After this chapter, you should be able to
- Distinguish AI output from an engineering decision.
- Define intended use, user, context and consequence.
- Assign accountable human roles and escalation paths.
- Recognise competence, ethics and public-safety duties.
- Set a proportionate no-go, research, pilot or bounded-assistance state.
Engineering context and methodSource §Lesson 01 · Engineering context and method · GOV-01 · GOV-02 · GOV-03 · GOV-11 · RES-01
AI is a family of statistical and computational methods, not an engineering authority. Begin with the decision, the evidence a competent person needs, the harm from false positive and false negative actions, and a transparent non-AI baseline. ASCE policy states that engineers retain responsibility for work products that use AI; the Code of Ethics keeps public safety, competence, truthfulness and disclosure duties in force. The course therefore treats AI as bounded evidence within professional control.
Verified worked exampleSource §Lesson 01 · Verified worked example · GOV-01 · GOV-02 · GOV-03 · GOV-11 · RES-01
Condition-report triage
A classifier ranks 60 bridge reports; the proposed use is reviewer prioritisation, not condition determination.
- Scope
Output = review priority only
No automatic asset status - Consequence
False negative → delayed review
High-consequence error - Control
All urgent cues + low-confidence cases → engineer
Human escalation retained
Result. A bounded triage pilot can be assessed; autonomous condition decisions remain out of scope.
Practical lab · 4 h lesson effortSource §Lesson 01 · Practical lab · 4 h lesson effort · GOV-01 · GOV-02 · GOV-03 · GOV-11 · RES-01
- Select one civil-engineering decision and write its decision owner, users, affected parties and existing workflow.
- Build a consequence matrix for false positive, false negative, abstention and system failure.
- Define a measurable baseline and the minimum evidence needed before any pilot.
- Write human review, override, escalation and audit requirements.
Failure modes to investigateSource §Lesson 01 · Failure modes to investigate · GOV-01 · GOV-02 · GOV-03 · GOV-11 · RES-01
- Starting with a fashionable model instead of a decision.
- Calling a probability an engineering conclusion.
- No named accountable owner or competent reviewer.
- Using average accuracy to hide a safety-critical error.
- Treating voluntary governance guidance as legal authorisation.
Knowledge checksSource §Lesson 01 · Knowledge checks · GOV-01 · GOV-02 · GOV-03 · GOV-11 · RES-01
| Question | Answer rationale |
|---|---|
| Who retains responsibility when AI is used? | The engineer and authorised organisation retain responsibility; the model cannot assume professional duty. |
| What comes before model selection? | The intended decision, consequence, baseline, evidence and control boundary. |
| Is human-in-the-loop a sufficient control by itself? | No; the person needs competence, time, information and real authority to disagree. |
| Can high accuracy authorise deployment? | No; representativeness, consequential errors, uncertainty and operating controls also govern. |
| What is a bounded use? | A declared task, population, conditions, outputs, users and abstention/escalation rule. |
Key points
- Start from the accountable engineering decision and its consequence.
- Compare against a transparent non-AI baseline.
- Validate on a split that represents intended use and retain human authority.
Source references recorded by the supplied chapter
- ABET, Criteria for Accrediting Engineering Programs 2026–2027, Civil Engineering program criteria.
- ASCE Policy Statement 573, Artificial Intelligence and Engineering Responsibility.
- ASCE Code of Ethics.
- International Engineering Alliance, Graduate Attributes and Professional Competencies v4 / 2021.1.