E077 Lesson 3 — Responsible and Trustworthy AI

WGU E077 Lesson 3 (3.1 Core Principles, 3.2 Data Ethics & Risk Areas, 3.3 Generative AI Ethical Concerns). Built from the lesson pages; A.I. Revealed Ch 6 readings not yet included.

Responsible AI

The practice of developing and using AI in ways that promote fairness, accountability, transparency, and ethical outcomes.

Fairness

The principle that outcomes and performance should be equitable across relevant people, groups, and contexts.

Bias

A systematic tendency to produce unbalanced outcomes due to data, assumptions, or design choices.

Accountability

The principle that people and organizations remain responsible for decisions made with AI support and for the effects of those decisions.

Oversight

The review and supervision of AI systems to help ensure responsible use.

Governance

Structures and frameworks that guide AI development and deployment so innovation is balanced with ethical considerations.

Transparency

Clear communication about why an AI tool is used, what it is intended to do, and what its limitations are.

Explainability

The ability to give understandable reasons for an AI output at a level suited to the audience and context.

Ethics

Values-based reasoning used to identify potential harm, weigh trade-offs, and prioritize human well-being when using AI.

Privacy risk

The possibility that personal information is collected, used, shared, or repurposed in ways that reduce individual control or cause harm.

Personal information

Data that can identify, describe, or be linked to an individual and may need special protection.

Behavioral data

Information about people's actions, activities, or interactions that AI tools may collect and use.

Surveillance

Monitoring enabled by data collection and analytics that can intrude on privacy and limit autonomy.

Autonomy

A person's ability to make independent choices without undue influence, including influence from monitoring or automated decision tools.

Accuracy

The extent to which information or AI outputs are reliable, correct, and suitable for supporting decisions.

Access

The ability of people or groups to benefit from and use AI-enabled services.

Security

Protection of data and systems from unauthorized entry, misuse, alteration, or harm.

System integrity

The condition in which data and tools remain correct, reliable, and protected from unauthorized changes.

Hallucination

An AI-generated output that is incorrect, unsupported, or misleading even though it may sound confident.

Misinformation

False or misleading content that can influence decisions, understanding, or trust when presented as if it were correct.

Transparency challenge

Difficulty understanding why an AI tool produced a particular output or what limits apply to it.

Human oversight

The involvement of people in checking AI outputs and deciding when extra scrutiny or intervention is needed.

Human review

The process of examining AI-generated content for errors or unsupported claims before it is used.

Risk mitigation

Actions taken to identify, reduce, or manage potential harms from AI-generated outputs.

Trustworthy AI

AI that earns and keeps confidence through appropriate oversight, responsible use, and attention to risk.

Higher-impact context

A situation where AI outputs may significantly influence decisions, communication, or outcomes affecting individuals or groups.

Skewed datasets

A source of bias in which the information used for training is unbalanced or unrepresentative.

Unconscious developer bias

A source of unbalanced outcomes that comes from builders' own unrecognized assumptions.

Planning

The first stage of governance processes, before evaluation, approval, and implementation.

Approval

The governance stage that comes after evaluation and before implementation.

Post-Launch Monitoring

The step in the policy-to-implementation flow that feeds back to Intake through a Continuous Improvement loop.

Intake

The first step of the responsible AI policy-to-implementation flow.

Responsible AI Office

In the governance hierarchy, the unit below Executive Leadership that oversees Research, Policy, and Engineering functions.

Board / Senior Oversight

The top level of the responsible AI governance hierarchy.

Accountability

After an AI recommendation harms a client, a manager says the model made the call, not the team. Which principle is being broken?

Transparency

Clinicians using a triage tool can't tell what it is meant to do or what its limits are. Which principle is weakest?

Explainability

A loan applicant is denied, and staff can't give any understandable reason for the tool's output. Which property is missing?

Privacy risk

A navigation app reuses location history to profile users' daily habits for ads without telling them. Which data-ethics concern area applies?

System integrity

An intruder quietly alters records in a training database. Which condition has been broken?

Access

A city benefits assistant works only in English and only on a smartphone app, so some residents can't use it. Which data-ethics risk area is this?

Hallucination

A customer-service chatbot confidently quotes a refund policy that doesn't exist. Which generative AI concern is this?

Misinformation

A nonprofit mass-produces social posts with generative AI, and a false claim spreads across several communities. Which concern is this?

Human review

Before a health organization publishes AI-drafted messages, qualified staff check each one for unsupported claims. Which practice is this?

Fairness

A model has high overall correctness, yet outcomes are worse for one group because its training data had gaps. Which principle is compromised?

Surveillance

An employer's analytics tool tracks every keystroke and location of staff, and they feel unable to act freely. What is this monitoring called?

Higher-impact context

Healthcare decision-making, financial lending, hiring, and criminal justice are lesson examples of this kind of situation.

Bias

Racial disparities in facial recognition and gender disparities in job recommendation algorithms are lesson examples of harms from this.

IEEE

An organization the lesson names for developing standards emphasizing transparency, accountability, and harm prevention.

Hallucination

A generative tool states wrong eligibility rules or event dates in a confident health message, an example of this concern.

Post-Launch Monitoring

Catching and correcting problems in AI-drafted public messages after they go live is an example of this step.