Fairness
Fairness is the principle or property of allocating treatment, opportunities, or outcomes equitably among relevant people, groups, or participants according to appropriate criteria.
Think Like a Software Engineering Manager
Fairness is the equitable and impartial treatment of people according to consistent standards and processes.
- How it is used
- Guides performance assessments, delegation, recognition, and equitable workplace treatment.
- How deep it goes
- Provides practical checks for standards, bias, appeals, and recognition while acknowledging that no single bias-reduction approach fits every situation.
- Examples
- Uses two employees’ performance assessments, caregiving accommodations, delegated intern mentoring, and recognition of technical-debt work.
- What you can do after
- Practice/perform fairer decisions by applying shared criteria, distributing opportunities, reviewing recognition patterns, and providing routes to appeal.
Read how Think Like a Software Engineering Manager explains fairness →
Generative AI in Action
Fairness is the property of treating people equitably and avoiding unjustified disparities in the outcomes or treatment produced by an AI system.
- How it is used
- Guides bias prevention and harm mitigation when designing and deploying enterprise generative AI systems.
- How deep it goes
- Defines sources and dimensions of bias and outlines layered mitigations, without specifying fairness metrics or a worked audit procedure.
- Examples
- Uses RAG bias risk and names Fairlearn, InterpretML, and IBM AI Fairness 360 as relevant ML toolkits.
- What you can do after
- Apply it via Fairlearn or IBM AI Fairness 360 to investigate ML-model bias, while combining data, output, and human safeguards.
Starting Data Analytics with Generative AI and Python
The property of treating relevant individuals or groups equitably and avoiding unjustified disparities in AI systems or their use.
- How it is used
- Guides responsible organizational use of generative AI through policy, team practices, ethical guidelines, and continuing review.
- How deep it goes
- Connects fairness to bias mitigation and lifecycle governance, but stops short of fairness metrics or technical model-internals analysis.
- Examples
- Uses a concise AI-use policy, diverse review teams, ethical guidelines, and continuous model monitoring as practical safeguards.
- What you can do after
- Apply fairness safeguards by writing policies, assembling diverse reviewers, defining ethical practices, and reassessing models and approved uses.
Read how Starting Data Analytics with Generative AI and Python explains fairness →
Introduction to Generative AI, Second Edition
Fairness is the property of treating individuals and groups equitably in an algorithmic system's design, operation, and outcomes.
- How it is used
- Positions fairness as a responsible-AI goal requiring governance, safeguards, oversight, and regulation.
- How deep it goes
- Frames fairness as a governance concern but stops short of defining tests, metrics, or operational procedures for measuring equitable treatment.
- Examples
- Contrasts voluntary AI-company self-policing with comprehensive regulation and highlights globally uneven participation in AI policy discussions.
Read how Introduction to Generative AI, Second Edition explains fairness →
Machine Learning System Design
The property that a model’s behavior or performance does not produce unjustified disparities across relevant demographic or other groups.
- How it is used
- Evaluates whether model improvements reduce error evenly across samples and monitors performance disparities across subpopulations.
- How deep it goes
- Defines residual inequality, demonstrates distribution comparisons, and gives Gini calculation; monitoring guidance stops at segment tracking.
- Examples
- Uses two ten-sample models with equal MAE gains, residual-distribution figures, and Gini coefficient geometry.
- What you can do after
- Evaluate model changes with residual distributions and Gini index, then monitor segment-level performance for discrepancies.
Inside AI
Fairness is the property of treating people equitably and avoiding unjustified disparities in system outcomes.
- How it is used
- Applies fairness when evaluating AI systems that affect citizens, alongside reliability, privacy, transparency, and explicability.
- How deep it goes
- Defines fairness through equitable treatment and bias avoidance, but does not specify metrics or implementation procedures.
- Examples
- Uses criminal-justice decisions about incarceration or probation, including medication and social support as alternatives.
- What you can do after
- Evaluate AI decisions affecting citizens by checking for bias, comprehensible logic, and explainability before deployment.
Causal AI
Fairness is the property that an algorithmic or statistical system treats individuals or groups according to specified equity criteria.
- How it is used
- Analyzes predictive algorithms by separating causal sources of statistical bias and informing fairer model design.
- How deep it goes
- Distinguishes direct penalties from indirect targeting pathways, but stops short of resolving normative definitions and fairness tradeoffs.
- Examples
- Uses loan repayment prediction, gendered digital-ad targeting, and a causal DAG to examine competing fairness interpretations.
- What you can do after
- Evaluate fairness by tracing bias to causal sources and designing models around variables causally relevant to repayment.
Building Reliable AI Systems
The property of an AI system treating people and groups equitably and avoiding unjustified disparities in outcomes.
- How it is used
- Treats fairness as a reliability concern enforced during fine-tuning and checked through production evaluation.
- How deep it goes
- Explains a group-gap training penalty and recommends group-wise audits and metrics, without specifying decision thresholds.
- Examples
- Uses unequal response lengths for different names and a LoRA loss penalty, then recommends demographic-group test sets.
- What you can do after
- Apply fairness through a LoRA fine-tuning loss and evaluate performance separately across demographic groups.
The Art of AI Product Development
Fairness is the property of treating relevant people or groups equitably in an AI system's behavior and outcomes.
- How it is used
- Uses fairness to guide AI safeguards, governance decisions, stakeholder communication, and regulatory compliance.
- How deep it goes
- Explains data bias, audits, monitoring, safeguards, and regulation, but stops short of detailed fairness metrics or implementation procedures.
- Examples
- Uses Amazon hiring bias and regional or seasonal prediction imbalance to show how unfair outcomes emerge and can be detected.
- What you can do after
- Apply fairness checks through diverse data, bias detection, audits, explainability, monitoring, and safeguards for affected groups.
Read how The Art of AI Product Development explains fairness →
Financial AI in Practice
The principle that decisions and outcomes should treat affected people or groups equitably and avoid unjustified discrimination.
- How it is used
- Constrains AI development and deployment in finance, where decisions have immediate effects and face regulatory scrutiny.
- How deep it goes
- Names fairness as a requirement but does not define measures, tests, or implementation steps.
Related concepts
- bias detectionmeasured by · Bias detection is the systematic assessment of data, decisions, or system behavior to identify distortions or unjustified disparities across groups or conditions.
- machine learningpart of · Machine learning is a field of computational methods in which systems learn patterns or relationships from data to make predictions, decisions, or other outputs.
- livenesskind of · Liveness is a correctness property requiring that a system eventually makes specified progress or produces a desired event or condition.
- Gini impuritymeasured by · Gini impurity is a measure of class-label heterogeneity in a set or decision-tree node, calculated from the proportions of its classes.
- diversity, equity, and inclusionpart of · Diversity, equity, and inclusion is an organizational principle and practice promoting representation, fairness, participation, and belonging for people with different identities and backgrounds.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.