ImpactMojoDigital Ethics 101www.impactmojo.in

ImpactMojo 101 Series · Free Forever

Digital
Ethics
101

Deploying Technology Responsibly: a Foundational Course on Data, AI & Digital Rights for Development Practitioners in South Asia

Research-BackedSouth Asia Focus100 SlidesFree Access

ImpactMojoDigital Ethics 101www.impactmojo.in

Agenda

What We Cover

Why Digital Ethics Matters

Slides 3–11

Data Privacy & Protection

Slides 12–20

Algorithmic Bias & Fairness

Slides 21–29

AI in Development

Slides 30–38

Digital Identity & Inclusion

Slides 39–47

Surveillance, Power & Consent

Slides 48–56

The Digital Divide

Slides 57–65

Platform Power & Data Colonialism

Slides 66–74

Misinformation & Online Harms

Slides 75–83

Responsible Design & Principles

Slides 84–91

Governance, Regulation & Practice

Slides 92–99

ImpactMojoDigital Ethics 101www.impactmojo.in

Section One

Why Digital Ethics Matters

ImpactMojoDigital Ethics 101www.impactmojo.in

The Promise

Technology entered development with a promise

Mobile money, biometric IDs, telemedicine, e-governance and AI have reached deep into the lives of the poor in South Asia: faster than almost anywhere on earth. The promise was real: leapfrog broken systems, cut leakage, deliver at scale.

But the same systems that include millions can exclude millions, and at this scale, a design flaw is not a bug, it is a policy that touches a billion people.

The promise	What it cost when it failed
Mobile money reaches the unbanked	Agent networks thin out exactly where cash is scarcest
Biometric ID ends duplicate claims	A worn fingerprint means no ration today
Direct transfer removes the middleman	A wrong account number is discovered a quarter later
E-governance makes the state reachable	Reachable by anyone who can use a form in English

Both columns are true at once, and this deck is not an argument for the second against the first. Digital delivery has reached people the paper system never found. The discipline is to hold the gain and the cost in view together, because programmes that see only one of them make predictable mistakes.

At national scale, a failure rate is a population. A system that works for 99 per cent of a billion people fails ten million of them, and they will be disproportionately the old, the manual labourers and the people furthest from a working network.

ImpactMojoDigital Ethics 101www.impactmojo.in

Definition

What we mean by digital ethics

Digital ethics

The study of how digital technologies (data, algorithms, platforms, AI) should be designed, deployed and governed so they respect rights, dignity and fairness, especially for the people with least power to refuse them.

It is not a compliance checkbox bolted on at the end. It is a set of questions asked from the first design meeting: who benefits, who is harmed, who can opt out, and who decides?

Asked at the design meeting	Asked at the launch review
Do we need this field at all?	How do we secure the field we already collect?
What happens to someone the system rejects?	What is our exception-handling backlog?
Who can refuse this, and at what cost?	What does the consent form say?
Cost to change: an hour	Cost to change: a rebuild

The left and right columns are the same questions at different prices. Nothing in this deck is technically difficult; almost all of it is cheap early and close to impossible late, which is why the timing is the substance of the argument.

Note the phrase “least power to refuse” in the definition. It is doing real work: a system used by people who can walk away raises different questions from one used by people whose pension depends on it.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Clash

'Move fast and break things' meets vulnerable people

Move fast and break things.

an early Silicon Valley motto

When the 'things' you break are someone's ration entitlement, pension, or right to be seen by the state, the cost of a failed experiment is not lost revenue: it is a missed meal.

Failed experiment in a startup	Failed experiment in a welfare system
Lost revenue, recoverable	A missed meal, not recoverable
Users churn, and are replaced	Users cannot leave; there is no second ration shop
Rollback in an afternoon	Rollback needs a policy decision and months
Failure shows up in the metrics	Failure shows up as an absence nobody logs

The fourth row is the one that makes the others dangerous. A person turned away does not appear in the dashboard as a failure; they appear as a transaction that never happened. The system reports success while excluding people, and it does so honestly.

The practical corollary: instrument the failures. A count of authentication rejections, by shop and by day, is a cheap piece of engineering and is usually the only way exclusion becomes visible to anyone with the power to act.

ImpactMojoDigital Ethics 101www.impactmojo.in

Tech4Dev

The double edge of every deployment

Promise

Reach the last mile cheaply
Cut leakage and ghost beneficiaries
Real-time monitoring of services
Give people a digital voice

Peril

Exclude those who cannot authenticate
Surveil the poor as a condition of aid
Automate and hide existing bias
Extract data, return little value

Neither column is the whole truth. Ethics is the discipline of holding both at once.

Claim made for a deployment	Question that tests it
“It cuts leakage”	How many genuine claimants were removed alongside the ghosts?
“It reaches the last mile”	Reaches, or requires the last mile to come to it?
“It gives people a voice”	Who reads what they say, and what happens next?
“It enables real-time monitoring”	Monitoring of the service, or of the people receiving it?

The first row is the most consequential question in this field. Removing ineligible claimants and removing eligible ones look identical in the headline number, because both appear as a fall in the beneficiary count. Only disaggregated follow-up distinguishes success from exclusion.

The last row names the recurring slippage. Monitoring built to hold a service accountable very often ends up monitoring the recipients instead, because they are the easier party to observe.

ImpactMojoDigital Ethics 101www.impactmojo.in

Scale

Why scale changes the ethics

A flawed clinic harms a village. A flawed national platform harms a nation. Digital systems concentrate decisions (one ruleset, one database, one model) so errors no longer cancel out; they replicate.

1 ruleset

decides eligibility for hundreds of millions

1 outage

can freeze entitlements across whole states

Paper system	Digital system
Many local rulesets; errors are local	One ruleset; errors replicate everywhere
Discretion allows a clerk to fix an edge case	Discretion is designed out, so edge cases fail hard
Failure is visible and local	Failure is distributed and invisible
Corruption is retail	Errors are wholesale, and so are fixes

The second row is the honest trade-off. Removing clerical discretion removes petty corruption and removes the human who could make an exception for the widow whose name is spelled two ways. Both effects are real, and the second is rarely costed.

Scale also cuts the other way. One fix propagates to everyone too, which is why a single well-designed exception route can protect millions. Concentration is the risk and the opportunity in the same property.

ImpactMojoDigital Ethics 101www.impactmojo.in

Who Is Affected

Practitioners sit on both sides

You may deploy digital tools: a survey app, a beneficiary database, a chatbot. You are also affected by them: the platforms you depend on, the IDs your participants must carry.

This course is for both roles: to build more carefully, and to push back on systems that harm the communities you serve.

As a deployer, you decide	As a user of others’ systems, you can
Which fields the survey app collects	Document the failures your participants hit
How long the beneficiary database is kept	Report exclusion upward with numbers, not anecdotes
Whether there is an offline route	Ask a vendor the questions in Section 11
Who can export the data, and to where	Refuse to require an ID the programme does not need

The left column is where most practitioners have real power and where it is least often exercised. A field officer choosing not to collect caste in a survey that does not need it has done more concrete good than a policy paper on data minimisation.

The right column compounds slowly. Documented exclusion, with counts and dates, is what eventually changes a national system, and field organisations are the only people positioned to collect it.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Core Questions

Five questions to ask of any digital system

Who benefits, and who bears the risk?
Who is excluded by the design, and how loudly do they fail?
What data is collected, and could it be used against people?
Who decides, and can an affected person contest a decision?
What if it breaks: is there a human fallback?

Question	A bad answer sounds like
Who benefits, who bears the risk?	“Everyone benefits”
Who does the design exclude?	“Coverage is 98 per cent”
What data, and could it be turned on people?	“It is all anonymised”
Who decides, and can it be contested?	“The system decides objectively”
What if it breaks?	“It has 99.9 per cent uptime”

Every bad answer in the right column is true and non-responsive, which is what makes them persuasive in a meeting. Each answers a question about the average case when the question asked was about the failing case.

Take these five to a vendor demonstration. They are the whole deck compressed, they need no technical vocabulary, and the quality of the answers tells you most of what you need to know about whether the supplier has thought about anyone but the median user.

ImpactMojoDigital Ethics 101www.impactmojo.in

Roadmap

How this course is built

The harms

Privacy, bias and AI
Identity and surveillance
The divide and platform power
Misinformation and online harm

The response

Responsible design principles
Privacy and ethics by design
Governance and regulation
A practical org checklist

Throughout, the examples are South Asian: Aadhaar, the DPDP Act, WhatsApp, the systems you actually meet at work.

Section	What you will be able to do
2 · Privacy	Apply the proportionality test; know your DPDP obligations
3–4 · Bias and AI	Ask for disaggregated error rates; match oversight to stakes
5–6 · Identity and surveillance	Insist on a manual fallback; spot function creep early
7–8 · Divide and platforms	Design for the unconnected; avoid rented-rails dependence
9–11 · Harms and governance	Run a procurement conversation and an ethics checklist

The examples are deliberately South Asian. Most digital-ethics material is written about European regulation and American platforms, and its assumptions (that consent is meaningful, that opting out is possible, that a regulator will act) hold differently here.

If you are reading before a specific decision: Section 11 has the checklist and the procurement questions, and Section 10 has the design principles that make the checklist answerable.

ImpactMojoDigital Ethics 101www.impactmojo.in

Section Two

Data Privacy & Protection

ImpactMojoDigital Ethics 101www.impactmojo.in

Foundations

What counts as personal data

Personal data

Any information that relates to an identified or identifiable individual: name, Aadhaar number, phone, location, biometric, health record, even a combination of seemingly harmless fields.

'It's just a phone number' is how most leaks begin. Personal data is anything that can be traced back to a person, alone or in combination.

Field	Alone	Combined with two others
Village	Thousands of people	Frequently one household, and sometimes one person
Age and gender	Broad categories
Caste and occupation	Common in the district

“It is just a phone number” is how most leaks begin, and the same logic applies field by field. Personal data is not a category of field; it is a property of a combination, which is why minimisation has to be judged across the whole record rather than one column at a time.

Small populations are the hard case. In a hamlet of two hundred, “woman, 60–65, widow, Scheduled Caste” may identify exactly one person, and disability, HIV status or a caste record attached to that row is a document that can be used against her.

ImpactMojoDigital Ethics 101www.impactmojo.in

Why Privacy

Privacy is not secrecy: it is control

Privacy is not having something to hide. It is the power to decide who knows what about you, and to live without being watched, profiled or sorted by default.

Arguing that you don't care about privacy because you have nothing to hide is like saying you don't care about free speech because you have nothing to say.

Edward Snowden

“Nothing to hide” assumes	Which fails when
The watcher is benign	Governments and employers change
The rules stay as they are	Data outlives the law that permitted it
Information is read in context	A clinic visit becomes a note in a file elsewhere
Everyone faces the same scrutiny	The poor are watched far more, for far less

Privacy is about control, not concealment. The question is not whether a fact is shameful but who gets to decide where it travels, and a person who cannot decide has lost something whether or not the fact is embarrassing.

The second row is the one that matters most in practice. A database built under one government, one law and one purpose will still exist under the next. Design as though the least trustworthy plausible future custodian will hold it, because eventually one will.

ImpactMojoDigital Ethics 101www.impactmojo.in

Landmark

Puttaswamy (2017): privacy is a fundamental right

In K.S. Puttaswamy v. Union of India (2017), a nine-judge bench of the Supreme Court of India held unanimously that the right to privacy is a fundamental right under Article 21 of the Constitution.

9 judges

unanimous Constitution Bench

Puttaswamy, 2017

Art. 21

privacy read into the right to life & liberty

Proportionality

any intrusion must be lawful, necessary & proportionate

What Puttaswamy settled	What it left open
Privacy is a fundamental right under Article 21	How the proportionality test applies case by case
It is not absolute; intrusion must be justified	Who decides whether a justification is adequate
It covers informational privacy, not only bodily	What remedy an excluded person actually has
Nine judges, unanimous	Later benches have applied it with varying strictness

The judgment is the foundation for everything in this section and is not self-executing. It supplies a test that any data-collecting programme can be measured against; it does not supply anyone to do the measuring, which is what Section 11 is about.

Useful in an internal argument. “Is this necessary and proportionate?” is a constitutional standard in India, not a preference, and framing a data question that way changes who has to justify what.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Test

When can the state intrude on privacy?

Puttaswamy set a proportionality test: a restriction on privacy is valid only if it clears every step. This is the lens for judging any data-collecting programme.

LEGALITY: backed by a valid law

→

LEGITIMATE AIM: a genuine state goal

→

NECESSITY: no less-intrusive way works

→

PROPORTIONALITY: benefit outweighs the harm

Step	Where programmes usually fail it
Legality	Collection authorised by a circular rather than a law
Legitimate aim	The aim is real, but stated so broadly it justifies anything
Necessity	A less intrusive option existed and was never assessed
Proportionality	The benefit is measured; the harm to the excluded is not

The third step is the one that decides most real cases. Almost every data collection has a legitimate aim; far fewer can show that the least intrusive means was considered. Asking “what would we do if we could not collect this?” is the practical form of the test.

The fourth step needs a number on both sides. A proportionality argument that quantifies savings and describes exclusion as “some initial difficulties” has not weighed anything; it has asserted a conclusion.

ImpactMojoDigital Ethics 101www.impactmojo.in

Principle

Consent: the floor, not the ceiling

Consent must be free, not the price of a basic service
Informed: people know what, why, for how long
Specific: for a stated purpose, not 'anything forever'
Revocable: people can withdraw without penalty

A thumbprint on a form nobody explained, given to receive a pension, is not free or informed consent.

Consent theatre	What would make it real
A thumbprint on an unexplained form	An explanation in the person’s language, before the form
“Agree to continue” with no alternative	A route to the service for someone who declines
Consent to “partners and affiliates”	Named purposes, with a stated retention period
Withdrawal that requires a district office visit	Withdrawal by the same channel as consent

The second row is the one that decides whether the rest matter. Consent given because the alternative is losing a pension is not free in any meaningful sense, however well the form is drafted, and no amount of explanation repairs it.

A test worth applying to your own consent process: in the last year, how many people declined? If the answer is none, you are not collecting consent; you are collecting signatures.

ImpactMojoDigital Ethics 101www.impactmojo.in

Principle

Purpose limitation & data minimisation

Purpose limitation

Data collected for one purpose should not be quietly reused for another. Immunisation records are not a recruitment list.

Data minimisation

Collect the least you need. Every extra field is a future liability: a breach risk and a tool that can be turned on people.

The safest data is the data you never collected. Ask of every field: do we truly need this?

Collected for	Later used for
Immunisation follow-up	Locating families for an unrelated enforcement drive
A livelihoods survey	Targeting lists shared with a funder or a vendor
Attendance monitoring	Evidence in a disciplinary or eligibility decision
Grievance records	Identifying complainants to the people complained about

The last row is the one NGOs get wrong most often. A grievance system that stores identifiable complaints without a strict access rule is a list of people who spoke up, and in a small community that is a dangerous document to hold.

The safest data is the data you never collected, and the second safest is data you deleted on schedule. Set a retention period at the point of collection; nobody ever sets one later, because by then someone can always imagine a use.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Law

India's Digital Personal Data Protection Act, 2023

The DPDP Act, 2023 is India's first comprehensive data-protection law. It applies to anyone processing digital personal data: including NGOs, researchers and small organisations. Its duties apply from 13 May 2027, under the commencement notification G.S.R. 843(E) of 13 November 2025; the Data Protection Board has existed since that date.

Process data only for a lawful purpose, with consent or a legitimate use
Honour rights to access, correction and erasure
Notify breaches and the Data Protection Board
Stronger safeguards for children's data and processing

'We're a small NGO' is not an exemption. Know your obligations before you collect a single record.

Obligation	What a small NGO actually has to do
Lawful purpose and notice	Write down why each field is collected, in plain language
Consent or legitimate use	Know which basis you are relying on, per dataset
Access, correction, erasure	Have a named person who can act on a request
Breach notification	Know who you would tell, and within what time
Children’s data	Stricter treatment; verify before collecting from minors

The Act does not exempt you for being small. The obligations scale with what you process, not with your budget, and an organisation holding beneficiary records is processing digital personal data in exactly the sense the Act means.

Take advice on the specifics. Rules and exemptions under the Act have been evolving since it was passed; this slide is orientation, not legal advice, and the compliance detail is exactly the kind that changes between when a deck is written and when it is read.

ImpactMojoDigital Ethics 101www.impactmojo.in

Re-identification

Anonymisation is harder than deleting names

Removing names does not make data safe. A combination of village + age + caste + occupation can re-identify one person, especially in small areas where few share those traits.

Direct IDs

Name, Aadhaar, phone: remove entirely

Quasi-IDs

Age + place + caste can re-identify: coarsen or aggregate

Technique	What it protects against
Drop direct identifiers	Casual lookup only: the weakest step
Coarsen quasi-identifiers (age bands, district not village)	Combination attacks, which are the real risk
Aggregate before release	Individual re-identification, at the cost of detail
Restrict access rather than publish	Everything, and it is the most-skipped option

Anonymisation is a spectrum, not a state. A dataset is re-identifiable in proportion to how many other datasets exist to join it against, and that number only grows. “It is anonymised” describes an intention rather than a guarantee.

The small-area problem is acute in our work. Development data is usually village-level and caste-disaggregated, which is precisely the combination that identifies people. Publishing at district level costs analytical detail and is often the honest choice.

ImpactMojoDigital Ethics 101www.impactmojo.in

Section Three

Algorithmic Bias & Fairness

ImpactMojoDigital Ethics 101www.impactmojo.in

The Myth

Algorithms are not automatically neutral

It is tempting to think a formula cannot discriminate: the maths just crunches numbers. But an algorithm learns from data produced by an unequal world, and it can faithfully reproduce that inequality at scale.

Bias does not come from 'the maths being racist'. It comes from biased data, biased proxies and biased objectives: all human choices.

Human choice	Where it enters
What to predict	“Risk of default” or “need for credit” are different models of the same people
What data to use	Whatever was already recorded, which is whatever was already served
What counts as success	Overall accuracy, or the error rate for the worst-served group
Where to set the threshold	A pure value judgement, usually made by whoever ships it

None of these is a mathematical question, which is the point. The maths is neutral in the trivial sense that it optimises whatever it is given; every choice about what to give it is a decision about people, made by people, and usually undocumented.

“The algorithm decided” is therefore an evasion, and it is a persuasive one because it sounds like a statement about physics. In every case there is a person who chose the objective, and asking who is the first useful question.

ImpactMojoDigital Ethics 101www.impactmojo.in

Definition

What algorithmic bias means

Algorithmic bias

A systematic, unfair difference in how an automated system treats different groups: producing outcomes that disadvantage people by gender, caste, religion, region, language or disability.

It is rarely intentional. It is usually inherited: from the data the model was trained on, and the world that data came from.

Where bias shows up in our sector	Form it takes
Credit and microfinance scoring	Thin-file applicants scored as risky for having been excluded
Beneficiary targeting models	Households missed by past surveys stay missing
Speech and language tools	Accuracy falls sharply outside dominant languages and accents
Facial recognition	Higher error rates for darker skin and for women, repeatedly measured
Fraud detection	Flags correlate with poverty markers, so the poor are investigated more

Every row shares one shape: the group that was under-served historically generates less and worse data, and the model reads that absence as a property of the group rather than a property of the record.

The last row is the one to watch in welfare. A fraud model trained on past investigations learns who was investigated, not who defrauded, and past investigation is itself a record of who was easy to investigate.

ImpactMojoDigital Ethics 101www.impactmojo.in

Source One

Bias enters through the training data

A model learns patterns from past examples. If history under-served women, recorded fewer female-headed loans, or skipped remote villages, the model learns that those people are 'lower priority', and repeats it.

The model is a mirror. If the data reflects a world that excluded people, the model will too: only faster and with an air of objectivity.

Gap in the data	What the model concludes
Fewer loans recorded to women	Women are a weaker lending prospect
Remote villages surveyed less often	Those villages have less need
Disability rarely recorded at all	Disability is not a relevant feature
Complaints logged only where offices exist	Service is satisfactory where there are no offices

The model is not wrong about the data; it is wrong about the world, and it has no way to tell the difference. Absence in a dataset means “not recorded”, and every model treats it as “not there”.

Which makes documentation the practical safeguard. A written note of who is missing from a training set, and why, is cheap to produce at collection time and impossible to reconstruct afterwards, and it is the thing an auditor will ask for first.

ImpactMojoDigital Ethics 101www.impactmojo.in

Source Two

Proxies smuggle in protected traits

Even if you remove caste, religion or gender from the data, other fields can stand in for them. Pincode can proxy for caste or religion; name can reveal community; occupation can track historical disadvantage.

Dropping the sensitive column does not remove the bias if a proxy remains. The model finds the back door.

Removed field	What still carries it
Caste	Surname, pincode, occupation, school attended
Religion	Name, locality, dietary fields, festival-linked activity
Gender	Name, occupation, phone-ownership patterns, time of use
Disability	Claim history, service-usage patterns, employment gaps

“We do not collect caste” is not a fairness claim. In South Asian data, surname and locality reconstruct it well enough for a model to act on, and dropping the column simply removes your ability to check whether it is doing so.

Which produces a genuine dilemma. To test whether a model discriminates by caste, you need caste data: the very field you were right to be wary of collecting. The usual resolution is to hold it separately, under tight access, for audit only, and never to expose it to the model.

ImpactMojoDigital Ethics 101www.impactmojo.in

Source Three

Feedback loops make bias self-fulfilling

Each loop deepens the gap: the group gets less, generates less data, and looks even 'riskier' next time.

The loop is self-confirming, and from inside it looks like accuracy. Each cycle the model’s predictions match outcomes more closely, because the model is causing the outcomes it predicts. Performance metrics improve while the exclusion deepens, so the dashboard reports success.

Breaking it costs measurable efficiency. It means spending a reserved share of outreach where the model says not to: the only way to generate data that could disconfirm the loop. Nobody does this by accident; it has to be a written rule.

ImpactMojoDigital Ethics 101www.impactmojo.in

Illustration

When a model fails one group more than another

Illustrative: a 'good' overall model with unequal error rates

Illustrative example, not real data

A 12% overall error can hide a 26% error for one group. Aggregate accuracy is not fairness.

ImpactMojoDigital Ethics 101www.impactmojo.in

Many Fairnesses

Fairness is not one thing

There are several mathematical definitions of fairness (equal error rates, equal selection rates, equal outcomes) and they often cannot all hold at once. Choosing one is a value judgement, not a technical default.

So the right question is not 'is it fair?' but 'fair in which sense, for whom, and who decided?'

Definition	What it equalises	What it gives up
Equal selection rate	Same share chosen from each group	May select less-qualified from one group
Equal error rates	Same chance of being wrongly rejected	Selection rates will differ
Equal precision	A positive means the same thing for everyone	Conflicts with equal error rates

These cannot all hold at once when base rates differ between groups: a result that is mathematically proved, not a matter of engineering effort. Any system claiming to be simply “fair” has chosen one and not told you which.

So the practical demand is disclosure, not perfection: which definition was chosen, by whom, and were the people affected part of that choice? That is a governance question with a technical surface, and it belongs in a procurement conversation rather than a model card.

ImpactMojoDigital Ethics 101www.impactmojo.in

Response

Auditing for bias is a practice, not a one-off

Disaggregate performance by gender, caste, region, language
Test for proxies: can removed traits be reconstructed?
Document training data, its gaps and known limitations
Re-audit as the system and population change over time

Fairness is not automatic and never finished. It has to be measured, on purpose, again and again.

Audit step	What to ask for
Disaggregate performance	Error rates by gender, caste, region, language, not overall accuracy
Test for proxies	Can a removed trait be predicted from the remaining fields?
Document the training data	Who is in it, who is missing, and over what period
Re-audit on a schedule	Populations drift; a model correct in 2024 is not therefore correct now
Record the decisions	Which fairness definition, which threshold, and who signed it off

The first row is the whole audit in one line. Aggregate accuracy is the number a vendor will lead with and the number that hides the harm; an overall error of 12 per cent is compatible with 26 per cent for the group that can least afford it.

Write the audit obligation into the contract. Asking for disaggregated performance after deployment is a favour; requiring it before signing is a term, and vendors who cannot supply it will say so at that point rather than a year later.

ImpactMojoDigital Ethics 101www.impactmojo.in

Section Four

AI in Development

ImpactMojoDigital Ethics 101www.impactmojo.in

The Moment

AI arrives in the development sector

From crop advisories and chatbots to diagnostic tools and beneficiary targeting, AI, including large language models, is now being piloted across South Asian programmes. The question is no longer whether, but how responsibly.

AI is a powerful amplifier. It amplifies good design and good data, and it amplifies bias, error and exclusion just as efficiently.

Question to settle before a pilot	Why it decides everything after
Is the model advising or deciding?	Sets the entire oversight requirement
Who is accountable when it is wrong?	If the answer is “the system”, nobody is
Was it evaluated on our population?	Performance elsewhere transfers poorly, especially across languages
What is the fallback when it is unavailable?	Uptime failures land on the same people as model failures

“Amplifier” is the right word for what AI does here. Applied to a well-designed process with good data, it extends reach. Applied to a process with an unexamined exclusion in it, it applies that exclusion faster, more consistently and with a confident interface.

Pilots have a specific failure mode in this sector: they are run with the vendor present, on the best-connected sites, with staff who were trained last week. None of those conditions holds at scale, and the pilot result is the number that gets quoted.

ImpactMojoDigital Ethics 101www.impactmojo.in

Opportunities

Where AI can genuinely help

Translation: bridge dozens of languages and scripts cheaply
Diagnostics: screen X-rays, retina scans, skin conditions where doctors are scarce
Targeting: find likely-eligible households for outreach
Triage: route calls, summarise field reports, flag urgent cases

Used as an assistant to stretched human workers, AI can extend reach into places services never went.

ImpactMojoDigital Ethics 101www.impactmojo.in

Risk One

Opacity: the black-box problem

Black box

A system whose internal reasoning cannot be inspected or explained: you see the input and the output, but not why it decided as it did.

If a model denies someone a benefit and no one can explain why, the person cannot contest it. Opacity quietly removes the right to an explanation, and to appeal.

What “explainable” has to mean	Test
To the affected person, not the engineer	Could a field officer say it aloud in the person’s language?
Specific to this decision	Not “the model considers many factors”
Actionable	Does it tell them what would change the outcome?
Available before the appeal deadline	An explanation that arrives late is not a remedy

Opacity removes a right without ever debating it. The right to contest an adverse decision depends on knowing the reason for it, so a system that cannot state a reason has quietly abolished the appeal while appearing merely technical.

A simpler model is often the ethical choice. Where a decision affects entitlements, a slightly less accurate rule that can be written on one page may serve people better than an opaque model that performs marginally better on average.

ImpactMojoDigital Ethics 101www.impactmojo.in

Risk Two

Automating exclusion at scale

When an AI system decides who is 'eligible', 'at risk' or 'likely fraudulent', it can lock millions out with no human in the room. The error is invisible to the operator and devastating to the excluded.

A false 'ineligible' is not an abstraction. It is a widow without a pension, a child without a scholarship: harm that compounds silently.

Error type	Who notices	What follows
False positive: wrongly included	Auditors, quickly	Tightening, and press coverage about leakage
False negative: wrongly excluded	Only the person excluded	Usually nothing; they stop coming

This asymmetry drives system design almost everywhere. Wrongful inclusion is measured, reported and punished; wrongful exclusion is invisible in the same data. Every incentive therefore pushes thresholds toward excluding more people, and the pressure is one-directional.

Correcting it requires deliberately counting the invisible error: surveying people who dropped out, auditing rejections rather than approvals, tracking the exception register. None of that happens by default, and all of it is cheap compared with the harm.

ImpactMojoDigital Ethics 101www.impactmojo.in

Risk Three

Hallucination & overtrust

Generative AI can produce confident, fluent answers that are simply wrong. In health, legal or entitlement advice, a plausible falsehood delivered with authority is dangerous.

The risk grows when users assume the machine must know better. Fluency is not accuracy. Always keep a verified human source for high-stakes advice.

Where fluency is most dangerous	Because
Health and medication advice	A confident wrong dose is acted on
Legal entitlement and rights	Invented sections and deadlines are unverifiable by the user
Scheme eligibility	A plausible answer stops someone applying
Translation into a low-resource language	Nobody in the room can check it

The last row is specific to our context and under-discussed. Generated text in a language the deploying team does not read cannot be reviewed by that team, and the fluency that makes it usable is exactly what makes an error invisible.

Overtrust compounds it. A first-time user meeting a system that answers everything instantly has no basis for calibrating when to doubt it, and the interface offers no signal of confidence that means anything.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Safeguard

Human-in-the-loop

AI suggests (screen, score, draft)

→

Human reviews with context & judgement

→

Human decides, and is accountable

→

Affected person can question the decision

For any decision that affects rights or entitlements, AI should inform a human, never replace one. Keep a person accountable and reachable.

ImpactMojoDigital Ethics 101www.impactmojo.in

Match the Stakes

Not every task needs the same caution

Use	Stakes	Guardrail
Draft a report summary	Low	Human edits before sending
Translate a notice	Medium	Native speaker verifies
Screen a medical scan	High	Clinician confirms every case
Decide benefit eligibility	Very high	Human decides; appeal route open

Calibrate oversight to the harm of being wrong. The higher the stakes, the more human judgement and the easier the appeal.

What to check at each level of stakes
Low: is a human reading it before it leaves?
Medium: is the reviewer competent in the language and the subject?
High: does every case get human confirmation, and is that logged?
Very high: does a person decide, and can the affected person reach a human who can reverse it?

Calibration is the point. Treating every use as high-stakes produces an oversight process nobody follows; treating every use as low-stakes produces the failures in the previous slides. The judgement is about the cost of being wrong, borne by whom.

Watch for stakes creeping upward after deployment. A tool introduced to draft summaries becomes the basis for a decision within a year, without anyone revisiting the guardrail that was set for drafting.

ImpactMojoDigital Ethics 101www.impactmojo.in

Before You Deploy

Questions before piloting any AI

What is the harm if it is confidently wrong, and who bears it?
Can we explain a decision to the person it affects?
Was it tested on people like ours: our languages, our context?
Is there a human fallback and a working appeal route?
Who is accountable when it fails, and can they be reached?

Question	What an inadequate answer looks like
Harm if confidently wrong?	“It is only a suggestion”, check whether anyone overrides it
Can we explain a decision?	“It uses many signals”
Tested on people like ours?	Benchmarks from another country, in another language
Human fallback and appeal?	A helpline number that rings in a city office
Who is accountable?	“The vendor”, with no clause saying so

The first row hides the most common failure. A tool described as advisory becomes decisive when overriding it requires justification and accepting it requires none. Ask what share of suggestions were overridden last quarter; the answer is usually close to zero.

Ask these before the pilot, in writing. Every one of them is answerable on day one and unanswerable once the system is embedded and the people who chose it have moved on.

ImpactMojoDigital Ethics 101www.impactmojo.in

Section Five

Digital Identity & Inclusion

ImpactMojoDigital Ethics 101www.impactmojo.in

Aadhaar

The world's largest biometric ID system

Aadhaar, run by the Unique Identification Authority of India (UIDAI), assigns a 12-digit number linked to fingerprints, iris scans and a photograph. It is the largest biometric identity system ever built.

145.32 cr

Aadhaar numbers generated, as of 28 September 2026

UIDAI enrolment dashboard

UIDAI

the statutory authority that runs it

What Aadhaar is	What it is often assumed to be
A number linked to biometrics, proving one person = one number	Proof of citizenship: it is not
An authentication service others can query	A database of everything about you: it holds a limited set
Statutorily backed, with a designated authority	Unregulated: there is a legal framework, contested but real
Optional to obtain (Aadhaar Act s3), but the government may require it for any subsidy, benefit or service paid from the Consolidated Fund (s7), which the Supreme Court upheld in 2018	Required for everything: banks and telecom companies may not refuse a service to someone who declines to use it (s4(6), after the 2018 judgment)

Getting these distinctions right matters for the argument. Critiques that overstate what the system holds are easy to dismiss and take the well-founded objections down with them. The strongest case against a design is one that describes it accurately.

The fourth row is where the ethics lives, and Section 5 returns to it: a system that is optional to obtain and a legal condition of receiving a subsidy raises the consent questions from Section 2 without ever having to answer them.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Scale

Enrolment at a scale without precedent

Aadhaar enrolment growth (illustrative trajectory toward 145 crore)

Illustrative trajectory to 2023; the 2026 point is UIDAI’s figure, 145.32 crore generated as of 28 September 2026 (enrolment dashboard)

The shape is illustrative; the headline is UIDAI’s own figure, 145.32 crore Aadhaar numbers generated by 28 September 2026. Scale this large makes both the gains and the harms enormous.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Gains

What digital ID made possible

A portable identity for people who had no documents at all
Direct benefit transfers into bank accounts, cutting some intermediaries
Removal of some duplicate and 'ghost' beneficiaries
Easier access to opening bank accounts and SIM cards

For someone previously invisible to the state, a verifiable identity can be genuinely empowering.

Who gains most	Why
People with no documents at all	Migrants, the landless and many women held nothing in their own name
Women opening an account independently	An identity not mediated through a husband or father
Inter-state migrants	Portability matters most to people who move
Anyone previously invisible to the state	Being countable is a precondition for being served

These gains are real and should not be conceded rhetorically. For a person who could not prove who they were, a portable verifiable identity is a genuine expansion of what they can do, and dismissing that makes the rest of the critique sound ideological.

The same property produces the harm, though. An identity that unlocks everything also gates everything, and the next slide is about the people for whom the gate does not open.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Harms

Authentication failure becomes exclusion

When entitlements are tied to biometric authentication, anyone the machine cannot read can be turned away. Worn fingerprints from manual labour, poor connectivity, server downtime, or a mismatch can mean no ration today.

Researchers and journalists have documented cases where authentication failures left the elderly and labourers unable to draw the food they were entitled to. Exclusion by error is still exclusion.

Failure	Who it hits hardest
Worn or damaged fingerprints	Manual labourers, older people: precisely the entitled
No connectivity at the point of sale	Remote villages, which are also the poorest
Server or service downtime	Everyone in that district, on that day
Demographic mismatch in records	People whose names are transliterated inconsistently
No linked mobile for the OTP fallback	Women, who own phones at markedly lower rates

Every failure mode correlates with the conditions the programme exists to address. That is not a coincidence: manual labour wears fingerprints, poverty tracks with poor connectivity, and inconsistent documentation follows from never having had documents. The errors are concentrated on the intended beneficiaries.

Researchers and journalists have documented specific cases in which authentication failures left entitled people without their ration. Treat the mechanism as established and the national magnitude as contested: that is the honest position, and it is enough to require a fallback.

ImpactMojoDigital Ethics 101www.impactmojo.in

Inclusion vs Exclusion

The same system, two outcomes

Dimension	Benefit	Harm
Identity	The undocumented become visible	Errors make the visible disappear
Delivery	Faster transfers, fewer ghosts	Failed auth blocks real beneficiaries
Mandate	One ID, many services	Function creep, no real opt-out
Bodies	Biometrics are hard to forge	Worn fingerprints exclude labourers

The lesson is not 'reject digital ID'. It is design for the person the machine fails, because at this scale, even a small failure rate is millions of people.

ImpactMojoDigital Ethics 101www.impactmojo.in

Design Fix

Always build a non-digital fallback

Any system where authentication gates a basic entitlement must have a guaranteed manual override: a way to receive your due when the technology fails, without being sent home.

Biometric fails

→

Try alternative (OTP, face, operator)

→

Still fails → manual exception register

→

Entitlement delivered: never denied for tech

A fallback that works	A fallback that does not
The dealer can issue today, and reconcile later	“Come back tomorrow”
Exception register the dealer is expected to use	An exception route that counts against the dealer
Logged and monitored centrally	Undocumented discretion, which invites abuse
Known to the claimant before they arrive	Known only to officials

Fallbacks fail on incentives, not on design. If exceptions are treated as suspicious, the person at the counter will avoid using them, and the route exists on paper only. Whether the operator is rewarded or penalised for using the override decides whether it works.

The logging requirement does double duty. It protects against abuse of discretion, and it is the only way anyone ever finds out how often authentication is failing and where, which is the exclusion data nobody otherwise collects.

ImpactMojoDigital Ethics 101www.impactmojo.in

Voluntary?

When 'optional' is not really a choice

An ID described as voluntary stops being voluntary if you cannot get a pension, ration or scholarship without it. Coerced consent is the central tension in mandatory-by-practice digital identity.

Ask of any 'voluntary' system: what happens to someone who says no? If the answer is 'they lose a basic service', it is not voluntary.

“Voluntary” system	What happens if you decline
ID for a subsidised ration	No ration
ID for a school scholarship	No scholarship
Aadhaar for a SIM card	In law, nothing: since the 2018 judgment the operator must accept another ID (Aadhaar Act s4(6)). The test is whether the person at the counter knows that
Biometric attendance for a wage	No wage that day

Apply the test to any system, including your own. “What happens to someone who says no?” converts a policy adjective into an observable fact, and a programme that cannot answer it has not thought about consent at all.

NGOs are frequently on the wrong side of this. Requiring an ID number to receive a service you provide, when the service does not need it, imports the coercion into your own programme: usually for reporting convenience.

ImpactMojoDigital Ethics 101www.impactmojo.in

Principle

Inclusion is the test, not enrolment

A digital ID is not successful because a billion people enrolled. It is successful only if the last person, the worn-fingerprint labourer in a low-connectivity village, can still claim what is theirs.

Measure systems by their failure cases, not their headline coverage. The margin is where the ethics lives.

Metric that flatters	Metric that tests
Enrolment coverage	Authentication success rate, by district and by age
Transactions completed	Transactions attempted and failed
Savings from removed duplicates	Genuine claimants removed alongside them
Average uptime	Outage hours in the lowest-connectivity blocks

The left column is what gets reported and the right column is what would tell you whether the system works. Every metric on the right is computable from data these systems already generate; the reason they are absent is that nobody required them.

This is the single most transferable idea in the deck. Measure a system by its failure cases, not its headline coverage, and it applies equally to your own beneficiary database, your helpline and your grievance process.

ImpactMojoDigital Ethics 101www.impactmojo.in

Section Six

Surveillance, Power & Consent

ImpactMojoDigital Ethics 101www.impactmojo.in

The Shift

From serving people to watching them

Digital systems built to deliver services also generate detailed records of where people go, what they receive and whom they know. Delivery and surveillance can run on the very same rails.

The question is rarely 'is there a camera?' It is 'who can see this data, link it, and act on it, and can the watched person say no?'

Built to deliver	Also produces
Ration transactions	A record of where a person was, and when, weekly
Attendance systems	Movement and association patterns
Health programme registers	Conditions, pregnancies, treatment histories
Grievance portals	A list of who complained, about whom

Nobody had to intend surveillance for this to happen. Delivery and monitoring run on the same rails, and the by-product of a well-designed delivery system is a detailed longitudinal record of the people it serves.

The question to ask is about linkage rather than collection. Each dataset above is defensible alone. The risk arrives when they can be joined on a common identifier, which converts four service records into one profile.

ImpactMojoDigital Ethics 101www.impactmojo.in

Function Creep

Data collected for X, used for Y

Function creep

When data gathered for one stated purpose is gradually repurposed for others: without fresh consent, often without the subject's knowledge.

A database built to deliver food can become a tool to track, profile or police the same people. The data outlives the promise made when it was collected.

Guard against creep	How
Bind the purpose in writing	State it at collection, and require a new basis for any new use
Set a deletion date	At collection, not later: retention only ever gets extended
Separate the identifier	Hold service data and identity data apart, joinable only under a rule
Log every access	Who read this record, when, and why

Function creep is rarely a decision; it is a series of reasonable requests. Each new use is individually defensible (an investigation, a study, an audit) and the cumulative result is a database doing something nobody would have approved at the outset.

The data outlives the promise. Whoever made the assurance at collection will not be in post when the request comes, and an undocumented promise is not a control. Write it into the schema, not the memo.

ImpactMojoDigital Ethics 101www.impactmojo.in

Whose Privacy

The surveillance of the poor

Those who depend on the state are watched most. To receive welfare, the poor must hand over biometrics, locations, family details and transaction histories: a level of scrutiny the wealthy never face to access their own money.

Surveillance is distributed unequally: those with the least power are watched the most.

a recurring finding in surveillance studies

To receive a subsidy, a poor household provides	To access their own savings, a wealthy person provides
Biometrics, at every transaction	A card, usually once
Household composition and income declarations	Nothing recurring
Location, weekly, by collection point	Not recorded as a condition
Justification for any change in circumstances	None required

The asymmetry is the finding, not the surveillance. Scrutiny is applied in inverse proportion to power, and the justification, that public money requires accountability, is never applied with equal force to tax expenditures or subsidised credit.

Worth stating in a design meeting. “Would we ask this of a person applying for a business loan?” is a quick test of whether a data requirement is about need or about suspicion.

ImpactMojoDigital Ethics 101www.impactmojo.in

Conditionality

Welfare conditionality as control

When aid is conditional on monitored behaviour (attendance tracked, accounts linked, movements logged) the price of help becomes constant observation. Dignity is quietly traded for entitlement.

Conditions framed as 'accountability' often land hardest on those least able to comply, and punish poverty rather than relieve it.

Condition	Who fails it, and why
Minimum attendance, biometrically verified	Sick, caring for someone, or without transport that week
Bank account linkage	People without documents, or with an account in another name
Repeated re-verification	Migrants, who are not present when the check runs
Digital grievance filing	The non-literate and the unconnected: the likeliest to have a grievance

Conditions sort by capacity to comply, not by need. The households that fail a condition are usually the ones facing the constraint the programme was designed for, so conditionality tends to remove exactly the people it was meant to reach.

The last row deserves separate attention. A grievance channel that requires the capability the grievance is about (being unable to authenticate, being unable to read the form) is a closed loop, and it is common.

ImpactMojoDigital Ethics 101www.impactmojo.in

Chilling Effects

Being watched changes behaviour

Chilling effect

When the knowledge or suspicion of being monitored makes people self-censor: avoiding lawful speech, association or services out of fear.

People may skip a clinic, a meeting or a complaint because a record might be kept and used against them. The harm is real even when no one is actually watching.

What people avoid when they suspect a record	Consequence
A clinic visit for a stigmatised condition	Untreated illness; onward transmission
A meeting of a union or a rights group	Weaker collective action, which is often the point
Filing a complaint against an official	Abuse continues, and appears to be absent
Seeking help after violence	The gravest, and the least visible in any dataset

The harm does not require anyone to be watching. Belief that a record exists and might be read is sufficient, which means chilling effects are produced by ambiguity, and ambiguity is the default state of most systems people encounter.

The counter is specificity. Telling people plainly what is recorded, who can see it, and how long it is kept reduces the chilling effect even when the answer is not ideal, because uncertainty is doing much of the damage.

ImpactMojoDigital Ethics 101www.impactmojo.in

Power

Surveillance is about asymmetry

They can see you

Location, transactions, contacts, biometrics: linked across databases into a single profile.

You cannot see them

Who holds the data, what rules apply, how to correct an error or contest a decision: opaque.

The ethical fault line is this asymmetry of visibility and power, not the technology itself.

They can	You cannot
Link your records across agencies	See what has been linked
Act on an inference about you	Learn what the inference was
Change the rules retrospectively	Withdraw data already given
Decline to explain a decision	Contest a reason you were never told

The technology is not the fault line; the asymmetry is. The same capability in a system where the subject can see, correct and contest is a different thing entirely, which is why the remedies in Section 10 are about visibility and redress rather than about restricting the technology.

A useful diagnostic: for any system, list what it knows about a person and what that person can find out about it. The gap between the two columns is the ethical exposure.

ImpactMojoDigital Ethics 101www.impactmojo.in

Consent Again

Consent under power imbalance

Consent means little when refusal means losing a livelihood or a service. A person desperate for aid cannot freely 'agree' to data terms they could never negotiate.

Where power is unequal, the burden shifts: the system must justify what it takes, rather than the person justify their refusal.

Where consent is weakest	Because
Welfare enrolment	Refusal means losing the entitlement
Employment conditions	Refusal means losing the job
NGO programme participation	Refusal is read as ingratitude, or as disqualification
Research participation	The researcher is often also the service provider

The third and fourth rows are ours. Where the person asking for consent is also the person who decides whether you receive help, consent is structurally compromised regardless of how carefully the form is worded or how sincerely it is offered.

The practical remedy is to separate the roles where you can, a different person seeking research consent from the one deciding eligibility, and, where you cannot, to lower what you ask for rather than improve how you ask.

ImpactMojoDigital Ethics 101www.impactmojo.in

Safeguards

Guardrails against surveillance creep

Purpose-bind data and delete it when the purpose ends
Separate service delivery from policing and profiling
Minimise linkage across databases by default
Independent oversight with teeth, and a real complaint route

Safeguard	What makes it real rather than stated
Purpose-binding and deletion	A date in the system, not a sentence in a policy
Separating delivery from policing	Different databases, different access, no shared key
Minimising linkage	Joins require an approval that is logged and refusable
Independent oversight	Power to compel disclosure and to stop processing
A complaint route	One a person without a smartphone can actually use

Each safeguard has a paper version and a working version, and the difference is whether it is enforced by the architecture or by good intentions. Architecture survives a change of management; intentions do not.

The second row is the one worth fighting for. Once a welfare database is queryable by law-enforcement, the chilling effects in the previous slides follow automatically, and no consent process can undo it.

ImpactMojoDigital Ethics 101www.impactmojo.in

Section Seven

The Digital Divide

ImpactMojoDigital Ethics 101www.impactmojo.in

Definition

The divide is not one gap but many

Digital divide

The unequal distribution of access to digital devices, connectivity, skills and meaningful use: cutting along gender, income, geography, language and disability.

It is not simply 'has a phone or not'. It runs from owning a device, to affording data, to being able to read the screen, to using it for anything that improves your life.

Level	What it takes to clear
A device in the household	Money, once, and whose device it is matters
A device you control	Frequently fails for women and younger family members
Affordable data and charging	Recurring cost, and reliable electricity
Skills and language	Literacy, and an interface in a language you read
Meaningful use	Something on the device that improves your life

The second row is invisible in most statistics. Household phone ownership is the standard measure and it hides intra-household distribution: a phone owned by a man is counted as access for his wife and daughters, who may use it under supervision or not at all.

Programme designs routinely stop at row one. Distributing devices closes the first gap and leaves the other four, which is why device-distribution projects so often report high coverage and low use.

ImpactMojoDigital Ethics 101www.impactmojo.in

Gender

The gender gap in access

Illustrative: mobile & internet access by gender

Illustrative pattern; a real gender gap is widely documented in South Asia

Figures are illustrative, but the pattern is well documented: women in South Asia are markedly less likely to own a phone or use the internet than men.

ImpactMojoDigital Ethics 101www.impactmojo.in

Geography

Rural, urban and the connectivity gap

Illustrative: internet use, urban vs rural

Illustrative pattern; an urban-rural gap is widely documented

Coverage maps flatter reality: a tower nearby does not mean an affordable, reliable signal inside a home.

ImpactMojoDigital Ethics 101www.impactmojo.in

Other Axes

Income, language and disability

Income: data, devices and electricity all cost money the poorest lack
Language: most services assume English or one dominant language
Literacy: text-heavy interfaces exclude non-readers
Disability: apps rarely work with screen-readers or for low vision

These axes stack. A poor, rural, non-literate woman with a disability faces every gap at once.

Axis	What it excludes in a typical service
Income	Recurring data cost, which is the binding constraint more often than the device
Language	Interfaces in English or one dominant state language
Literacy	Text-heavy flows with no voice or icon path
Disability	Apps that break screen readers; no captions; low-contrast text
Age	Small targets, jargon, and no assisted-use mode

The axes stack rather than substitute. A poor, non-literate rural woman with a disability faces all five at once, and a design that fixes one of them changes nothing for her, which is why single-axis accessibility fixes so often fail to move usage.

The cheapest high-value fix is usually voice. An IVR path covers literacy, language and low-end devices simultaneously, works over a voice network where data is unreliable, and is unglamorous enough that it rarely gets funded.

ImpactMojoDigital Ethics 101www.impactmojo.in

Layers

Access, skills, and meaningful use

LEVEL 1: Is there a device & signal?

→

LEVEL 2: Can they afford to use it?

→

LEVEL 3: Do they have the skills & language?

→

LEVEL 4: Does it actually improve their life?

Closing only the first gap and declaring victory leaves most people stranded at the higher levels.

Level	Typical intervention	What it leaves unsolved
1 Access	Distribute devices; build towers	Cost of use, skills, relevance
2 Affordability	Subsidised data	Skills, language, relevance
3 Skills	Digital literacy training	Whether anything useful is there
4 Meaningful use	Services worth using, in the right language	Nothing: this is the point of the other three

Most digital-inclusion funding sits at levels one and two, where progress is countable, and most of the benefit sits at level four, where it is not. That mismatch is the standard reason inclusion programmes report success and change little.

Measure at level four or do not claim inclusion. “Devices distributed” is an input; “people who completed something they needed” is the outcome, and only the second tells you whether the first mattered.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Trap

The risk of 'digital by default'

When a service goes online-only (applications, grievances, payments) the people on the wrong side of the divide are pushed out of the systems they need most.

Digital-by-default can become digital-or-nothing: efficient for the connected, a wall for everyone else.

Gone digital-only	Who is pushed out
Scheme applications	Anyone without a device, data or the literacy to use it
Grievance filing	The people with the most grievances
Appointment booking	Walk-in users, who were the majority
Payment and verification	Households where the phone belongs to someone else

Digital-by-default becomes digital-or-nothing wherever the offline channel is quietly withdrawn. It is rarely announced; the counter closes, the form goes out of print, and the staff who handled paper are reassigned. The policy says both channels exist.

The rule worth defending: for any essential service, the offline route must remain funded, staffed and advertised, not merely permitted. A channel nobody is told about and nobody is paid to run has been closed.

ImpactMojoDigital Ethics 101www.impactmojo.in

Intermediaries

When access depends on a middleman

Many people reach digital services only through an intermediary: a relative, a shopkeeper, a kiosk operator. That helps, but it means handing over passwords, OTPs and personal data, and trusting someone else with your transactions.

Dependence on intermediaries creates new risks: fees, errors, exclusion of those without a trusted helper, and fresh privacy exposure.

What the intermediary makes possible	What it costs
Completing a transaction at all	Handing over the OTP, and so the account
Reading and filling the form	Disclosure of income, health, family details
Reaching a service that is far away	An informal fee, unregulated and unrecorded
Help for those with a trusted relative	Nothing for those without one

Assisted use is the norm, not an edge case, and almost no system is designed for it. Interfaces assume a single user who is the account holder, so every real transaction involving a helper is technically a security violation.

Designing for it is possible and rare: delegated access with its own credential, a visible record of who acted, and a limit on what a helper can do. That is better than pretending assistance does not happen and leaving people to share passwords.

ImpactMojoDigital Ethics 101www.impactmojo.in

Design Response

Designing for the unconnected

Keep an offline / in-person channel for every essential service
Support local languages, voice and icons, not just English text
Build for low-end phones, patchy networks and assisted use
Measure who is left out, not just who is served

Design choice	Who it brings in
Voice and IVR alongside text	Non-readers, and anyone on a feature phone
Works on low-end devices and 2G	The poorest, and most rural users
Local languages, including script choice	Most of the country
Screen-reader compatibility and contrast	Users with disabilities, and everyone in sunlight
An assisted-use mode	The majority who need help, safely

Every row helps more people than the group it targets. Captions help in noisy places, high contrast helps in sunlight, low-bandwidth builds help everyone on a bad day. Accessibility work is rarely a minority accommodation in practice.

“Measure who is left out” is the instruction that makes the rest work. Usage statistics describe the people your design already fits, so they will never tell you about the people it excludes. That has to be sought deliberately.

ImpactMojoDigital Ethics 101www.impactmojo.in

Section Eight

Platform Power & Data Colonialism

ImpactMojoDigital Ethics 101www.impactmojo.in

The Gatekeepers

A few platforms shape the digital world

A handful of Big Tech firms run the search engines, app stores, social networks, cloud servers and operating systems that most digital life in South Asia depends on, and most of them are headquartered far away.

When the rails are owned by a few firms, they set the rules: what is visible, what is allowed, what data is taken, and what it costs to take part.

Layer	What it controls
Operating system and app store	What can be installed, and what a developer must pay
Search and feed ranking	What is findable, and therefore what exists in practice
Cloud hosting	Whether your service runs at all tomorrow
Payments and identity	Who can transact, and on what terms

Gatekeeping compounds down the stack. An organisation can change its content but not its ranking, change its app but not the store’s terms, change its provider but not at short notice. Each layer is a dependency with its own unilateral terms.

The jurisdictional point matters here. Most of these firms are headquartered elsewhere, so the rules governing a South Asian user’s experience are set under another country’s law and enforced through a support queue.

ImpactMojoDigital Ethics 101www.impactmojo.in

Concept

Data colonialism

The large-scale extraction of data from people and communities, often in the Global South, for the benefit of firms elsewhere, echoing older patterns of resource extraction.

The raw material is human life itself: behaviour, location, relationships, attention, mined, processed and monetised far from where it was produced.

The analogy holds	Where it strains
Raw material extracted; value added elsewhere	Data is not depleted by being taken
Terms set by the extracting party	Users receive a real service in return
Infrastructure and expertise stay abroad	No territorial control or coercion in the historical sense
The producing region holds little of the upside	Some local firms capture value too

Present both columns. The framing is analytically useful and it is a metaphor with limits; stating the limits is what stops it becoming a slogan and makes the underlying observation, that value flows in one direction, harder to dismiss.

The sharpest version of the claim is about AI training. Models are trained on data produced in one place, and sold as services back to it, which is close to the classic pattern, including in who holds the resulting capability.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Flow

Where the value goes

Users in the Global South generate data

→

Platforms collect & process it abroad

→

Profiles, ads & AI models are built

→

Profit & control accrue elsewhere

The people who produce the data rarely own it, see the profit, or shape the terms. The value flows one way.

ImpactMojoDigital Ethics 101www.impactmojo.in

Extraction

The Global South as a data source

Fast-growing user bases in South Asia are highly valuable: huge markets, rich data, and often weaker bargaining power and regulation. The region supplies users and data, but holds little of the infrastructure or the upside.

Cheap or 'free' services are paid for in data. The product offered for free is usually the user's attention and information.

What makes the region attractive	What it means locally
Very large, fast-growing user base	Scale gives the platform leverage in any dispute
Rich behavioural and language data	Valuable for training models sold back into the region
Lighter or newer regulation	Practices tested here before markets with older law
Price-sensitive users	“Free” wins, and free is paid for in data

The third row cuts both ways and is worth stating carefully. Newer regulation can also be better designed, having learned from elsewhere; the DPDP Act exists because that argument was won. The risk is the gap between a law passing and a regulator being able to act on it.

“Free” is the mechanism, not a bonus. Zero-price services win overwhelmingly among price-sensitive users, which means the population with least capacity to bear a data cost is the population most fully enrolled in paying it.

ImpactMojoDigital Ethics 101www.impactmojo.in

Asymmetry

Who captures the value

Platforms gain

Ad revenue and market dominance
Training data for AI models
Lock-in and network effects

Users get

A free service: for now
Little say over terms or data
No share of the value created

ImpactMojoDigital Ethics 101www.impactmojo.in

Dependence

When development runs on rented rails

NGOs and governments increasingly build on platforms they do not control: cloud servers, messaging apps, ad systems. A change in pricing, policy or access can disrupt critical services overnight.

Ask before you build: what happens to our beneficiaries if this platform changes its rules, raises its price, or shuts our account?

Dependency	What a change would do to your programme
A messaging app as the delivery channel	A policy change ends your outreach, without notice
A cloud provider holding beneficiary data	A pricing change or account action freezes operations
An ad platform for recruitment	A category ban removes your reach overnight
A proprietary data format	Migration costs exceed the value of moving

Ask the question before you build, not during the incident: what happens to the people we serve if this platform changes its rules, raises its price, or closes our account? A dependency with no answer is a single point of failure sitting outside your control.

Two cheap mitigations cover most of it: keep your own copy of essential data in an open format, and know the manual process that would run for a fortnight if the platform disappeared. Neither requires leaving the platform.

ImpactMojoDigital Ethics 101www.impactmojo.in

Sovereignty

Data sovereignty & local alternatives

Data sovereignty is the idea that a community or country should have meaningful control over data about its people. Public digital infrastructure and open-source tools are part of the response.

The goal is not autarky but agency: the ability to set terms, keep critical data local, and not be wholly dependent on a single foreign firm.

What data sovereignty can mean	Trade-off
Data localisation: stored in-country	Also makes it easier for the local state to reach
Public digital infrastructure	Reduces platform dependence; concentrates state power
Open-source and open standards	Portability, at the cost of engineering capacity to maintain it
Community control over community data	The strongest version, and the hardest to institutionalise

Sovereignty is not automatically protective. Moving data from a foreign firm to a domestic government changes who holds it, not whether the person it describes has any say, and for some populations the domestic state is the more immediate risk.

Agency is the better test than location. Can the people described set terms, see what is held, and refuse? A dataset that meets those conditions abroad is safer than one that fails them locally.

ImpactMojoDigital Ethics 101www.impactmojo.in

Practitioner Stance

Using platforms with eyes open

Prefer open, portable formats so you are not locked in
Keep your own copy of essential data, off the platform
Read the data terms: what does the platform take from your users?
Favour public or community infrastructure for critical services where viable

Practice	Effort
Export your data in an open format, on a schedule	A recurring job; set it once
Read the terms for what the platform takes from your users	An afternoon, once per platform
Prefer formats and standards you could move	A choice at procurement, free at that moment
Use public or community infrastructure for critical paths	Real work, and only where it is viable

The realistic stance is not refusal. Most organisations cannot leave the platforms their participants already use, and telling them to is advice nobody can act on. Using them with eyes open, and with an exit that exists, is what is actually available.

The second row is the one with an obligation attached. If you ask participants to reach you through a platform, you have enrolled them in its data terms, and you should be able to say what those terms are.

ImpactMojoDigital Ethics 101www.impactmojo.in

Section Nine

Misinformation & Online Harms

ImpactMojoDigital Ethics 101www.impactmojo.in

The Landscape

Information disorder in everyday life

False and misleading content spreads through the same channels people rely on for news, health advice and family contact. In South Asia, much of it flows through WhatsApp and other messaging apps.

Misinformation

False, but shared without intent to harm: a worried relative forwarding a 'cure'.

Disinformation

False and spread deliberately: to deceive, profit or incite.

Why the same message spreads further here
It arrives from a relative, so it carries their credibility rather than a source’s
There is no like-count, no label and no visible correction inside a closed group
Encryption hides the origin, so “who started this” has no answer
Fact-checking capacity is thinnest in the languages with the most users

The distinction between misinformation and disinformation decides the response. A worried relative is a target, not an author, and responding to them as though they ran the campaign loses the only channel into that group.

ImpactMojo’s Post-Truth Politics 101 covers this in depth: the taxonomy, the spread evidence and the correction technique. This section covers what a development organisation specifically has to do about it.

ImpactMojoDigital Ethics 101www.impactmojo.in

Why It Spreads

The mechanics of a rumour

Encrypted, closed groups make forwarded messages feel personal and trustworthy: they come from family, not a stranger. There is no visible source, no correction, and outrage travels faster than fact.

A message from your uncle's group carries the weight of your uncle, even when its content is false. Trust in the messenger launders the message.

Open platform	Closed group
Researchers can see what circulates	Nothing is observable from outside
Content can be labelled or down-ranked	There is nothing to label
A correction can reach the same audience	You cannot post into a group you are not in
Credibility comes from the source	Credibility comes from the sender

Trust in the messenger launders the message, and this is why closed-network falsehood is harder to counter than open-platform falsehood, not easier. No label competes with an uncle.

Which points at the one intervention that works: one competent, well-liked person inside the group who checks things. That is a community-mobilisation problem, which development organisations are unusually well placed to solve and platforms are not.

ImpactMojoDigital Ethics 101www.impactmojo.in

Real Harm

When rumours turn to violence

Viral rumours, for example false 'child-kidnapper' messages forwarded on WhatsApp, have been linked to mob violence and killings in India. Online falsehood becomes offline harm.

This is not a debate about opinions. Misinformation at scale can get people hurt and killed, and inflame communal tension.

What turned a message into violence	What has helped
A rumour circulating faster than any check	Forward limits, which cut velocity without reading content
No authoritative account for hours or days	Police and administration publishing verified facts early
A stranger present in an unfamiliar place	Local volunteers who can say “this is not from here”
Recaptioned footage from elsewhere	Reverse image search, taught at community level

These are documented, recurring events, not a hypothetical. False child-kidnapping messages forwarded through messaging apps have been linked to mob killings in India, and the pattern has repeated across states and years.

The second row is the intervention within reach of most organisations. Rumour fills a vacuum; saying something true, early and locally, even “here is what we know so far”, is the cheapest thing that reduces the space it fills.

ImpactMojoDigital Ethics 101www.impactmojo.in

Targeting the Poor

Scams that prey on the vulnerable

Fake 'KYC update' messages that steal banking OTPs
Bogus government-scheme and lottery messages demanding a fee
Loan-app traps with hidden charges and harassment
Phishing aimed at first-time, low-literacy internet users

New users with little digital experience and thin financial buffers are prime targets, and lose the most when caught.

Scam	Why new users are the target
Fake KYC-update messages	Real KYC requests exist, so the pretext is plausible
Bogus scheme or lottery fees	Genuine schemes do transfer money unexpectedly
Loan apps with hidden charges	Formal credit is unavailable, so the offer is welcome
OTP requests from a “bank officer”	Assisted use has trained people to share OTPs

Each scam borrows the shape of a legitimate process, which is why generic warnings do not work. “Never share your OTP” is undermined every time an intermediary legitimately needs it to complete a transaction on someone’s behalf.

Loan-app harassment deserves separate mention. Recovery practices have included contacting a borrower’s entire phone book, and the harm falls hardest on women, for whom the reputational threat is the leverage.

ImpactMojoDigital Ethics 101www.impactmojo.in

Health & Elections

Where false information does the most damage

Health

Vaccine rumours, fake cures and dangerous remedies cost lives, as seen vividly during COVID-19.

Democracy

Coordinated falsehoods can distort elections, inflame division and erode trust in institutions.

ImpactMojoDigital Ethics 101www.impactmojo.in

Moderation

Content moderation in Indian languages

Platforms moderate far better in English than in Hindi, Tamil, Bengali, Marathi and dozens of other languages. Harmful content in under-resourced languages and code-mixed scripts often slips through.

The places with the most users and the highest real-world stakes frequently get the least moderation investment. Language is a safety gap.

What makes moderation weak in a language	Consequence
Few trained human reviewers	Escalations sit unread, or are auto-closed
Little training data for classifiers	Automated detection performs poorly or not at all
Code-mixing and Roman script	Text falls outside whatever model exists
Local slurs and dog-whistles unmapped	The most dangerous content is the least detectable

Investment tracks revenue, not risk. The languages with the most users and the highest real-world stakes are frequently the least resourced, because moderation spending follows advertising value per user rather than potential harm.

The practical implication for organisations: do not assume a platform report will be actioned. Build a relationship with a trust-and-safety contact before you need one, and keep a record of what you reported and when.

ImpactMojoDigital Ethics 101www.impactmojo.in

Tension

Moderation vs free expression

Removing harmful content and protecting free speech pull against each other. Over-removal silences dissent and minorities; under-removal lets harm spread. Neither side is automatically right.

Beware moderation powers framed as safety that become tools to suppress legitimate criticism. Who decides, and who can appeal, matters as much as the rule.

Over-removal	Under-removal
Dissent and minority speech disappear first	Harmful content spreads unchecked
Automated systems misread satire and reclaimed slurs	Coordinated campaigns operate freely
Rules become a tool against critics	Targeted groups bear the cost
Appeal routes are slow or absent	Reports go unanswered

Both columns describe real failures, and neither side is automatically right. A deck that presented only one would be doing advocacy; the honest position is that this is a genuine trade-off with no setting that avoids both errors.

Watch for safety framing used against criticism. Powers created to remove genuinely harmful content are the powers later used on inconvenient reporting, which is why who decides and who can appeal matter as much as what the rule says.

ImpactMojoDigital Ethics 101www.impactmojo.in

What Helps

Responses that actually work

Digital literacy: teach people to pause, check and verify before forwarding
Friction: limits on mass-forwarding slow viral spread
Local fact-checking in the languages people actually use
Trusted messengers: community voices counter rumours best

Response	Evidence and cost
Forward limits and friction	Reduces velocity measurably; needs no content decisions
Prebunking: teach the technique before the lie	Generalises to claims not yet invented; effects decay
Local fact-checking in the right language	Essential, and always second to the rumour
Trusted community messengers	The strongest single factor; slow to build, cheap to keep

The fourth row is where development organisations already have the asset. Community health workers, self-help group leaders and teachers are trusted messengers who exist, are known and are reachable, which is exactly what no platform intervention can manufacture.

A concrete ask for a programme: add rumour monitoring to your existing community feedback channels, and name a person who can respond within a day. Both are cheap, and the second is what is missing when the day comes.

ImpactMojoDigital Ethics 101www.impactmojo.in

Section Ten

Responsible Design & Principles

ImpactMojoDigital Ethics 101www.impactmojo.in

The Shift

Ethics at the start, not the audit

Most digital harm is designed in early and discovered late. Responsible design moves ethical questions to the first meeting, where they are cheap to fix, not the launch review, where they are not.

You cannot bolt ethics on at the end. By then the data is collected, the model is trained and the exclusions are baked in.

Decision	Cost to change at design	Cost to change after launch
Which fields to collect	A conversation	Data already held; schema migration
Whether there is an offline route	A requirement line	New staffing, budget and policy
What the model optimises	A choice	Retraining, and a year of decisions to revisit
Whether decisions can be appealed	A workflow step	A new institution

This table is the argument for the whole section. Ethics late is not merely less effective; for several of these decisions it is effectively unavailable, because the cost of reversal exceeds what any organisation will spend on a harm that is already priced in.

The cheapest intervention available to a manager: put the five questions from Section 1 on the agenda of the first design meeting. It costs twenty minutes and it is the moment when all four rows above are still in the left column.

ImpactMojoDigital Ethics 101www.impactmojo.in

By Design

Privacy & ethics by design

Privacy by design

Building privacy protections into a system's architecture and defaults from the outset: rather than adding them later as an afterthought or an opt-in.

Privacy-protective defaults, not opt-out traps
Collect the minimum; delete when done
Security and access controls built in from day one

Default that protects	Default that harms
Sharing off unless chosen	Sharing on with an opt-out buried in settings
Retention set at collection	Keep indefinitely, review never
Access limited to those who need it	Whole-team access because it is simpler
Optional fields genuinely optional	Required fields with no operational purpose

Defaults decide outcomes for almost everyone. The share of users who change a default is small in every study of it, so the setting you ship is, in practice, the policy: regardless of what the consent screen offers.

The third row is the commonest failure inside NGOs. Everyone can see the beneficiary database because restricting it seemed bureaucratic, and access control is then impossible to introduce once people are used to it.

ImpactMojoDigital Ethics 101www.impactmojo.in

A Real Framework

The Principles for Digital Development

The Principles for Digital Development are a widely endorsed set of guidelines for using technology in development work: a real, sector-wide framework. First published in 2014, they were revised in 2024 into the nine principles below (digitalprinciples.org).

Understand the existing ecosystem; share, reuse and improve
Design with people; design for inclusion
Build for sustainability; establish people-first data practices
Create open and transparent practices; anticipate and mitigate harms; use evidence to improve outcomes

Principle	The question it forces
Design with people	Did anyone affected see this before it was built?
Understand the existing ecosystem	What is already working that we are about to replace?
Build for sustainability	Who runs this when the grant ends?
Use evidence to improve outcomes	What would tell us this is not working?
Establish people-first data practices; anticipate and mitigate harms	What is the worst use of this data, and by whom?
Share, reuse and improve; create open and transparent practices	Could someone else take this over?

The Principles are a real sector framework, endorsed by many donors, which makes them useful in a proposal as well as a design meeting. Citing them converts an ethical argument into a compliance argument, which is frequently the argument that wins.

Their weakness is that they are easy to claim. Every proposal says it designs with people; the right-hand column is there because a specific answerable question is harder to assert falsely than a principle.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Rule

Do no (digital) harm

The humanitarian principle of do no harm applies fully to data and technology. Before deploying, ask how the system could be misused, and who would be hurt if it were.

Map who could be harmed

→

Map how data could be misused

→

Design the safeguard & the fallback

→

Decide: is the benefit worth the residual risk?

Ask	Concretely
Who could be harmed?	Name groups, not “users”: women, minorities, undocumented migrants
How could the data be misused?	By a future government, an employer, a family member, a hostile neighbour
What is the safeguard?	Something in the architecture, not the policy document
Is the residual risk worth it?	A decision someone signs, on a date, in writing

The second row is where most exercises are too polite. The realistic misuse case is often a family member with access to a woman’s phone, or a local official with a grudge, not a sophisticated attacker, and not the scenario a security review covers.

The fourth row is what makes it accountable. Residual risk is unavoidable; an unnamed acceptance of it is not. Record who decided and on what basis, so that the decision can be revisited when the context changes.

ImpactMojoDigital Ethics 101www.impactmojo.in

Core Practice

Data minimisation as a habit

The single most protective habit is to collect less. Every field you do not gather is a breach that cannot happen and a tool that cannot be turned against people later.

Collect less

fewer fields, less risk, less liability

Keep less

delete on a schedule; don't hoard 'just in case'

Field routinely collected	Is it used?
Caste	Only if disaggregation is actually reported; otherwise a liability
Full date of birth	Age band is almost always sufficient
Precise GPS of a household	Village or block is usually enough, and far safer
Phone number of every member	One contact serves the purpose
Photograph	Rarely used after collection, and hard to secure

Run this audit on your own current survey form. Most organisations find several fields that nobody has analysed in years, collected because the template had them, and each one is a breach that could not have happened if it were not there.

Coarsening is the underused option. Age bands rather than dates, block rather than GPS, presence of a condition rather than a diagnosis: the analysis usually survives, and the re-identification risk falls sharply.

ImpactMojoDigital Ethics 101www.impactmojo.in

Participation

Design with, not for

The people affected by a system know its failure modes best. Involving them in design surfaces exclusions, language gaps and risks that outsiders simply cannot see.

Nothing about us without us. Co-design is not a courtesy: it is how you find the harms before they reach a million people.

What participation surfaces	That outsiders miss
Who in the household controls the phone	Designs that assume personal device access
Which words the interface should use	Translations that are technically correct and unread
When people can actually attend or transact	Office hours that assume no wage work
What a failure costs them	Retry flows designed for people with time

“Nothing about us without us” is a design method here, not a slogan. Each row above is knowledge that exists only in the affected population and is not recoverable by testing, because a test tells you that something failed and not why.

Participation done badly is worse than none: a consultation held after the build, with people selected by the implementer, produces endorsement rather than information, and the endorsement is then cited.

ImpactMojoDigital Ethics 101www.impactmojo.in

Accountability

Build in redress from the start

A clear, reachable way to contest an automated decision
A human who can override the system when it is wrong
Plain-language notice of what data is held and why
An exit: people can correct, delete, or refuse

A system with no appeal route is not efficient: it is unaccountable.

Redress element	Test of whether it is real
A route to contest a decision	Usable by someone without a smartphone or literacy
A human who can override	They have the authority and are not penalised for using it
Plain-language notice	In the language spoken, at the moment data is taken
An exit	Correction and deletion actually happen, within a stated time

A system with no appeal is not efficient; it is unaccountable. Efficiency claims usually count the cost of handling appeals and not the cost of wrong decisions that stand, because the second falls entirely on people outside the organisation.

Redress is also your best source of failure data. Every appeal is a report of a case the design got wrong, which makes a working grievance channel the cheapest quality-assurance mechanism a programme can have.

ImpactMojoDigital Ethics 101www.impactmojo.in

Section Eleven

Governance, Regulation & Practice

ImpactMojoDigital Ethics 101www.impactmojo.in

Who Governs

The layers that hold systems to account

LAW: the DPDP Act & constitutional rights

→

REGULATOR: a data-protection authority

→

ORGANISATION: your own policies & reviews

→

PRACTITIONER: the daily choices you make

Governance is not only the regulator's job. The last and most frequent line of defence is the person designing the form.

Layer	What it can do	What it cannot
Law	Set rights and obligations	Notice your specific form
Regulator	Investigate, penalise	Reach most small deployments
Organisation	Policy, review, training	Survive a change of leadership without structure
Practitioner	Decide what goes on the form	Fix a system designed elsewhere

The last layer acts most often and is trained least. The person deciding which fields go on a survey makes hundreds of small data-protection decisions a year, and usually receives no guidance at all on any of them.

Which is where a short internal standard pays. Two pages saying what is never collected, what is always deleted and who to ask does more, in aggregate, than a compliance review that happens annually.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Authority

Data protection authorities

The DPDP Act establishes a Data Protection Board of India to handle breaches and grievances. Globally, independent data protection authorities enforce rules, investigate complaints and impose penalties.

A law is only as strong as the body that enforces it. Watch for independence, resources and real power to act, not just words on paper.

What makes an authority effective	Warning sign
Independence in appointment and removal	Members serve at the pleasure of the executive
Adequate staff and technical capacity	A handful of officers for a national caseload
Power to compel and to penalise	Recommendations only
Broad remit	Wide exemptions for state processing

A law is only as strong as the body enforcing it, and these four properties, not the wording of the statute, determine whether a data-protection regime does anything. They are also observable, so the assessment does not require legal training.

Check the current position before relying on this slide. The composition, powers and exemptions under the DPDP framework have been contested and are still settling; what is written here is the shape of the question, not a snapshot of the answer.

ImpactMojoDigital Ethics 101www.impactmojo.in

Rights

The rights people hold over their data

Access: know what data is held about you
Correction: fix what is wrong
Erasure: have data deleted when no longer needed
Grievance: complain and seek redress

Help the communities you work with understand these rights. A right nobody knows about protects nobody.

Right	What it looks like when exercised
Access	“Show me what you hold about me and who you shared it with”
Correction	“My name is spelled two ways; fix the record”
Erasure	“The programme ended; delete my data”
Grievance	“I was refused and nobody will tell me why”

A right nobody knows about protects nobody, and communities you work with are unlikely to hear about these anywhere else. Explaining them is squarely within what a field organisation can do and rarely does.

Test your own readiness first. If a participant made an access request tomorrow, could you answer it, and how long would it take? Most organisations discover the answer is no and several weeks, which is itself the finding.

ImpactMojoDigital Ethics 101www.impactmojo.in

Checklist

An organisation's digital-ethics checklist

Area	Ask before you deploy
Purpose	Is each data field truly necessary?
Consent	Is it free, informed, specific, revocable?
Inclusion	Who does this design exclude, and is there a fallback?
Bias	Have we tested outcomes across groups?
Security	Who can access this, and is it protected?
Redress	Can a person contest a decision and reach a human?

ImpactMojoDigital Ethics 101www.impactmojo.in

Procurement

Ethics when you buy, not just when you build

Most organisations buy or adopt tools rather than build them. The ethics questions still apply: ask the vendor, and make the answers part of the contract.

Where is data stored, and who can access it?
Has the tool been tested for bias and accessibility?
What happens to our data if we leave the vendor?

Ask the vendor	Put in the contract
Where is data stored, and who can access it?	Named jurisdictions; a list of sub-processors
Has it been tested for bias and accessibility?	Disaggregated performance, supplied annually
What happens to our data if we leave?	Export in an open format; deletion certified
What is the uptime and the fallback?	An agreed manual process during outages
Who is liable when it is wrong?	A clause, rather than an assurance in a meeting

Most organisations buy rather than build, so procurement is where most of their digital ethics is actually decided: usually by people who were not in any of the conversations this deck describes.

The right-hand column is the one that matters. A verbal assurance is worth nothing after the account manager changes; the same commitment as a contract term survives, and costs nothing extra to ask for before signature.

ImpactMojoDigital Ethics 101www.impactmojo.in

The Takeaways

If you remember five things

Behind every record is a person: usually with little power to refuse
Collect less: minimisation is the strongest safeguard
Design for the person the system fails, not the headline coverage
Bias is inherited from data and proxies: audit, don't assume
Keep a human accountable and an appeal route open

If you do one thing	Where it is in this deck
Delete a field from your survey form	Sections 2 and 10
Ask a vendor for disaggregated error rates	Sections 3 and 11
Build a working manual fallback	Section 5
Set a deletion date on an existing dataset	Sections 2 and 6
Write down what happens to someone who says no	Sections 2 and 5

Every item is inside the authority of an ordinary programme manager, which is deliberate. A deck about a structural problem that recommends only structural change leaves its reader with nothing to do on Monday morning.

And the structural point stands alongside it. Individual care does not substitute for law, regulators and platform accountability; it is what fills the gap while those are built, and it is the part you control.

ImpactMojoDigital Ethics 101www.impactmojo.in

Keep Learning

Source	Read it for
Eubanks, Automating Inequality	How welfare systems automate exclusion: closest to Sections 5–6
O’Neil, Weapons of Math Destruction	Feedback loops and opaque scoring, in plain language
D’Ignazio & Klein, Data Feminism	Power in data collection and classification
Couldry & Mejias, The Costs of Connection	The data-colonialism argument, from its authors
Principles for Digital Development	The sector framework, usable in a proposal