Skip to content
NTPU AI Learn

GoAI 101 (6-Hour Edition) Practice Center

Six judgment activities; each output feeds the same integrated Canvas (Government AI Governance & Acceptance Canvas).

Info

Every scenario is simulated teaching material; agency names, people, and figures are examples only and do not represent any real institution.

Warning

Do not enter real national ID numbers, real citizen personal data, sensitive government information, account passwords, API keys, or other unauthorized data into these exercises. Use de-identified, simulated, or public data only.

Practice challenges

Activity 1: AI use judgment card

Given a simulated government work scenario, judge how far AI should go (Assist / Recommend / Execute) and state the human responsibility. The output feeds Canvas fields A use case name, B business goal, C AI role, and K human owner.

Steps

  1. Read the case background and write the business goal and the users
  2. Choose the AI role (Assist / Recommend / Execute) and give the reason and the consequence of an error
  3. List the actions AI is permitted to take and those it is prohibited from taking
  4. Name the accountable human role and write the main risks

Deliverables

  • An AI use judgment card (case name, business goal, AI role and reason, permitted / prohibited actions, human owner)
  • One sentence on why AI should not do more (or less)

Activity 2: Fix a flowchart with human gates

Given a half-finished flowchart, add human approval, escalation, and hand-back-to-human paths, and make sure responsibility is clear at every step. The output feeds Canvas field D workflow steps and human/AI division.

Steps

  1. Check the input, output, and owner (AI or human) of each step
  2. Find high-impact steps that lack human approval and add the gate
  3. Add exception handling: escalation, hand-back to a human, and the failure path
  4. Write the completion conditions of the flow

Deliverables

  • A corrected flowchart with human gates (AI decision points and human decision points marked)
  • A gap list: which gates or exception paths the original was missing

Activity 3: Data and tool permission table

Use one table to cover both what AI may know (data) and what it may do (tools). The output feeds Canvas fields E context and trusted data and F tools and permissions.

Steps

  1. List candidate data and tools, stating the purpose of each
  2. Classify data as Include / Retrieve / Remember / Exclude, and mark source, version, and sensitivity
  3. For tools, mark the access scope, separate query / write / send actions, and apply least privilege
  4. Mark high-impact actions that need approval and the approver, and write the limits and the fallback when something fails

Deliverables

  • One integrated table (item, purpose, source or access scope, permitted action, approver, limits)
  • A short note on which data or actions were excluded and why

Activity 4: Governance and control checklist

Go back over the earlier flowchart and permission table, find the governance gaps, and add a control, an owner, and evidence for each risk. "A human will look at it" is not complete governance. The output feeds Canvas fields G governance and human-in-the-loop and L risk and mitigation.

Steps

  1. Check the flow and the permission table item by item for places with no owner or no control
  2. For risks such as prompt injection, data leakage, tool misuse, and over-authorization, write one concrete control each
  3. Assign an owner to each control and distinguish human approval gates from system control gates
  4. Write the evidence each control can leave behind, and the incident response: stop, report, human takeover

Deliverables

  • A governance and control checklist (risk, control, owner, evidence)
  • At least one governance gap and how to fix it

Activity 5: Three types of test cases

Using a template, complete normal, edge/ambiguous, and failure/restricted test cases for the same case. A test that was not actually run must be labelled "test design", never presented as obtained evidence. The output feeds Canvas field H reliability tests.

Steps

  1. Write the input scenario for each case type
  2. Write the expected behavior and the prohibited behavior
  3. Write the required evidence and the pass/fail criterion
  4. For the simulated outputs provided, fill in the actual behavior and judge pass or fail; label tests not actually run as "test design"
  5. Write which regression tests to rerun after a model or knowledge update

Deliverables

  • Three test cases (normal, edge/ambiguous, failure/restricted), each with input, expected behavior, prohibited behavior, and required evidence
  • A marking of which results have evidence and which are only "test design"

Activity 6: Rewrite a vague requirement and make an acceptance decision

First rewrite one vague requirement into a verifiable one, then review the provided evidence package against the 14 acceptance dimensions and give a decision. The output feeds Canvas fields I procurement requirement mapping, J acceptance decision, M evidence and traceability, and N open issues.

Steps

  1. Rewrite the vague requirement so it states the business purpose, the concrete requirement, the verification method, the acceptance criterion, and the delivery evidence
  2. Review the evidence package against the 14 acceptance dimensions (Function, Workflow, Accuracy / Quality, Knowledge / RAG, Permission, Security / Privacy, Tool Use, Human Approval, Logging / Traceability, Error Handling, Reliability, Version Control, Regression Test, Operations & Maintenance)
  3. Mark the dimensions that are unmet or lack evidence
  4. Decide Accept / Accept with conditions / Reject, and write the reasons and the open issues

Deliverables

  • One rewritten verifiable requirement (requirement, verification method, acceptance criterion, delivery evidence)
  • An acceptance decision: the decision, the reasons and the evidence relied on, and the open issues

Completion marks are stored only in this browser.

Sync my progressJoin a session to keep progress across devices and share it with your instructor

Info

After you join, your completed lesson numbers are sent to your instructor together with your learner code, so they can see your progress. Use the codes your instructor gave you, and do not enter your name, company secrets or other personal data. Progress is kept with the rest of that session's data.

What next?