You walk into a chalk talk. The room is full of people who will decide your future. You’re told to demonstrate your thinking — live, unassisted, in real time. So you do what any modern professional does: you reach for the tool that powers your actual daily work.
And they stop you.
No ChatGPT. No AI. No prompting. Just you, a whiteboard, and the expectation that you’ll perform cognition like it’s 1997.
This happened. A researcher — someone whose daily scientific practice involves typing prompts, receiving structured responses, editing them, and producing results — was told they couldn’t use the one tool that defines how they actually work. They called it discrimination. The internet laughed.
But here’s the thing: they weren’t entirely wrong.
We don’t test architects by making them carve stone by hand. We test them on whether the building stands.
The comment sections lit up with accusations of plagiarism. “You’re just copying what the AI says and calling it your own thinking!” And sure, if you’re pasting raw ChatGPT output verbatim and slapping your name on it, that’s lazy at best and dishonest at worst. But that’s not what we’re talking about, and you know it.
You know it because you do this every day. You prompt, you refine, you iterate, you edit. You take a rough structure and shape it into something precise. The synthesis — that’s yours. The judgment about what’s good, what’s wrong, what’s missing — that’s yours. The AI is a tool, not a ghostwriter. It’s a calculator for language.
And yet, the moment we step into an evaluation room, we pretend the tool doesn’t exist.
Think about what a chalk talk actually tests. It tests your ability to generate ideas under pressure, unaided, in front of an audience. That’s a real skill. But is it the skill that matters most in 2024? When was the last time you produced your best work alone, unassisted, on a whiteboard, with strangers watching?
The institution isn’t measuring competence. It’s measuring obedience to a world that no longer exists.
Here’s the uncomfortable truth nobody in the comments wanted to engage with: if AI-assisted work consistently produces better results than unassisted thinking — and in many domains, it already does — then the problem isn’t the employee’s methods. The problem is the test.
We’ve been here before. Mathematicians resisted calculators. Writers resisted word processors. Programmers resisted Stack Overflow. Every time a tool amplifies human capability, the gatekeepers panic and demand proof that people can still do it the hard way. And every time, they eventually lose, because the hard way stops being the real way.
The programmer in the comments section got it right: “In my experience with building and delivering software, it’s not really about memorization. Search ability has been far more important. Wouldn’t it be fair to think of AI as just another search ability?”
Yes. It would. And if your evaluation can’t tell the difference between someone who searches well and someone who thinks well, your evaluation is broken.
Now, let’s be clear about something before the strawman army arrives. This isn’t an argument for letting people off the hook entirely. Pure prompt-and-paste is not a skill. If you can’t explain what the AI produced, can’t identify its errors, can’t improve on its output — you’re not augmented, you’re dependent. And dependency should absolutely be screened out.
But that’s not what chalk talks do. Chalk talks screen out the augmented along with the dependent. They throw out the baby, the bathwater, and the entire plumbing system.
If AI-assisted work produces superior results to unassisted thinking, the problem is the evaluation metric, not the employee’s methods.
The real question institutions should be asking isn’t “Can this person think without AI?” It’s “Can this person think with AI — critically, creatively, and with judgment?” Because that’s the job. That’s every job now. The professional who can’t leverage AI effectively is the one who should worry. The one who can is the one you should hire.
So no, calling it “discrimination” was probably the wrong word. But the instinct behind it was right. When your evaluation method penalizes the exact practice that makes someone effective at their job, you’re not protecting standards. You’re protecting nostalgia.
And nostalgia has never built anything worth keeping.
FAQ
Q: Isn't using AI during an evaluation just cheating by another name?
A: Not if the evaluation is testing real-world capability. If your job involves AI every day, testing without it measures a skill you'll never use. The real question is whether you can critically evaluate, edit, and improve AI output — not whether you can pretend it doesn't exist.
Q: What should institutions do instead of banning AI?
A: Redesign evaluations around synthesis and judgment. Let candidates use AI, then test whether they can identify its errors, defend its outputs, and improve on its suggestions. The skill that matters is critical editing, not unassisted generation.
Q: Doesn't this just let lazy people hide behind AI?
A: Only if your evaluation is bad. A well-designed assessment exposes dependency instantly — ask someone to explain, critique, or extend what the AI produced. The lazy ones can't. The augmented ones thrive. Banning AI is the lazy solution for lazy evaluators.