Bias detection and mitigation, fairness metrics, privacy frameworks, consent models, transparency requirements, and accountability structures for data science practice. Covers algorithmic bias sources, disparate impact testing, differential privacy, GDPR principles, model cards, datasheets for datasets, responsible AI frameworks, and the organizational governance needed to make ethics actionable. Use when auditing models for bias, designing privacy-preserving systems, establishing governance processes, or evaluating the social impact of data-driven decisions.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Bias detection and mitigation, fairness metrics, privacy frameworks, consent models, transparency requirements, and accountability structures for data science practice. Covers algorithmic bias sources, disparate impact testing, differential privacy, GDPR principles, model cards, datasheets for datasets, responsible AI frameworks, and the organizational governance needed to make ethics actionable. Use when auditing models for bias, designing privacy-preserving systems, establishing governance processes, or evaluating the social impact of data-driven decisions.
Data science operates on people's data, affects people's lives, and encodes human decisions into automated systems. Ethics in data science is not an afterthought or a compliance checkbox -- it is a design requirement. Ruha Benjamin's concept of the "New Jim Code" names the reality that automated systems can reproduce and amplify existing social inequalities while appearing objective. This skill covers the principles, frameworks, and practices that make ethical data science concrete and actionable.
Bias can enter at every stage of the data science workflow. It is not a property of algorithms alone -- it is a property of the system: data, design decisions, deployment context, and feedback loops.
Stage
Bias type
Example
Problem formulation
Framing bias
Defining "success" as engagement maximizes addictive behavior
Data collection
Selection bias
Training a facial recognition system on predominantly light-skinned faces
Data labeling
Annotation bias
Labelers' cultural assumptions influence what counts as "toxic" speech
Feature engineering
Proxy bias
ZIP code encodes race due to residential segregation
Model training
Optimization bias
Minimizing overall error ignores disparate performance across subgroups
Evaluation
Metric bias
Reporting aggregate accuracy hides poor performance on minority groups
Deployment
Automation bias
Decision-makers defer to model output without scrutiny
Feedback loops
Amplification bias
Predictive policing increases patrols in targeted areas, generating more arrests, confirming the model
Historical Bias vs. Representation Bias
Historical bias: The world is unequal, and data reflects that inequality. A hiring model trained on historical decisions inherits past discrimination. Even a "perfect" model of biased reality produces biased outputs.
Representation bias: The training data does not represent the deployment population. A speech recognition system trained on American English performs poorly on other dialects. This is not a bug in the algorithm -- it is a gap in the data.
Both are real. Neither is solved by "better algorithms" alone. The fix requires changes to data collection, problem formulation, and deployment monitoring.
Fairness Metrics
Impossibility Theorem
Chouldechova (2017) and Kleinberg, Mullainathan, and Raghavan (2016) independently proved that three natural fairness criteria cannot all be satisfied simultaneously when base rates differ between groups:
Calibration: Among those predicted positive, the fraction truly positive is the same across groups.
Equal false positive rate: The rate of incorrectly predicting positive is the same across groups.
Equal false negative rate: The rate of incorrectly predicting negative is the same across groups.
When the base rate (actual positive rate) differs between groups, satisfying any two of these requires violating the third. This is not a technical limitation to be solved -- it is a value judgment about which type of error matters more. The choice must be made explicitly, not hidden inside a loss function.
Common Fairness Definitions
Metric
Definition
When appropriate
Demographic parity
P(positive prediction) is equal across groups
When the prediction itself causes differential treatment
Equalized odds
TPR and FPR are equal across groups
When false positives and false negatives have different costs
Equal opportunity
TPR is equal across groups (weaker than equalized odds)
When false negatives are the primary concern (e.g., loan approval for qualified applicants)
Predictive parity
Precision is equal across groups
When the model's positive predictions trigger consequential actions
Individual fairness
Similar individuals receive similar predictions
When you can define a meaningful similarity metric
Counterfactual fairness
Prediction would be the same in a counterfactual world where the individual belonged to a different group
When causal reasoning is possible and the causal model is trusted
Measuring Disparate Impact
The four-fifths rule (EEOC, 1978): if the selection rate for a protected group is less than 80% of the rate for the most-selected group, there is evidence of adverse impact.
Disparate impact ratio = (selection rate for protected group) / (selection rate for most-selected group)
If this ratio < 0.8, investigate. This is a screening heuristic, not a legal standard -- but it is widely used as a first check.
Privacy
Privacy Principles (GDPR Framework)
Principle
Meaning
Practical implication
Lawfulness
Legal basis for processing
Document the legal basis (consent, legitimate interest, contract, etc.)
Purpose limitation
Collect for specified purposes only
Do not repurpose data without new consent or legal basis
Data minimization
Collect only what is necessary
Every field in the dataset should have a documented purpose
Accuracy
Keep data correct and current
Provide mechanisms for correction; audit data quality
Replace specific values with ranges (age 34 -> 30-39)
Reduces data utility
k-Anonymity
Every record is indistinguishable from at least k-1 others on quasi-identifiers
Vulnerable to homogeneity and background knowledge attacks
l-Diversity
Each equivalence class has at least l distinct sensitive values
Better than k-anonymity but still vulnerable
t-Closeness
Distribution of sensitive attribute in each class is close to overall distribution
Strong but complex to implement
Differential privacy
Add calibrated noise so no individual's inclusion changes the output significantly
Mathematical guarantee; degrades with composition
Differential Privacy
Differential privacy (Dwork, 2006) provides a mathematical guarantee: the output of an analysis is approximately the same whether any individual is in the dataset or not.
The privacy budget epsilon controls the tradeoff: smaller epsilon = stronger privacy = more noise = less accuracy. Epsilon is spent with each query, and it does not regenerate -- this is the composition theorem.
Practical deployment: Apple (emoji usage), Google (Chrome usage), US Census (2020). Each chose an epsilon value that balanced utility and privacy for their specific context. There is no universally "correct" epsilon.
Consent
Informed Consent Requirements
Purpose: What the data will be used for, in plain language.
Scope: What data is collected and how long it is retained.
Rights: How to access, correct, or delete data.
Risks: Potential consequences of participation.
Voluntariness: Participation is optional; no penalty for declining.
Third parties: Whether data is shared and with whom.
Consent Models
Model
Description
Strength
Weakness
Opt-in
User actively agrees before data is collected
Respects autonomy
Lower participation rates
Opt-out
Data is collected by default; user can withdraw
Higher participation
Default bias; many users never opt out
Dynamic consent
Ongoing, granular control over data uses
Maximum user control
Complex to implement; user fatigue
Broad consent
Consent for a category of future uses
Enables secondary research
Vague; user may not understand implications
Tiered consent
Multiple options (e.g., anonymized only, full research use)
User chooses comfort level
More complex consent forms
When Consent Is Insufficient
Consent does not make harmful uses ethical. If a system causes disparate impact, user consent to data collection does not excuse the harm. Similarly, consent from one population does not extend to another. Consent is necessary but not sufficient.
Transparency and Accountability
Model Cards
Mitchell et al. (2019) proposed model cards as standardized documentation for deployed models:
Performance metrics overall and disaggregated by group
Training data
Source, size, preprocessing, known limitations
Evaluation data
Source, demographics, relationship to deployment population
Ethical considerations
Known biases, risks, mitigation strategies
Caveats and recommendations
Known limitations, suggested monitoring
Datasheets for Datasets
Gebru et al. (2021) proposed datasheets as standardized documentation for datasets, covering: motivation, composition, collection process, preprocessing, uses, distribution, and maintenance.
Why this matters: A model is only as good as its data. Without dataset documentation, users cannot assess whether the data is appropriate for their task.
Algorithmic Impact Assessment
Before deploying a model that affects people's lives:
Identify affected populations. Who does this system affect? Who is most vulnerable?
Assess potential harms. What happens when the model is wrong? Are errors distributed equitably?
Evaluate alternatives. Is automation necessary? Would a simpler rule or human judgment be more appropriate?
Plan monitoring. How will you detect degradation, drift, or emerging bias after deployment?
Establish recourse. How can affected individuals challenge or appeal automated decisions?
Organizational Governance
Building an Ethics Practice
Ethics is not a one-time review. It requires ongoing organizational commitment:
Ethics review board: Multidisciplinary group (data scientists, ethicists, domain experts, community representatives) that reviews high-risk projects before deployment.
Bias bounty programs: Internal or external programs that reward finding bias in deployed systems.
Incident response: Process for handling ethical incidents (biased output discovered in production, privacy breach, consent violation).
Training: Regular ethics training for all team members, not just compliance officers.
Diverse teams: Teams with diverse backgrounds are more likely to identify blind spots in data, design, and deployment.
Responsible AI Frameworks
Framework
Source
Key contribution
Fairness, Accountability, Transparency (FAccT)
ACM Conference
Academic research community and standards
Responsible AI Practices
Google
Practical guidelines for industry
Ethics Guidelines for Trustworthy AI
EU High-Level Expert Group
Seven requirements including human agency, robustness, privacy
Algorithmic Accountability Act
US proposed legislation
Requires impact assessments for automated critical decisions
Race After Technology
Ruha Benjamin
Critical examination of how technology reproduces racial inequality
Common Mistakes
Mistake
Why it fails
Fix
"The algorithm is objective"
Algorithms encode human decisions in data, features, and metrics
Audit for bias; algorithms inherit their creators' and data's biases
Ethics as compliance checkbox
Checking a box does not prevent harm
Build ethics into the design process, not just the review process
Aggregate metrics only
95% accuracy overall can mask 70% accuracy for a subgroup
Always disaggregate metrics by protected groups
Anonymization = privacy
Re-identification is possible from "anonymized" data
Use differential privacy or formal anonymization guarantees
"We didn't intend bias"
Intent does not determine impact
Measure impact, not intent
Consent theater
Long, unreadable consent forms do not constitute informed consent
