How to Answer AI Ethics Questions in a Medical School Interview

Home • Interview Prep • How to Answer AI Ethics Questions in a Medical School Interview

Here is one of the biggest mistakes applicants make when preparing for medical school interviews:

They try to discover the “correct” answer to every ethical question.

Then they memorize it.

This works beautifully until the interviewer asks:

“Why?”

Or:

“What if I change the situation?”

And suddenly the carefully memorized answer collapses like an IKEA bookshelf assembled without the instructions.

AI questions are particularly good at exposing this problem.

There often is no simple answer.

Should physicians use AI? Sometimes.
Should patients be told AI was used? Depends.
Should physicians trust AI? Depends.
Can AI improve healthcare? Yes.
Can AI harm patients? Also yes.

Welcome to medicine.

A Better Framework: P-E-P-H-A

When you encounter an AI ethics scenario, think through five areas:

P — Patient

What does the patient need, want, understand, and value?

E — Evidence

How good is the evidence supporting this AI system?

P — Privacy & Equity

How is information protected, and does the tool work fairly across populations?

H — Human Oversight

What meaningful human review exists?

A — Accountability

Who is ultimately responsible?

You do not have to announce:

“I will now apply my PEPHA framework.”

Please don’t.

Use it internally to organize your thinking.

Scenario 1: AI Disagrees With You

You diagnose pneumonia.

An AI clinical decision-support system flags the patient as high risk for pulmonary embolism.

What do you do?

A weak applicant says:

“I would trust my medical judgment.”

Another weak applicant says:

“I would follow the AI because it has access to more information.”

A stronger applicant says:

The disagreement itself deserves investigation.

Ask:

What information caused the AI to flag PE?
Did I miss something?
Is the system validated for this population?
Are there symptoms, risk factors, labs, or imaging findings I should reconsider?
Could both conditions be present?
Do I need additional testing?

The principle is simple:

Do not outsource your judgment to either ego or automation.

Scenario 2: The AI Is Better Than You

Now the interviewer changes the question:

“Studies show this AI system is significantly more accurate than physicians. Would you follow it now?”

Good.

Now the problem is harder.

Ask:

What studies?
How large?
Which patients?
What outcome?
External validation?
What error rate?
What type of errors?
Was the patient population representative?
How is the system monitored after deployment?

“More accurate” is useful information.

It is not the end of the analysis.

Scenario 3: AI Performs Worse for Black Patients

Suppose a hospital uses an algorithm that performs very well overall but substantially worse in Black patients.

Should the hospital continue using it?

Now you need to recognize a fundamental problem with averages.

Imagine:

Group A: 98% accuracy
Group B: 72% accuracy
Combined average: impressive enough to put in a brochure

That does not mean the system is equitable.

A strong answer should discuss:

Validation across subgroups
Potential harm
Transparency
Alternative tools
Human oversight
Correction or suspension if disparities cannot be addressed

This is precisely why AI and health equity belong in the same conversation.

Scenario 4: AI Could Help a Rural Community

A rural clinic has limited access to specialists.

An AI system could assist primary-care physicians in interpreting certain studies.

Do you support using it?

Do not immediately answer yes.

Do not immediately answer no.

Ask what happens without it.

Ethical analysis requires comparison with the real alternative.

If the alternative is immediate evaluation by an expert specialist, AI may add little.

If the alternative is waiting three months or traveling 200 miles, the analysis changes.

This is why medical ethics rarely works well when removed from context.

Scenario 5: A Patient Says ChatGPT Diagnosed Them

A patient enters the office and says:

“ChatGPT says I have multiple sclerosis.”

What do you say?

Not:

“Don’t use ChatGPT.”

That conversation is already over.

Instead:

“I can understand why that would concern you. Let’s go through the symptoms you entered and what it told you.”

Now evaluate:

What symptoms?
How long?
What pattern?
What red flags?
What does the actual evidence suggest?

Maybe the AI output was nonsense.

Maybe it raised a legitimate possibility.

Either way, the patient came to you.

Do not punish them for trying to understand their health.

Scenario 6: AI Makes a Medical Error

Suppose AI recommends a medication.

The physician accepts the recommendation.

The patient experiences severe harm.

Who is responsible?

This is where applicants often desperately try to assign exactly 37% responsibility to the physician, 42% to the hospital, and 21% to the developer.

You are not an insurance adjuster.

Discuss the layers:

Was the physician expected to independently review the recommendation?
Was the AI approved for this use?
Was the system functioning correctly?
Did the hospital properly validate it?
Was the risk known?
Was the physician trained?
Was there an obvious contraindication?

Responsibility can be distributed across a system.

But a physician cannot simply say:

“The computer told me to do it.”

Professional accountability does not disappear because software was involved.

Scenario 7: AI Wants Your Patient’s Data

An AI system could potentially improve dramatically if it had access to millions of patient records.

Should the hospital provide them?

Now balance:

Potential societal benefit

against

Privacy and autonomy

Questions include:

Is the information identifiable?
How is it secured?
Who controls it?
Who profits from it?
Can patients opt out?
What happens if the data are breached?
Could the data later be used for another purpose?

Responsible innovation requires more than asking:

“Could this technology help?”

You must also ask:

At what cost, to whom, and under what safeguards?

Scenario 8: AI Saves Physicians Two Hours Per Day

Here is an interesting one.

Suppose an AI documentation tool reduces each physician’s administrative work by two hours daily.

Is this good?

Probably.

But what happens to those two hours?

Does the physician spend more time with patients?

Great.

Does the hospital simply schedule six additional patients?

Different conversation.

Technology exists inside economic systems.

Efficiency does not automatically translate into better patient care.

That is a sophisticated point to recognize in an interview.

Scenario 9: AI Delivers Bad News

Should AI ever tell a patient they have cancer?

This question gets directly to the human side of medicine.

An AI system might technically be able to communicate diagnostic information.

But communication is not merely transmission.

A physician:

Observes emotion.
Answers questions.
Clarifies uncertainty.
Understands family context.
Responds to silence.
Recognizes when the patient did not understand a word after hearing “cancer.”

Medicine is not simply getting the correct information from database A into human B.

The relationship matters.

Scenario 10: AI Is the Only Option

Suppose an underserved region has no radiologist available overnight.

An AI system can interpret emergency scans with reasonable accuracy.

Would you use it?

Again:

Compared with what?

A slightly imperfect tool may be unacceptable when an excellent human alternative is immediately available.

The same tool may be lifesaving when the alternative is no interpretation at all.

Ethics depends on context.

What Interviewers Are Actually Evaluating

An AI ethics station may appear to be about technology.

It is often testing something much older:

Can you identify competing values?
Can you tolerate ambiguity?
Can you explain your reasoning?
Can you change your position when new information appears?
Can you acknowledge uncertainty?
Can you prioritize patient welfare?
Can you disagree respectfully?
Can you avoid pretending you know things you do not know?

That last one is particularly important.

If an interviewer asks you something about AI you genuinely do not understand, you are allowed to say:

“I don’t know enough about that specific technology to make a confident judgment, but I would want to understand…”

Then explain what information you would seek.

That can be a very strong answer.

Doctors say “I don’t know” every day.

The good ones follow it with:

“…but I know how I would find out.”

What NOT to Do

Don’t be blindly pro-AI.

“AI is the future and will revolutionize everything.”

Maybe.

That sentence also sounds like you are trying to raise venture capital.

Don’t be blindly anti-AI.

“Medicine requires humans, so AI should not be involved.”

AI is already involved.

Don’t reduce everything to empathy.

Human empathy matters enormously.

It is not the only issue.

Don’t pretend AI is objective.

Algorithms can inherit bias from data, design, implementation, and the systems surrounding them.

Don’t assume AI is always less accurate than physicians.

In some tasks, an AI tool may outperform humans.

That is what makes the ethical question interesting.

Don’t memorize.

Your interviewer can destroy a memorized answer by changing one fact.

And they should.

The Best General Position

You do not need a political platform on artificial intelligence.

A sensible medical position is:

Use evidence-supported technology when it improves patient care, but maintain appropriate human judgment, transparency, privacy protection, equity, monitoring, and accountability.

Interestingly, that is very close to the themes the AAMC itself emphasizes in its responsible-AI principles.

That is not because the AAMC has discovered a magical answer.

It is because these principles reflect the same responsibilities medicine has always had.

AI is new.

The obligation to protect patients is not.

Final Advice

Do not prepare for AI questions by trying to become an AI expert.

Prepare by becoming better at reasoning.

Take a scenario.
State the problem.
Identify the stakeholders.
Ask what evidence is available.
Consider benefits and harms.
Consider equity.
Consider privacy.
Preserve human oversight where appropriate.
Clarify accountability.
Then make a decision.

And if the interviewer changes one fact?