I still remember the faculty meeting back in late 2024 when someone confidently declared that while AI could write code, it would never have “true creativity.” Well, that’s not entirely accurate—by now, you’ve probably seen the benchmarks. GPT-4 didn’t just pass the Torrance Tests of Creative Thinking; it crushed them. We’re talking top 1% scores for originality. And if you’re an educator still assigning “come up with 10 unique ideas for a marketing campaign” as a graded homework assignment, you are wasting your time. Your students know it.

I’ve been wrestling with this in my own design curriculum for the last six months. The reality isn’t that human creativity is dead. It’s that our definition of “creative work” in schools was lazy. We conflated “generating lots of random ideas” with actual creativity. AI proved that generating ideas is the easy part. It’s the cheap part.

The “Idea Hose” Experiment

I got tired of the theoretical debates, so last Tuesday I ran a little test during my morning lab. The prompt was simple: “List 50 unconventional uses for a defunct shopping mall.”

My students had 20 minutes. I ran the same prompt through a Python script hitting the OpenAI API. To make it fair, I didn’t just use the default settings. I tweaked the parameters to encourage divergence:

• Model: gpt-4-turbo-2025-11-06 (I stick to this snapshot for consistency)
• Temperature: 1.1 (high enough for chaos, low enough to avoid gibberish)
• Frequency Penalty: 0.6 (to stop it from repeating themes)

The results were humbling. The class average was 18 ideas. Most were variations of “paintball arena” or “homeless shelter.” Solid, socially conscious, but predictable.

The script returned 50 ideas in roughly 14 seconds. Cost me about three cents. But here’s the kicker—the quality wasn’t just “okay.” It was weirdly specific. It suggested converting the atrium into a vertical hydroponic mushroom farm, using the escalator shafts for gravity-fed logistics, and flooding the lower level for indoor kayaking.

The Death of the Blank Page

This breaks the fundamental loop of how we teach creative writing and design. For decades, the hardest part was the “blank page problem.” Getting started. Generating the raw material.

That problem is solved. It’s gone.

If you ask a student to brainstorm, they can now generate 200 viable starting points before they finish their coffee. This shifts the cognitive load entirely. The skill isn’t generation anymore; it’s curation.

Technical Reality Check: It Still Has No Taste

Here is where the AI falls on its face, and where we need to focus our teaching. While GPT-4 scored in the top percentile for originality (uniqueness of ideas), its ability to judge the quality of those ideas is still abysmal.

In my shopping mall experiment, the AI also suggested “fill the mall with bees to make honey.” Unique? Yes. Practical? No. Safe? Absolutely not.

How I’m Changing My Grading for 2026

I’m done fighting the tools. If a calculator can do long division, we teach calculus. If an LLM can brainstorm, we teach curation and synthesis.

Here is the rubric I’m using this semester. It’s messy, but it works:

1. Prompt Engineering (20%): Show me the inputs. Did you ask for “ideas,” or did you ask for “sustainable, low-cost adaptive reuse scenarios specific to the Rust Belt”?
2. The Curation Layer (40%): You generated 100 ideas. You showed me 3. Why those 3? I want a 500-word defense of your selection criteria.
3. Implementation Details (40%): Take the idea and make it real. The AI can dream up a “vertical mushroom farm,” but it can’t figure out the load-bearing capacity of a 1990s food court floor. You can.

I ran a pilot of this rubric with a senior project group last month. One student used the API to generate 500 potential names for a startup, filtered them programmatically by available .com domains (using a separate script), and then manually selected the final brand based on phonetics and cultural fit.

That is the workflow. Machine for volume, human for taste.

We need to stop treating these high test scores as a sign of the apocalypse or a miracle. It’s just a metric. The “top 1%” headline is catchy, but it misses the point. The AI isn’t a genius. It’s just a very fast, very eager improviser that never gets embarrassed by a bad idea.

Our job isn’t to compete with it on speed or volume. You will lose that race every time. Our job is to be the editor. The director. The one who looks at the pile of 50 weird ideas and says, “Okay, number 42 is actually brilliant, but we need to tweak the ending.”

If you’re still grading on “who can come up with the most ideas,” you’re grading a metric that stopped mattering in 2023. Let the bots have the brainstorming sessions. We have real work to do.