Quiz: Generative AI for Process Control
 |  |  |  |  |
|---|
1. A student pastes an assignment into an AI assistant, submits the AI's answer, and moves on. What is the main problem with this approach in an engineering course?
A. The AI's answer is always wrong
- Incorrect. AI answers are often correct, especially on standard textbook problems. The problem is not primarily the answer quality.
B. The student builds no understanding and cannot verify or defend the result
- Correct. The purpose of an assignment is the understanding built while doing it. An engineer must be able to verify and take responsibility for any result submitted, and unverified AI output fails on exams, in projects, and in the workplace.
C. Using AI tools is prohibited in engineering practice
- Incorrect. AI tools are widely used in engineering practice. The issue is how they are used, not whether.
D. The AI response is too long to submit
- Incorrect. Length is not the issue. Understanding and verification are.
2. Which prompt is most likely to build understanding of first-order dynamics?
A. "Solve problem 3 from my homework and show all steps."
- Incorrect. This produces a solution, not understanding. It skips the practice and self-testing that build skill.
B. "Write Python code to simulate a first-order system."
- Incorrect. Useful for productivity, but it does not test or build your own understanding of the concept.
C. "Quiz me with 5 questions about time constants, one at a time. Grade my answers and list my misconceptions at the end."
- Correct. Retrieval practice with feedback is one of the most effective ways to learn, and this prompt makes the AI a tutor that tests understanding instead of doing the work.
D. "Summarize the first-order systems page in 3 bullet points."
- Incorrect. A summary is passive. Reading a summary feels like learning but does not test whether you can apply the concept.
3. An AI assistant derives an energy balance for a stirred tank and reports
$$m\,c_p \frac{dT}{dt} = U\,A\,(T-T_a) + \alpha Q$$
The tank is hotter than the surroundings (`T > T_a`) and the heater is off (`Q=0`). What does checking this limiting case reveal?
A. The equation is correct because the units balance
- Incorrect. The units do balance, but a units check is not the only verification. Check the sign: with `T>T_a` and `Q=0`, this equation predicts the temperature rises without any heat input.
B. The sign of the heat transfer term is wrong: a hot tank should cool toward ambient
- Correct. With `T>T_a` and `Q=0`, the equation gives `\frac{dT}{dt}>0`: the hot tank heats up forever with no energy input. The convection term should be `UA(T_a-T)`. Limiting-case checks catch sign errors that units checks miss.
C. Nothing; limiting cases only apply to steady state
- Incorrect. Limiting cases apply to any condition where the correct behavior is known, including transient behavior like cooling toward ambient.
D. The equation must be solved numerically before it can be checked
- Incorrect. The sign error is visible by inspection, without solving anything. Verification is often faster than simulation.
4. Which element is missing from this prompt? "Explain PID control."
A. Context, constraints, and a verification step
- Correct. A strong prompt states who you are and the system you care about (context), the level, format, and scope (constraints), and how the answer will be tested (verification), e.g., "then ask me 3 questions to check my understanding."
B. Nothing; shorter prompts are always better
- Incorrect. Short prompts produce generic answers. A prompt is a small specification: the response quality follows the specification quality.
C. A request for Python code
- Incorrect. Code is not always the goal. The missing elements are context, constraints, and verification.
D. The AI model version number
- Incorrect. The model version rarely changes how you should structure a learning prompt.
5. During the course project, your AI assistant confidently recommends wiring a heater directly to a microcontroller output pin. What is the appropriate response?
A. Follow the recommendation; the AI has read more electronics documentation than you
- Incorrect. Confidence is not correctness. Microcontroller pins typically supply tens of milliamps; a heater draws far more and would damage the board. Safety-relevant advice must always be verified.
B. Verify the current requirements against the datasheets, check it with the safety guidelines, and use a driver circuit (e.g., a transistor) if the load exceeds the pin rating
- Correct. The engineer verifies before acting, especially where hardware damage or safety is involved. Datasheets and safety guidelines outrank AI confidence, and "the AI said so" is never a justification.
C. Ask the same AI if it is sure
- Incorrect. Asking "are you sure?" invites the model to double down or flip without new information. Verification requires an independent source such as a datasheet.
D. Never use AI for hardware questions
- Incorrect. AI is useful for explaining circuits, generating checklists, and debugging. The correct posture is use with verification, not avoidance.