liveBook · Manning

Bias

Also called biases

Bias is a systematic tendency or deviation that distorts measurements, estimates, judgments, or outcomes relative to a relevant reference or target.

How 38 Manning books explain it

Grokking Machine Learning

Systematic distortion in data or model behavior that can produce unfair or unrepresentative outcomes.

How it is used
Uses bias as the constant that helps regression predict prices and perceptrons classify sentences, while distinguishing historical bias in data.
How deep it goes
Defines the constant, illustrates its effect on regression lines and classification thresholds, and sketches repeated updates rather than deriving them here.
Examples
Uses a $100 house base price, a word-count sentiment threshold, and historical Titanic passenger data.
What you can do after
Build it from scratch by adjusting a regression model’s bias alongside its weights to improve predictions.

Read how Grokking Machine Learning explains bias →

Inside AI

AI bias is systematic unfairness or distortion in an artificial-intelligence system's outputs or decisions.

How it is used
Highlights fairness risks when AI informs consequential decisions about people.
How deep it goes
Explains algorithmic and data sources of bias through examples, and calls for audits without detailing an audit procedure.
Examples
Illustrates bias with UK A-level grading, criminal-justice risk scores, and COVID-19 imaging models.
What you can do after
Evaluate AI applications for potential unfairness by considering data, algorithm design, and consequences.

Read how Inside AI explains bias →

Natural Language Processing in Action, Second Edition

Word vector bias is systematic distortion in learned word associations caused by stereotypes and imbalances in the training corpus.

How it is used
Connects corpus stereotypes to biased word associations and their consequences for NLP systems.
How deep it goes
Explains corpus-dependent associations and compares word distances, while identifying bias compensation as a challenge rather than detailing a mitigation procedure.
Examples
Compares “man” and “woman” distances to “nurse” and contrasts finance and geology meanings of “bank.”
What you can do after
Evaluate learned associations by comparing word-model distances and considering the training corpus.

Read how Natural Language Processing in Action, Second Edition explains bias →

LLMs in Production

A systematic pattern in model outputs that reflects skewed data or assumptions and produces inaccurate or unfair representations.

How it is used
Treats bias as a data-quality and responsible-AI concern in language-model development.
How deep it goes
Explains training-data origins and group-based evaluation with a metric example, while stopping short of a comprehensive mitigation procedure.
Examples
Illustrates skewed leader images and compares gendered prompt completions with WinoBias and HONEST.
What you can do after
Apply it via HONEST to compare the hurtfulness of supplied completions across gender groups.

Read how LLMs in Production explains bias →

Grokking Statistics

Model bias is systematic prediction error arising when a model's rigid assumptions prevent it from adequately representing the data.

How it is used
Use bias to question whether data, metrics, and predictive models support the claims made from them.
How deep it goes
Defines sampling and model bias, then explains self-selection, measurement choices, and the bias-variance tradeoff without formal derivation.
Examples
Uses family photographs, a dorm sleep survey, 3D-printer filament measurements, and a toddler-meltdown model.
What you can do after
Evaluate/understand whether a dataset or model warrants its intended claim by checking representation, context, and rigidity.

Read how Grokking Statistics explains bias →

Demand Forecasting Best Practices

Forecast bias is the systematic tendency of forecasts to overestimate or underestimate actual demand.

How it is used
Evaluates whether demand forecasts systematically over- or under-estimate demand alongside accuracy for planning decisions.
How deep it goes
Defines bias, derives it from average forecast error, works an example, and explains metric and process limitations.
Examples
Uses a four-period demand-and-forecast table and contrasts accurate biased forecasts with inaccurate unbiased forecasts.
What you can do after
Apply it via Excel or KPI reporting to calculate bias, pair it with MAE, and investigate persistent over- or under-forecasting.

Read how Demand Forecasting Best Practices explains bias →

AI Model Evaluation

A bias is a learnable additive parameter that shifts a model computation independently of its inputs.

How it is used
Identifies bias as a learnable parameter within the anatomy of an AI model.
How deep it goes
Mentions biases alongside weights but does not explain their calculation or training mechanics.
Examples
Names personalized recommendations as a model application without illustrating a bias parameter.

Read how AI Model Evaluation explains bias →

Starting Data Analytics with Generative AI and Python

Bias is a systematic preference or assumption that can distort the selection or interpretation of information, including generative-model outputs.

How it is used
Frames bias as a data-quality and fairness risk in generative AI-assisted analysis and decisions.
How deep it goes
Classifies sources of bias and outlines monitoring and mitigation techniques, while emphasizing domain-specific judgment and limits of post-processing.
Examples
Uses unrepresentative customer reviews, keyword selection, and residential loan underwriting to illustrate potential bias.
What you can do after
Evaluate generative AI outputs for possible bias and consider monitoring, provider feedback, or cautious post-processing.

Read how Starting Data Analytics with Generative AI and Python explains bias →

The Art of AI Product Development

Bias is a systematic skew in data, model behavior, or outcomes that produces unequal or inaccurate results.

How it is used
Frames bias mitigation as an AI governance responsibility spanning product development, deployment, and stakeholder communication.
How deep it goes
Details data audits, fairness testing, explainability, human review, and monitoring without mathematical derivations.
Examples
Uses applicant screening and customer-support chatbot scenarios to illustrate indirect discrimination and gaps in representation.
What you can do after
Practice identifying bias sources and planning checks across data, model outcomes, and user feedback.

Read how The Art of AI Product Development explains bias →

Effective Data Analysis

Bias is a systematic pattern of error or unequal treatment that produces unfair or disadvantageous outcomes for particular groups.

How it is used
Uses bias to assess the risks of data-informed and automated decisions for affected groups.
How deep it goes
Illustrates how data and rules can lead to biased outcomes, without specifying a quantitative fairness measure or audit procedure.
Examples
Illustrates bias with a resume-ranking model and a support-ticket categorization system.
What you can do after
Evaluate how automated decisions might disadvantage groups and why their outputs require caution.

Read how Effective Data Analysis explains bias →

Hacking Cryptography

A systematic deviation in the probabilities of RC4 key-scheduling or keystream outcomes from the uniform behavior expected under the ideal model.

How it is used
Motivates why nonuniform random choices and RC4 outputs enable practical key guessing.
How deep it goes
Works through the FMS procedure from weak-IV construction and KSA equations to frequency-based key recovery.
Examples
Uses date-shaped passwords and WEP packets with [L, 255, X] IVs, including high- and low-volume recovery runs.
What you can do after
Diagnose/debug with it by recognizing when repeated observations and frequency counts can expose biased key or keystream values.

Read how Hacking Cryptography explains bias →

Introduction to Generative AI, Second Edition

Bias in an AI model is a systematic tendency for its outputs or errors to reflect unfair, skewed, or otherwise undesirable patterns.

How it is used
Frames identity-based disparities as a safety and fairness risk in large language models.
How deep it goes
Defines disparate outputs and traces sources and consequences, while stopping short of a detailed mitigation procedure.
Examples
Uses a resume-screening audit and outdated responses about gender identity or climate activism to illustrate disparities and gaps.
What you can do after
Supports evaluating how model outputs may reinforce discrimination in decision-making.

Read how Introduction to Generative AI, Second Edition explains bias →

Machines that Think

A systematic distortion in data, model behavior, or decisions that produces uneven or unjustified outcomes.

How it is used
Explains how inherited distortions can shape model behavior and consequential decisions.
How deep it goes
Explains sources, auditing practices, and mitigation points while noting that benchmarks and interventions have limits.
Examples
Uses hiring by school, résumé name changes, and demographic differences in generated occupations and conversational tone.
What you can do after
Evaluate and understand possible bias by comparing cases, inspecting subgroup outcomes, and interpreting benchmarks cautiously.

Read how Machines that Think explains bias →

Machine Learning for Tabular Data: XGBoost, Deep Learning, and AI

Bias is systematic prediction error arising from restrictive modeling assumptions or insufficient model flexibility.

How it is used
Guides model and ensemble choices by weighing systematic prediction error against variance and generalization.
How deep it goes
Defines bias intuitively and traces its trade-offs through linear features, regularization, sampling, and randomized trees without formal derivation.
Examples
Uses transformed linear-regression features, bagged models, and Airbnb price classification with extra-trees.
What you can do after
Evaluate/understand whether model simplification, sampling, or tree randomization exchanges prediction bias for lower variance appropriately.

Read how Machine Learning for Tabular Data: XGBoost, Deep Learning, and AI explains bias →

Experimentation for Engineers

A systematic measurement error that consistently shifts results above or below the value being estimated.

How it is used
Diagnoses systematic distortions in experimental measurements, business-metric predictions, and adaptive decision logs.
How deep it goes
Explains sampling and model bias through numerical simulations and feedback loops, then gives safeguards for common experimental biases.
Examples
Uses exchange execution costs, time-of-day effects, missing recommendation outcomes, and refitted policies as concrete cases.
What you can do after
Diagnose bias with randomization, replication, exploration, peer review, holdouts, and reverse A/B tests.

Read how Experimentation for Engineers explains bias →

Modern C, Third Edition

Bias is a systematic deviation between an observed measurement and the property it is intended to represent.

How it is used
Uses bias to qualify performance measurements that can alter the execution they observe.
How deep it goes
Explains timing overhead and compiler and runtime effects, then outlines repeated measurement and statistics; defers a full treatment of measurement.
Examples
Uses repeated timespec_get calls and timed loops to illustrate measurement overhead.
What you can do after
Evaluate timing results by considering instrumentation effects and summarizing repeated measurements.

Read how Modern C, Third Edition explains bias →

Build AI Drug Discovery Pipelines

Statistical bias is systematic error caused by a model's simplifying assumptions that prevent it from capturing the true structure of the data.

How it is used
Guides model selection and error diagnosis for solubility prediction and classification.
How deep it goes
Defines bias, relates it to training and test error, and diagnoses it with learning and validation curves.
Examples
Uses second- and twentieth-order polynomials, RANSAC and SVR learning curves, and regularized decision trees.
What you can do after
Diagnose underfitting with error curves and adjust model complexity, data collection, or regularization.

Read how Build AI Drug Discovery Pipelines explains bias →

Machine Learning System Design

A systematic error caused by a model's limited capacity to capture useful signals, leading it toward simplified assumptions about the data.

How it is used
Uses bias to reason about underfitting, validation reliability, and the business consequences of forecast errors.
How deep it goes
Defines model bias and explains the bias–variance tradeoff, then gives practical validation guidance without a mathematical derivation.
Examples
Uses promotional sales residuals and airline ticket forecasts to illustrate systematic prediction errors and their consequences.
What you can do after
Diagnose/debug with it by inspecting residual patterns, considering missing features, and checking validation design for biased estimates.

Read how Machine Learning System Design explains bias →

Building Reliable AI Systems

Bias is a systematic tendency in an AI system that produces unfairly skewed or discriminatory outcomes.

How it is used
Frames bias as a production fairness risk across data collection, fine-tuning, and deployment.
How deep it goes
Explains bias amplification and sketches a fairness-loss penalty, although the supplied implementation does not measure the intended group difference.
Examples
Uses recruiting discrimination, a customer-service chatbot, and résumé comparisons with varied names to illustrate demographic disparities.
What you can do after
Evaluate and understand how training data and fine-tuning can produce disparate outcomes and where fairness interventions might apply.

Read how Building Reliable AI Systems explains bias →

AI Governance

A systematic distortion in data or model behavior that can produce unfair or skewed outcomes.

How it is used
Treats bias as a trustworthiness risk that can persist despite security and privacy safeguards.
How deep it goes
Defines algorithmic bias and traces historical-data and proxy failures, while recommending testing and correction without specifying a detailed mitigation procedure.
Examples
Uses claims decisions, résumé rankings, hospital prioritization, and occupational image generation to illustrate unequal outcomes.
What you can do after
Evaluate and understand bias risks by comparing outcomes across demographic variants and considering oversight for high-impact decisions.

Read how AI Governance explains bias →

First-Party Data Activation

Model bias is systematic distortion in an artificial-intelligence model’s outputs that disadvantages or misrepresents certain groups or cases.

How it is used
Frames bias as an ethical risk in personalized and generative marketing.
How deep it goes
Explains biased training data and discriminatory outcomes, then recommends mitigation practices without specifying testing methods.
Examples
Uses personalized campaigns and generative marketing adoption to illustrate where bias matters.
What you can do after
Evaluate AI marketing risks and identify governance practices for mitigating bias.

Read how First-Party Data Activation explains bias →

Generative AI in Action

Bias is a systematic tendency in data or model behavior that produces unfair, distorted, or unbalanced outcomes.

How it is used
Addresses bias as an enterprise generative-AI risk in prompting, deployment, and responsible-AI governance.
How deep it goes
Identifies data, model assumptions, and algorithmic design as bias sources and calls for detection and correction without detailing procedures.
Examples
Contrasts recipe errors with machinery instructions and proposes diverse retrieval sources for enterprise decisions.
What you can do after
Evaluate and understand where generative-AI bias can arise and consider oversight, prompt vigilance, and diverse retrieval as precautions.

Read how Generative AI in Action explains bias →

Data Storytelling with Altair and AI

Bias is a systematic distortion or inclination that influences decisions, perceptions, judgments, or representations in favor of or against a particular subject.

How it is used
Uses bias to evaluate misleading charts and potential harms in generative-AI-assisted data stories.
How deep it goes
Defines bias, distinguishes data from algorithm bias, and works through normalization without technical AI internals.
Examples
Uses Italian homelessness rates and repeated DALL-E image outputs from simple person-description prompts.
What you can do after
Evaluate/understand when raw comparisons or AI outputs may create systematically misleading or exclusionary representations.

Read how Data Storytelling with Altair and AI explains bias →

How Large Language Models Work

Bias is a systematic skew in information, representation, or model behavior toward particular perspectives, groups, or time periods.

How it is used
Evaluates how LLM training, prompting, and deployment can reproduce or amplify harmful patterns.
How deep it goes
Explains examples, data sources, and testing limits, but does not develop a formal statistical account of bias.
Examples
Uses race and gender evaluations, African American Vernacular English, business-meeting images, and older public-domain texts.
What you can do after
Evaluate model outputs and intended uses for stereotypes, historical skew, untested contexts, and possible downstream harm.

Read how How Large Language Models Work explains bias →

Grokking AI Algorithms, Second Edition

Bias is a systematic distortion in data selection, recording, interpretation, or resulting conclusions.

How it is used
Frames bias as a data-quality and responsible-AI concern, while also defining its separate role as a perceptron parameter.
How deep it goes
Explains sampling, representation, and interpretation with examples, then gives an intuitive perceptron analogy without developing the parameter mathematically.
Examples
Uses cherry-picked observations, stereotypical doctor and nurse images, and a master-gain analogy for perceptron bias.
What you can do after
Evaluate datasets for sampling and representation problems, and understand how bias shifts a perceptron’s combined input.

Read how Grokking AI Algorithms, Second Edition explains bias →

Transfer Learning for Natural Language Processing

Bias is systematic skew in judgments or outcomes, including skew introduced when people prepare training labels.

How it is used
Highlights risks in training labels and model outputs when NLP systems are developed or deployed.
How deep it goes
Connects label quality and model predictions to practical harms, without detailing bias-mitigation methods.
Examples
Uses fake-news labels, Twi masked-word predictions, and gender-classification disparities as examples.
What you can do after
Evaluate/understand how label choices and model outputs can reproduce skew and why deployment requires scrutiny.

Read how Transfer Learning for Natural Language Processing explains bias →

Transformers in Action

How it is used
Identifies potential demographic and cultural distortions in LLM behavior to support fairer applications.
How deep it goes
Explains how pretraining representation can contribute to bias and demonstrates toxicity screening, without providing a comprehensive demographic-bias test.
Examples
Uses Llama demographic distribution figures and a RoBERTA-Toxicity classifier applied to a generated response.
What you can do after
Apply RoBERTA-Toxicity to screen model responses and evaluate pretraining-data representation for potential bias.

Read how Transformers in Action explains bias →

Build an Advanced RAG Application (From Scratch)

Bias is a systematic difference in a model's behavior or outcomes that disadvantages particular groups or viewpoints.

How it is used
Frames bias as a limitation requiring evaluation and mitigation when using large language models.
How deep it goes
Defines biased behavior and identifies its training-data origin, but stops short of measurement procedures or mitigation algorithms.
Examples
Uses sexist, racist, and homophobic outputs plus demographic testing to illustrate biased model behavior.
What you can do after
Evaluate LLM behavior across diverse demographics and recognize outputs that may reproduce discriminatory training-data patterns.

Read how Build an Advanced RAG Application (From Scratch) explains bias →

Causal Inference for Data Science

Bias is systematic distortion of an estimated effect caused by differences in assignment or other factors associated with the outcome.

How it is used
Flags threats to causal effect estimates from confounding or changing relationships in observational and time-series analyses.
How deep it goes
Explains confounder identification and adjustment, then states how unstable time-series relationships can bias counterfactual estimates.
Examples
Uses physician treatment choices, patient characteristics, hospitals, surgery comparisons, and changing movie-viewing relationships.
What you can do after
Diagnose potential bias by checking confounders and whether control-outcome relationships remain stable around a time-related event.

Read how Causal Inference for Data Science explains bias →

Data Without Labels

A systematic preference or prejudice reflecting unequal social attitudes or structures that can appear in data or model outputs.

How it is used
Frames bias as an ethical risk to address when developing and integrating generative AI systems.
How deep it goes
Names bias as a risk in generated outputs and recommends policies, but gives no fairness measures or bias-audit procedure.
What you can do after
Evaluate/understand how generated text can reflect societal biases in training data and why human oversight matters.

Read how Data Without Labels explains bias →

Build a Text-to-Image Generator (from Scratch)

Model bias is a systematic tendency in a model's outputs that reflects or amplifies unequal or stereotypical patterns associated with social groups or cultural contexts.

How it is used
Frames bias as a social and ethical concern in text-to-image generation, alongside misuse, resource consumption, and responsible development.
How deep it goes
Defines stereotypical outputs and their inequality effects, but stops short of specifying audit or mitigation procedures.
Examples
Uses “a doctor,” “a leader,” and Nazi-era soldier depictions to illustrate stereotyping and historically inaccurate overcorrection.
What you can do after
Evaluate generated depictions for stereotypes and understand why mitigation requires both technical and social judgment.

Read how Build a Text-to-Image Generator (from Scratch) explains bias →

Data-Driven Cybersecurity

Bias is a systematic distortion in an AI model or its outputs that produces consistently skewed assessments across cases or groups.

How it is used
Guides validation of AI-driven cybersecurity assessments before they influence fraud detection, threat response, or security decisions.
How deep it goes
Explains causes and consequences through scenarios, then recommends validation and oversight without detailing mathematical bias measures.
Examples
Uses cross-industry fraud misclassification and security incidents emphasized by training data as concrete examples.
What you can do after
Evaluate AI outputs for skewed assessments and apply human review and layered controls before acting on them.

Read how Data-Driven Cybersecurity explains bias →

The Complete Obsolete Guide to Generative AI

Bias is a systematic tendency for judgments or decisions to favor or disadvantage people for reasons unrelated to relevant qualifications.

How it is used
Positions bias as an ethical risk to manage when AI supports hiring and startup decisions.
How deep it goes
Frames bias through hiring criteria, auditing, and fairness caveats without providing measurement methods or technical mitigation procedures.
Examples
Uses AI candidate screening, video interviews, online tests, simulations, and startup AI oversight as concrete contexts.
What you can do after
Diagnose/debug AI-supported decisions by auditing outputs against relevant criteria and ethical concerns.

Read how The Complete Obsolete Guide to Generative AI explains bias →

Learn AI-Assisted Python Programming, Second Edition

Bias in AI results is systematic skew in generated outputs that favors or disfavors particular perspectives, groups, or outcomes.

How it is used
Warns programmers to supervise AI-generated names, visualizations, and code for patterns inherited from training data.
How deep it goes
Defines bias through concrete outputs and links it to demographic underrepresentation, but does not explain measurement or mitigation methods.
Examples
Uses generated names, graphs, and code styles to demonstrate how bias can appear in AI-assisted programming.
What you can do after
Evaluate generated outputs for reproduced training-data biases before adopting them.

Read how Learn AI-Assisted Python Programming, Second Edition explains bias →

LLM Customization and Fine-Tuning

A systematic skew in model behavior or outcomes.

How it is used
Addresses fairness risks when fine-tuning models on concentrated domain data.
How deep it goes
Defines data-related risks and practical controls, including targeted testing, without specifying a bias metric.
Examples
Uses culturally diverse names, varied language styles, and demographic or contextual group tests.
What you can do after
Evaluate fine-tuned outputs for uneven performance by reviewing data representation and testing relevant groups.

Read how LLM Customization and Fine-Tuning explains bias →

Deep Learning with R, Third Edition

Bias is a systematic tendency in data, judgments, or model outputs that produces uneven or distorted representations of people or phenomena.

How it is used
Frames machine-learning projects as ethical choices whose data and prediction targets can reproduce human prejudice.
How deep it goes
Defines how labeling bias becomes encoded in models and scaled, but stops short of presenting mitigation methods.
Examples
Uses facial trustworthiness ratings to illustrate prejudiced labels, questionable targets, and scaled algorithmic harm.
What you can do after
Evaluate machine-learning proposals for questionable targets, biased labels, and harmful effects on people.

Read how Deep Learning with R, Third Edition explains bias →

Causal AI

Statistical bias is a systematic difference between an estimator's expected value or observed pattern and the relevant population quantity or target relationship.

How it is used
Diagnoses disparities in training data and model decisions by connecting observed patterns to causal pathways.
How deep it goes
Distinguishes direct, indirect, fair, and unfair sources with a DAG, but stops short of prescribing a complete correction method.
Examples
Uses online advertising, loan approval, gender disparities, sexist officers, and gender proxies in applications.
What you can do after
Diagnose bias with causal pathways and distinguish targeting effects from discriminatory decision policies.

Read how Causal AI explains bias →

Timeless Algorithms, The Foundational Ideas

Bias is a systematic deviation that can consistently distort data, estimates, or predictions.

How it is used
Frames bias as one side of the regularization tradeoff controlled by C in SVM boundary fitting.
How deep it goes
Defines bias and relates it to slack, margin width, underfitting, and C, while treating these effects as tendencies rather than guarantees.
What you can do after
Evaluate candidate C values by weighing the risks of rigid, high-bias boundaries against overly responsive, high-variance boundaries.

Read how Timeless Algorithms, The Foundational Ideas explains bias →

Also covered in

What to know first

Related concepts

Questions about bias

What is bias?

Bias is a systematic tendency or deviation that distorts measurements, estimates, judgments, or outcomes relative to a relevant reference or target.

Which Manning books explain bias?

Grokking Machine Learning, Inside AI, Natural Language Processing in Action, Second Edition, LLMs in Production, Grokking Statistics, Demand Forecasting Best Practices, AI Model Evaluation, Starting Data Analytics with Generative AI and Python and 30 more.

What should I know before learning bias?

Manning books that teach bias usually introduce language model, machine learning, prompt and data source first.

What kinds of bias are there?

Kinds of bias include selection bias, training data bias, measurement bias and central tendency bias.

What is bias also called?

Bias is also called biases.

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.