startups like Playment have emerged in response to the inadequacies such as the lack of consistency in
quality from legacy platforms likea Amazon Mechanical Turk. Playment seeks to democratize the
availability of labelled data for other developers and machine learning engineers. Previously dominated
by big tech, many companies have started to build their own models, open-source tools and platforms
for annotation. Over the past decade, the nature of labelling and annotation work has evolved in
complexity, type of work and expertise required. While initially labelling and annotation often involved
straightforward tasks such as image classification, the work has now evolved requiring the annotation
of for example 3D models, or medical imagery and scans. Though the crowd-work model is still
employed, startups are leaning towards BPOs and impact sourcing firms, or in-house staff to build a
secure and consistent pool of labellers that possess the requisite knowledge and skills to carry out
tasks.
During the first half of the roundtable, the three startups listed below described their efforts to
respond to this evolving landscape and how they have chosen to structure/implement their labelling
and annotation work. They also discussed some of the challenges faced with their approaches.
iMerit leverages technology and skill development to engage women and youth from newly digital
communities and develop them into expert technology workers. This advanced workforce labels, enriches,
augments and audits data to shape algorithms. iMerit has over 2500 fulltime data experts with 52%
women and over 80% employees from under-resourced backgrounds, with an average age of under 25.
iMerit works with leaders across sectors like Autonomous Vehicles, AgTech, Medical Imaging, e-Commerce
and Financial
Services.
Playment's mission is to expedite the AI age by availing large quantities of high quality labeled
datasets to ML teams. Playment provides software and services for data labeling in computer vision
space, has served 200 customers across 12 countries with more than 10,000 crowd-workers and
BPO workers in India. Mercedes, Samsung, Intel, Ford and Sony are some of their key customers.
TaskMonk aims to change the way annotated data is being procured by enterprises for training AI /
ML Models. Within a year of launch Taskmonk has scaled to deliver million/tasks per month for its
customers serving process involving Text, Image, Video, Audio, Sensor and Geo-Spatial data.
PLAYMENT
TASKMONK
IMERIT
PART 1: Connecting with the Startup Ecosystem
PART 1: Connecting with the Startup Ecosystem
Playment labelers are typically housewives, recent
graduates and retired people/veterans
iMerit has an inclusive hiring policy – 50% of their
employees are women and 80% of their workforce
comes from disadvantaged communities defined
based on respective communities.
TaskMonk’s annotation partners and BPOs engage
workers from Tier II & Tier III cities in India
“We don't have criteria, we just need a skill set,
and an aptitude test that a person needs to pass.
If they passed, no matter which department
qualification, age, sex, gender, we will accept
them. Of course, there is definitely a working
minimum of 18 years of age.”
- Ajinkya Malasane, Playment
Playment adopts a hybrid crowdsource and BPO sourcing model, where labelers are onboarded if they meet
basic requirements and pass an aptitude test.
iMerit structures work through a formal, full-time employment contract and labelers are often hired from
poor and marginalized communities.
TaskMonk has created a technology platform in which their clients can carry out in-house labelling using tools
or by onboarding annotation partners.
HOW IS WORK STRUCTURED AND WHAT PROCESSES ARE INVOLVED?
Data Labelling work is typically structured through crowd-work, outsourced to
registered BPOs or carried out in-house with full-time employees.
“If you want to do annotation work or
computer vision, or a text transcription or
text translation, we use a lot of the cloud
models that are already available as a
starting point and the annotation partners
become the checker for that”
- Sampath Herga, TaskMonk
Automation is used to allocate tasks to labellers and provide them with guidance in
the annotation process
WHO ARE THE LABELLERS?
The demographic profile of data annotators
and labellers vary across platforms.
Playment's platform allocates tasks to labellers based on
their degree of experience and performance ranking
TaskMonk allocates tasks based on the workers’ degree of
familiarity, which also enables them to gain more domain or
subject matter expertise. Alternatively, a more complex
task is split up into chunks, allocated to different workers
and finally stitched together once complete.
"Quantity, diversity and quality of data is what matters
for machine learning engineers"
- A J I N K Y A M A L A S A N E , P l a y m e n t
PART 1: Connecting with the Startup Ecosystem
Consensus Based Model
This entails inserting pre-annotated images in the labelling set. If labellers annotate these images
correctly, it serves as a basis for how the rest of the labelled data is assessed for quality. Playment
argues that this model is flawed as it is not representative or indicative of high quality labelling
processes.
Gold Standard
In this model multiple annotators carry out the same labelling task, which is more effective and
arguably produces better quality output. However, given that it requires several annotators, it is both
more expensive and time consuming to employ as a standard.
Maker/Editor/Checker Logic Models
This model is employed by Playment and ranks annotators based on degree of expertise and skill set.
Based on their proficiency with tasks, they are either classified as ‘Makers’, ‘Editors’ or ‘Checkers’.
Checkers tend to possess the most experience and are required to verify the work of Makers.
Checkers report to their respective project managers. Based on the work required, Editors can be
brought in to supplement the work of checkers. It combines the value of the Gold Standard as tasks
are reviewed by at least 4-6 workers. Workers who are labelled as checkers are able to build their
familiarity and expertise in annotation.
Quality of labelled data can be improved by standards and by
building better and smarter tools for annotators. Some
startups like Playment and TaskMonk use a series of
processes and quality standards to address this. Consensus
Based, Gold Standard & 'Maker/Editor/Checker' Logic Models
are three approaches for ensuring quality of data is
maintained through the labelling process.
WHAT ARE THE CHALLENGES?
“For a data labeling provider the biggest problem
to solve is quality.” - AJ Malasane, Playment
“Most of the quality requirements are a little
higher than what it was earlier. Most of the low
hanging fruit is done. We need quality work and
data annotation which works well with the
managed services model”-
Sampath Herga, TaskMonk
"It's a unique problem of building software and enabling human
work in tandem, which makes it neither a software industry nor a
complete services industry. It's a hybrid"
- A J I N K Y A M A L A S A N E , P l a y m e n t
PART 1: Connecting with the Startup Ecosystem
Workers that demonstrate high learnability can be guided
and their capacity can be built effectively, even if they
possess limited formal education and training.
and retaining data labellers and annotators while considering the evolving
requirements and complexity of tasks
“We are having to build a people architecture that
stays one step ahead of the algorithm. This means
being able to do increasingly complex work, and
being able to do increasingly nuanced work and
being able to provide insight to the makers of the
algorithms.”- Jai Natarajan, iMerit
“A lot of people have what we might call street
smarts or learnability. When they're presented with
the right curriculum in context, they're able to
convert that into high data labeling quality.”
- Jai Natarajan, iMerit
RESPONSE FROM RESEARCHERS & SCHOLARS
“As the complexity of the task increases, the complexity of the
annotations required increase, you'd need a little bit more
expertise than just throwing it at an unknown crowd. So, skilling
and training is a very important part.”-
Kalika Bali, Microsoft Research
There is often a hierarchy of expertise in the types of
tasks required, some of this complexity can be overcome
by skilling & training.
2. Uniqueness & Complexity of Tasks
To address new labelling demands and requirements from the
industry such as labelling data in 3D models for Augmented
Reality/Virtual Reality functions, startups either adopt a tools
agnostic approach combined with a focus on dynamic training
(iMerit) or prioritize the creation of smarter tools to enable
workers with limited training (Playment).
“No two projects and jobs and data labeling activities
are similar. I call this the ‘how to ground truth, the
ground truth’ - there’s no solution available on the
market”” - AJ Malasane, Playment
“I believe in making better (easier, efficient) tools to
do things at scale. Tough to train and expect from
10k people. The better the tool, the lesser the
variability”- AJ Malasane, Playment
“The difference between the person who uses a
basic image editing like what we have in
PowerPoint, versus somebody who knows all the
tricks of Photoshop, the shortcut, the magic wand
etc - that difference is substantial” - Jai Natarajan,
iMerit
RESPONSE FROM RESEARCHERS & SCHOLARS
“We have done some small data collection for Natural Language
Inference recently, and it is super difficult to even design the task,
let alone the issues of evaluation.”
- Kalika Bali, Microsoft Research
Labelling image/video data is very different from
labelling language or speech data for NLP models
SENIOR RESEARCHER, MICROSOFT | SAFRA CENTRE FOR ETHICS FELLOW | FACULTY AFFILIATE,
BERKMAN KLEIN CENTRE FOR INTERNET & SOCIETY, HARVARD UNIVERSITY | FACULTY - SCHOOL OF
INFORMATICS, COMPUTING & ENGINEERING, INDIANA UNIVERSITY
Mary, an anthropologist and media scholar by training, focuses on how everyday uses of technologies transform
people’s lives. At MSR, Mary works on the social impact of digital labor through the case of on-demand labor. In May
2019, with computer scientist Siddharth Suri, Mary published five years of collaborative research in the book, Ghost
Work: How to Stop Silicon Valley from Building a New Global Underclass published by HMH Books. Their research
team combined methods from anthropology and computer science, to amass the largest data set about on-demand
work ever collected. They conducted hundreds of in-person interviews and more than 10,000 surveys of workers in
the US and India.
MARY L. GRAY
PART 2: Exploring futures of work in AI Data Labelling
ASSOCIATE PROFESSOR, IIT HYDERABAD
Nimmi is an Associate Professor at the Kohli Centre on Intelligent Systems, Indian Institute of Information
Technology, Hyderabad. She brings an anthropological lens in understanding the impacts of AI research and praxis.
Nimmi is also an Adjunct Professor at the Indian Institute of Technology, IIT, Hyderabad where she teaches courses
at the intersections of society and technology. Previously, Nimmi was a senior research scientist and led the Human
Interactions research area at the Xerox Research Center India. Nimmi previously worked with Microsoft Research,
with a focus on a combination of theoretical analysis and ethnographic field research to understand technology use in
developing countries.
DR. NIMMI RANGASWAMY
SENIOR RESEARCHER, MICROSOFT RESEARCH
At MSR Vivek is broadly interested in designing efficient computer systems, and more recently, in developing
technologies for societal impact. One of his current areas of focus includes Project Karya, a new platform to provide
digital work to rural communities. The word “Karya” literally means “work” in a number of Indian languages. Project
Karya enables rural communities to participate in crowdsourcing. Vivek primarily works with the Technology for
Emerging Markets group at Microsoft Research India. He received his . in Computer Science from Carnegie
Mellon University. His . thesis was on new memory abstractions for efficient memory systems, with specific focus
on DRAM-based main memory.
DR. VIVEK SESHADRI
PRINCIPAL RESEARCHER, MICROSOFT RESEARCH
Kalika works in the areas of Machine Learning, Natural Language Systems and Applications, as well as Technology for
Emerging Markets. Her research interests lie broadly in the area of Speech and Language Technology especially in the
use of linguistic models for building technology that offers a more natural Human-Computer as well as Computer-
Mediated interactions. She is currently working on Project Mélange where she tries to understand, process and
generate Code-mixed language data for both text and speech. Code-mixing or use of more than one language in a
single conversation or utterance is a phenomenon that is observed in all multilingual societies.
DR. KALIKA BALI
RESEARCHERS & SCHOLARS:
PART 2: Exploring futures of work in AI Data Labelling
While crowd work and content moderation processes have been significantly documented, limited research exists on
what this may involve for data labelling processes, necessary for building AI. Often rendered invisible, the labour
behind these data labelling processes are more frequently being carried out by workers in the global south. These
experiences are further mediated by dimensions of class, gender, language proficiency and technical expertise. This
section provided researchers with the opportunity to share learnings from their research, explore avenues for
collaboration, and raise questions. While this section of discussions was limited by time constraints, key themes and
issues raised by researchers set the stage for further discussion and areas for future research.
LEARNINGS FROM EXISTING RESEARCH:
Prioritize and value the inclusion of women and people from underrepresented
and marginalized communities
KALIKA BALI
There are often assumptions around people who carry out platform based work and limited understanding
of how this work can improve or perpetuate existing patriarchal structures. Further research needs to be
done on gender and its relation with social capital
“Is it really true that women are able to take their own decisions because they're earning
money? Is it just that the money is just handed over to the head of the family and he takes all
the decisions? - Kalika Bali
Acknowledge that data labelling/annotation is not ‘click-work’ or a niche job
MARY GRAY
The processes, expertise, time and energy spent on these tasks should not be rendered invisible but instead
recognized so we can build better support systems.
“Imagining we are somehow reducing what it is that people are doing fails to see just how
incredibly cognitive cognitively challenging it is to be able to be constantly offering that very
human capacity for spontaneous responsive insight” - Mary Gray
PART 2: Exploring futures of work in AI Data Labelling
Re-orient workplace environments and consider what can be done differently to
strengthen well-being, conditions and the ecosystem in which workers operate
How can we re-shape the Future of Work away from white, neo-liberal conceptions of productivity and
accomplishment?
“What are ways in which we would recognize what are people trying to get out of these forms of
employment? How does it speak to them meeting needs that are not met in the way we try to formally
organize employment now?” - Mary Gray
Foster a culture of collaboration between workers both on and offline
Whether or not this is encouraged by platforms, workers typically find peers and form networks that they
then use to ask for help and collaborate.
“We accomplish what we accomplish in our day to day employment, through groups of people
coordinating and collaborating” - Mary Gray
Build relevant methodological approaches to enable ethnographic study on the
Future of Work
DR. NIMMI RANGASWAMY
From the perspective of anthropology, what is the methodology and approach to adopt when studying
automation and the impact of artificial intelligence?
Explore how digital work (annotation and labelling) can be made a more
accessible livelihood opportunity for rural communities
DR. VIVEK SESHADRI
Through Project Karya, Vivek works with rural communities who help label and digitize regional language
and speech sets to build natural language processing models for the next billion users in India.
“Language based skills are one of the strong skills of people in rural communities. What we essentially
want to do is take the wave of digital work that is imminent in language based machine learning
models and build a platform that can take that work to the communities which are skilled in those
types of tasks.” - Seshadri
本报告来源于三个皮匠报告站（）,由用户Id:1079574下载,文档Id:909811,下载日期:2025-09-12
PART 3: Learnings for Industry & the Way Forward for Research
KEY LEARNINGS FOR INDUSTRY:
Building a community of workers that is highly skilled, motivated and incentivized to
continue learning is a value in itself as it enhances worker well-being; it is also
associated with better quality output.
Diverse workforces are good for business and can improve retention and commitment
to learning and building skill sets for success
It is critical to consider how to shape supportive peer networks for data labellers as this is
instrumental in determining work-experience and subsequently in driving better quality
output
RESPONSE FROM RESEARCHERS & SCHOLARS
“There's so much energy spent on how to get high quality out of people when the answer has been
there all along. It's you care about them, you care about them having an environment in which
they can do good work, you get high quality data.”
- Mary Gray
“ . . .we wanted to bui ld a strong community, which doesn’t f ight against each other for
the monetary benefits but l ikes working together to make the end job successful” -
AJ Malasane, Playment
“80% of who we hire are actual ly from very poor backgrounds, who are completely
locked out. What we get is phenomenal retention, phenomenal motivation. We get
people that for whom this work is transformational in their communities” -
Jai Natarajan, iMerit
iMerit relies on in-house domain experts who design
curricula based on labellers pre-existing knowledge.
As a result, labellers with limited educational
qualifications are able to effectively annotate medical
imagery and carry out other complex tasks.
“A lot of people have what we might call
street smarts or learnability. And when
they're presented with the right curriculum in
context, they're able to convert that into high
data labeling quality”- Jai Natarajan, iMerit
TRAINING: Intuitive training designed and lead by domain experts can supplement
labellers with limited expertise and enable more inclusive hiring policies
‘Super labelers’, ‘Checkers’ and formally employed data labelers often deliver higher quality output as a
result of familiarity with processes and learned expertise.
PART 3:
“there was a time we had 500,000 annotators on the platform. We kind of squeeze that down to
really high quality individuals who take this as for their l iving, their incentives are in l ine. They can
spend time they understand higher levels of concepts” - Ajinkya Malasane, Playment
HUMAN RESOURCES: Building sustainable career trajectories for workers by
investing in their development is critical for enhancing livelihood opportunities and
can also create positive business outcomes
TRAINING: Investing in dynamic training and skill ing of data
labelers enables them to rise up a skill curve
APPROACHES TO IMPLEMENTATION:
Learnings for Industry & the Way Forward for Research
Investing in workers and their career development in a way that facilitates growth and mobility within the
company can have an impact on the individual and their respective communities. As a company that is
dedicated to creating lasting impact, iMerit claims that this approach does not hinder performance, cost,
quality or agility compared to its industry competitors. In contrast, they argue that the next upcoming trend
in the industry is to see the human or ‘worker’ in the loop even after the labelling/annotation is deployed.
“We have two PhDs in l inguistics. They sit down with the customer and they unpack the
customer's jargon rich domain and they convert that to a data labeling model” -
Jai Natarajan, iMerit
“We've done surveys of our employees in the neighborhoods, the communities that we operate in.. .We
find that over 80% of them [workers] pay back in the community, they educate a sibling or a family
member or in turn they pursue their own education in night college. They build things l ike sanitation
in their communities and in their homes. We find that virtuous cycle really working.”- Jai Natarajan,
iMerit
AREAS FOR FUTURE RESEARCH
AND POLICY CONSIDERATIONS:
PART 3: Learnings for Industry & the Way Forward for Research
"The future of work is not a what, but a how"
- J A I N A T A R A J A N , I M E R I T
TECHNICAL AND SERVICE ISSUES:
The areas for future research and policy considerations we present below have been drawn from
the roundtable presentations and discussions. We’ve aimed to identify areas that we feel demand
future study but that might also have impactful implications for the varied stakeholders and actors.
We have broken down the areas into three groupings of issues. These are intended as an aid for
organising the content, and we recognise the issues likely intersect in multiple ways.
How can we best support an evolving industry in which labelling is moving from
simple tasks to more complicated skilled/knowledge work?
The changing nature of the tasks and activities coming from data labelling customers is presenting
new challenges in terms of measuring both output and quality. This presents more than merely
technical hurdles. It raises issues around how work should be judged fairly; the responsibility the
industry should take in supporting knowledge work; and how such work should receive recognition.
It also speaks to a wider set of concerns surrounding the ways in which skilled labour is being
increasingly parameterised and valued in purely monetary terms. As the data labelling industry
moves towards measuring output and especially quality, further research is needed to explore how
best to recognise the labour force, and how to do more than provide individual financial reward, and
consider the prosperity and sustainability of communities.
How should we make sense of a focus on platform building in provisioning data
labelling services?
All the startups contributing to the roundtable discussions spoke of their efforts to build bespoke
platforms. These focused on providing access to labellers, quantifying labour and
supporting/measuring quality. A variety of questions arise around this, including how an ecosystem
(or competing landscape) of heterogeneous platforms should be understood, mapped and regulated;
how such technical platforms distribute and organise labour and skill; and where these systems
locate control and authority?
PART 3: Learnings for Industry & the Way Forward for Research
SOCIOTECHNICAL ISSUES:
How can we begin to understand the role of language in data labelling and
specifically the role of labelling speech and language use?
With the increasing trend towards more complex labelling activities to support AI systems, some
platforms are trying to cope with complex language labelling tasks. This has shown the underlying
challenges faced in this ostensibly technical work. Language is being shown to be unlike, for example,
image labelling, as it requires a much more sophisticated understanding of the structures of language
use and contexts in which language is used. Furthermore, we need to recognise there will be a
variability of language use by labellers, one that intersects with social/cultural dimensions such as
education, class/caste, geography, coloniality, etc. (see below). Language understanding is seen as an
important area in AI and machine learning, but through the work of data labelling it reveals the need
for significant future study.
How can labelling platforms enhance and build on collective/communal labour?
Labelling and annotation platforms currently apply an individual user model, aiming to maximise the
labelled outputs individuals produce. As the industry moves towards quality as a metric, a greater
understanding of collective or communal work is needed. From cases such as micro loan schemes,
we know that many of the people doing labelling work live in and draw on communal skills and
knowledge. This invites future work understanding the relations between labelling work and
collective skills, and where platforms might emphasise not just drawing on but sustaining
communities.
STRUCTURAL ISSUES
What are the ways in which gender intersects with data labelling work? What are
the opportunities, outcomes and impact on women’s l ives and social capital?
One of the startups described its efforts to hire women, with a workforce made up of 60% women.
This was presented as an emancipatory venture, offering women a source of income, and greater
independence and skill/expertise. Beyond these benefits, we also need to consider how such
changes might reconfigure wider social dynamics, ranging from changes in the domestic realm to the
wider labour market.
PART 3: Learnings for Industry & the Way Forward for Research
How should we account for other wider structural issues that intersect with labour,
such as race and coloniality?
As in other parts of the technology sector, what is seen as the laborious work of innovation is being
exported to regions in the Global South, such as India, Vietnam, Kenya and Uganda. As well as issues
of recognition and reward, this raises questions about the role of race and coloniality, and, specifically
for labelling, how histories of cultural integration (and power and subordination) play into the work of
category making.
What is required at a systemic and institutional level to enable justice for workers? How can
we frame regulation or policy keeping in mind the diversity of lived realities, experiences
and differences in how work may be structured by platforms?
We must ask how equity and justice are achieved for the workforce with the labelling industry
becoming a substantial sector in global economies, particularly in emerging markets. The structures at
play in sustaining the industry are built on logics that rely on exploitation, and alongside it the
datafication of labour and digital surveillance. Moreover, workers who are an essential part of making
AI have no role in governance. Amidst these conditions, we need to explore how the conditions for
equity and justice are afforded and ensured.
