Prof. Dr. Larry AdamsAcademic, Author & Researcher
The IEEE Series: AI, Machine Learning & Data Science · Book 9

IEEE-Oriented Responsible AI, Safety, Security and Ethics

Intended audience: Cross-disciplinary learners, engineers, managers, and practitioners

Dr. Lance Adams and Prof. Dr. Larry Adams

Free to read online. For a copy of this book, or to quote, highlight or reproduce any part of it, please request permission through the contact page beforehand.

Editorial and publication note

This is a proposed educational series. IEEE-style technical writing and citations do not imply IEEE endorsement, publication, accreditation, certification, or affiliation. Any standard, designation, author credential, publication detail, or institutional relationship must be verified before publication. These Word files are expanded educational drafts and chapter-development manuscripts for editorial review; they are not represented as final peer-reviewed or publication-ready books.

The proposed series title is descriptive only. Do not use IEEE logos or imply IEEE affiliation without authorisation. Any references to standards should be checked in the official IEEE Standards Association catalogue and used only within their stated scope.

About this book

Examines privacy, bias, transparency, AI safety, adversarial threats, human oversight, governance, auditing, and standards-aware engineering.

Who should use this book

This volume is designed for cross-disciplinary learners, engineers, managers, and practitioners. Instructors should adapt pacing, depth, tools, and examples to learner needs and local curriculum requirements.

Learning outcomes

  • Explain the main concepts and vocabulary introduced throughout the volume.
  • Apply a structured method to a relevant practical problem.
  • Interpret evidence, identify limitations, and communicate results clearly.
  • Consider privacy, fairness, safety, security, and human oversight where relevant.

How to use this book

Read the chapter explanation, complete the activity, answer the review questions, and keep a learning journal recording assumptions, decisions, mistakes, and improvements. For technical work, save code, data dictionaries, test results, and software-version details.

Table of contents

  • Chapter 1: Foundations of Responsible AI
  • Chapter 2: Ethical Principles in Intelligent Systems
  • Chapter 3: Privacy, Consent, and Data Protection
  • Chapter 4: Bias, Fairness, and Representation
  • Chapter 5: Transparency and Explainability
  • Chapter 6: AI Safety and Risk Assessment
  • Chapter 7: Secure AI Development
  • Chapter 8: Adversarial Attacks and Data Poisoning
  • Chapter 9: Human Factors and Human Oversight
  • Chapter 10: Reliability, Robustness, and Testing
  • Chapter 11: Governance and Accountability
  • Chapter 12: AI in Education, Healthcare, Finance, and Public Services
  • Chapter 13: Standards-Based Engineering and Conformity
  • Chapter 14: Auditing and Documenting AI Systems
  • Chapter 15: Responsible AI Impact Assessment

Chapter 1: Foundations of Responsible AI

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in foundations of responsible ai.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Responsible AI translates values and obligations into design, deployment, and monitoring practices. Principles such as fairness, accountability, privacy, safety, and transparency need operational definitions for each context.

Worked example and application

Worked example: imagine a small team considering this concept for a clearly scoped, low-risk task. First define the intended outcome and inputs; then choose a simple baseline, document assumptions, test representative examples, inspect errors, and explain when the method should not be used. Compare the result with a human-reviewed alternative before proposing deployment.

Practical activity

Identify stakeholders and draft context-specific responsible-use requirements.

Chapter review

  • What is the main idea in 'Foundations of Responsible AI'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 2: Ethical Principles in Intelligent Systems

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in ethical principles in intelligent systems.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Ethical analysis considers benefits, harms, rights, duties, distributional effects, and stakeholder perspectives. Principles may conflict, so decisions should be documented with rationale and review mechanisms.

Worked example and application

Worked example: list the system's users, affected people, assets, likely failure modes, and potential consequences. Select preventive and detective controls, assign an accountable owner, and record the residual risk that remains after controls.

Practical activity

Analyse a fictional allocation system from the perspectives of several affected groups.

Chapter review

  • What is the main idea in 'Ethical Principles in Intelligent Systems'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 3: Privacy, Consent, and Data Protection

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in privacy, consent, and data protection.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Privacy work examines necessity, purpose limitation, consent where applicable, access, retention, security, and data subject rights under relevant law. Avoid collecting sensitive information merely because it is available.

Worked example and application

Worked example: use a small, fictional dataset with a documented meaning for each column. Check completeness and units, calculate a simple summary, visualise the result, and write one conclusion that is directly supported by the observations. State what the dataset cannot establish.

Practical activity

Create a data-flow map and identify unnecessary fields and privacy safeguards.

Chapter review

  • What is the main idea in 'Privacy, Consent, and Data Protection'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 4: Bias, Fairness, and Representation

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in bias, fairness, and representation.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Bias may arise from sampling, labels, measurement, historical practice, model design, or deployment. Fairness metrics can conflict and require a context-specific choice with stakeholder involvement.

Worked example and application

Worked example: define a real-world task, list the inputs and constraints, sketch the steps or system components, and test the proposed solution with at least one ordinary case and one edge case. Record the outcome and what you would improve.

Practical activity

Compare subgroup error rates in a fictional dataset and discuss what they do and do not establish.

Chapter review

  • What is the main idea in 'Bias, Fairness, and Representation'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 5: Transparency and Explainability

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in transparency and explainability.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Transparency includes communicating purpose, limitations, data sources, and governance; explainability concerns methods for understanding particular outputs or behaviour. Neither alone guarantees fairness or accuracy.

Worked example and application

Worked example: define a real-world task, list the inputs and constraints, sketch the steps or system components, and test the proposed solution with at least one ordinary case and one edge case. Record the outcome and what you would improve.

Practical activity

Draft a user-facing explanation and a technical model documentation summary.

Chapter review

  • What is the main idea in 'Transparency and Explainability'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 6: AI Safety and Risk Assessment

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in ai safety and risk assessment.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Risk assessment identifies hazards, affected parties, likelihood, severity, exposure, mitigations, and residual risk. Controls should be tested and assigned to accountable owners.

Worked example and application

Worked example: imagine a small team considering this concept for a clearly scoped, low-risk task. First define the intended outcome and inputs; then choose a simple baseline, document assumptions, test representative examples, inspect errors, and explain when the method should not be used. Compare the result with a human-reviewed alternative before proposing deployment.

Practical activity

Prepare a risk register for an AI-assisted service, including residual risk and review dates.

Chapter review

  • What is the main idea in 'AI Safety and Risk Assessment'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 7: Secure AI Development

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in secure ai development.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Security requires threat modelling across data, models, interfaces, dependencies, and deployment. Apply least privilege, secure development, access controls, monitoring, and incident response.

Worked example and application

Worked example: imagine a small team considering this concept for a clearly scoped, low-risk task. First define the intended outcome and inputs; then choose a simple baseline, document assumptions, test representative examples, inspect errors, and explain when the method should not be used. Compare the result with a human-reviewed alternative before proposing deployment.

Practical activity

Create a threat model and map each threat to prevention, detection, and response controls.

Chapter review

  • What is the main idea in 'Secure AI Development'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 8: Adversarial Attacks and Data Poisoning

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in adversarial attacks and data poisoning.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Adversarial examples, poisoning, prompt injection, model extraction, and abuse of interfaces can undermine system integrity. Threats depend on architecture and access; no single control addresses all cases.

Worked example and application

Worked example: use a small, fictional dataset with a documented meaning for each column. Check completeness and units, calculate a simple summary, visualise the result, and write one conclusion that is directly supported by the observations. State what the dataset cannot establish.

Practical activity

Review a fictional attack scenario and propose layered mitigations and tests.

Chapter review

  • What is the main idea in 'Adversarial Attacks and Data Poisoning'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 9: Human Factors and Human Oversight

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in human factors and human oversight.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Human oversight is effective only when people have suitable information, time, authority, and competence to intervene. Automation bias can cause users to accept incorrect outputs uncritically.

Worked example and application

Worked example: define a real-world task, list the inputs and constraints, sketch the steps or system components, and test the proposed solution with at least one ordinary case and one edge case. Record the outcome and what you would improve.

Practical activity

Design an escalation pathway and define when human review is mandatory.

Chapter review

  • What is the main idea in 'Human Factors and Human Oversight'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 10: Reliability, Robustness, and Testing

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in reliability, robustness, and testing.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Reliability testing examines expected conditions, edge cases, distribution changes, and failures. Robustness is task-specific and should be measured rather than asserted.

Worked example and application

Worked example: define a real-world task, list the inputs and constraints, sketch the steps or system components, and test the proposed solution with at least one ordinary case and one edge case. Record the outcome and what you would improve.

Practical activity

Build a test matrix covering normal, boundary, subgroup, and out-of-distribution cases.

Chapter review

  • What is the main idea in 'Reliability, Robustness, and Testing'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 11: Governance and Accountability

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in governance and accountability.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Governance establishes ownership, decision rights, documentation, escalation, audit, and lifecycle review. Accountability must be assigned to real roles rather than an abstract 'AI team'.

Worked example and application

Worked example: define a real-world task, list the inputs and constraints, sketch the steps or system components, and test the proposed solution with at least one ordinary case and one edge case. Record the outcome and what you would improve.

Practical activity

Create a responsibility matrix for model owner, data steward, security lead, and decision owner.

Chapter review

  • What is the main idea in 'Governance and Accountability'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 12: AI in Education, Healthcare, Finance, and Public Services

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in ai in education, healthcare, finance, and public services.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

High-impact contexts differ in regulation, consequences, data sensitivity, and professional responsibilities. Use cases should be evaluated within their legal and operational setting rather than assuming one universal checklist is enough.

Worked example and application

Worked example: imagine a small team considering this concept for a clearly scoped, low-risk task. First define the intended outcome and inputs; then choose a simple baseline, document assumptions, test representative examples, inspect errors, and explain when the method should not be used. Compare the result with a human-reviewed alternative before proposing deployment.

Practical activity

Compare the risk profile of two domains and specify additional domain review needed.

Chapter review

  • What is the main idea in 'AI in Education, Healthcare, Finance, and Public Services'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 13: Standards-Based Engineering and Conformity

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in standards-based engineering and conformity.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Standards can provide agreed requirements or processes within defined scopes. Confirm exact designation, edition, status, and applicability; citing a standard does not prove conformity or approval.

Worked example and application

Worked example: define a real-world task, list the inputs and constraints, sketch the steps or system components, and test the proposed solution with at least one ordinary case and one edge case. Record the outcome and what you would improve.

Practical activity

Create a standards verification worksheet with official source, edition, scope, and evidence required.

Chapter review

  • What is the main idea in 'Standards-Based Engineering and Conformity'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 14: Auditing and Documenting AI Systems

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in auditing and documenting ai systems.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Audits examine design, data, tests, decisions, incidents, and governance evidence. Documentation should be accurate, versioned, accessible to relevant reviewers, and updated when systems change.

Worked example and application

Worked example: imagine a small team considering this concept for a clearly scoped, low-risk task. First define the intended outcome and inputs; then choose a simple baseline, document assumptions, test representative examples, inspect errors, and explain when the method should not be used. Compare the result with a human-reviewed alternative before proposing deployment.

Practical activity

Prepare a compact audit evidence checklist for a deployed model.

Chapter review

  • What is the main idea in 'Auditing and Documenting AI Systems'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Chapter 15: Responsible AI Impact Assessment

Learning outcomes

By the end of this chapter, the learner should be able to:

  • Explain the central concepts introduced in responsible ai impact assessment.
  • Apply the ideas to a clearly defined example or practical task.
  • Identify at least one limitation, assumption, or responsible-use consideration.

Core explanation

Impact assessment connects the proposed use to affected stakeholders, potential harms, mitigations, residual risks, monitoring, and accountability. Assessments should be revisited when context or capabilities change.

Worked example and application

Worked example: imagine a small team considering this concept for a clearly scoped, low-risk task. First define the intended outcome and inputs; then choose a simple baseline, document assumptions, test representative examples, inspect errors, and explain when the method should not be used. Compare the result with a human-reviewed alternative before proposing deployment.

Practical activity

Produce an impact assessment with stakeholder map, controls, residual risks, and review triggers.

Chapter review

  • What is the main idea in 'Responsible AI Impact Assessment'?
  • Which assumptions or data requirements matter most in this topic?
  • What could cause the approach to fail or produce a misleading result?
  • How would you explain the result or limitation to a non-specialist?

Key takeaways

  • Start with a clear question, objective, or system requirement.
  • Document inputs, assumptions, methods, and limitations.
  • Test the approach with relevant examples and review the consequences of errors.

Capstone project

Prepare an AI system risk and impact assessment identifying affected stakeholders, foreseeable harms, mitigation controls, residual risks, monitoring, and accountability arrangements. Include an evidence register and reassessment triggers.

Suggested project deliverables

  • A short problem statement and intended audience.
  • A description of data, inputs, assumptions, tools, and method.
  • Evidence of testing, analysis, or evaluation appropriate to the project.
  • A clear explanation of results, limitations, and foreseeable risks.
  • A final report or presentation and a reflection on possible improvements.

Assessment guidance

Assess stakeholder analysis, contextual risk reasoning, control quality, evidence traceability, standards accuracy, and explicit residual-risk ownership.

Glossary

Residual risk: Risk remaining after controls have been applied.

Model card: A structured document describing a model's intended use, evaluation, and limitations.

Threat model: A systematic description of assets, threats, attack paths, and controls.

Human oversight: Arrangements that enable appropriately qualified people to understand, supervise, and intervene in a system.

Conformity assessment: A process for determining whether specified requirements have been met.

Responsible practice checklist

  • Use data and materials only when their source and permitted use are understood.
  • Avoid exposing personal, confidential, or credential information.
  • Document assumptions, missing information, and important limitations.
  • Test ordinary cases, boundary cases, and foreseeable failure conditions.
  • Use human review where errors could materially affect people, safety, rights, or resources.
  • Do not claim that a system is fair, safe, secure, compliant, or reliable without appropriate evidence.

References and further reading

The source material supplied for this project does not provide a complete, volume-specific bibliography. Before publication, add verified primary sources, textbooks, official documentation, and standards applicable to this volume. Confirm exact titles, editions, publication dates, URLs, licences, and access dates. Do not cite a standard as evidence of conformity unless the required assessment has actually been performed.

[1] C. M. Bishop, Pattern Recognition and Machine Learning. New York, NY, USA: Springer, 2006.

[2] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. Cambridge, MA, USA: MIT Press, 2016.

[3] National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1, Jan. 2023.

[4] IEEE, “IEEE Author Center.” [Online]. Available: https://journals.ieeeauthorcenter.ieee.org/

[5] IEEE Standards Association, “IEEE Standards.” [Online]. Available: https://standards.ieee.org/