Creative Diversity Gym: A Mindware Tool for Creativity.
A practical workflow for human-first creativity in the age of fluent AI based on the H-AGI Protocol: IQMindware Strategy Lab
H-AGI Protocol: Human-Artificial General Intelligence Protocol
This tool is part of the IQMindware.com suite of evidence-based general itnelligence augmentation tools. This tool is free. A new tool will be available each week for problem solving, decision making, argumentation, strategic action, negotiation, and so on for paid subscribers to this Substack.
AI can now help you write articles, generate business ideas, draft lesson plans, produce social media posts, outline books, design products and brainstorm almost anything.
That is obviously powerful, but it’s also risky.
Same Altman identified the global demand for intelligence, and envisaged providing it at scale, flooding the market with IQ as embodied in ChatGPT.
AI is useful enough that millions of people may start using the same tools, the same models and the same default patterns of reasoning to generate their ideas.
The result may not be bad work.
It may be polished, competent and fluent work.
But also increasingly similar work.
That is the problem behind this week’s H-AGI tool from IQMindware’s Strategy Lab.
It is designed to help you use AI without letting AI quietly compress your creative search space.
The problem: AI can raise individual creativity while reducing collective diversity
A recent paper by Kibum Moon, Adam Green and Kostadin Kushlev looked at a subtle but important risk of large language models.
The authors were not simply asking whether AI can produce creative writing. In many cases, it can. Other studies have found that LLMs can perform impressively on creative tasks, and many of us have already experienced this directly. Ask a good model for article ideas, metaphors, explanations, titles or campaign angles, and it will often produce something useful.
But Moon and colleagues asked a deeper question:
What happens to creative diversity when many outputs are generated by the same kind of model?
That matters because creativity is not only about whether one idea is good. It is also about whether a group, team, field or culture continues to generate many different kinds of ideas.
A single AI-generated essay may look original.
A hundred AI-generated essays may start to reveal a pattern.
The researchers compared human-written college admissions essays with GPT-4-generated essays. Across three preregistered studies involving 2,200 essays, they found that human-written essays contributed more new semantic diversity as more essays were added.
They used a measure called the diversity growth rate. In plain English, this asks:
When you add one more piece of work, does it add genuinely new ideas, or does it mostly repeat the same territory?
The striking result was that GPT-4 essays showed a lower diversity growth rate than human-written essays. Even when the researchers tried to make the AI outputs more diverse through prompt changes, parameter changes and chain-of-thought prompting, the human-written essays still contributed more collective diversity.
This is the key point:
AI may help one person generate a better-looking output, while still making many people’s outputs more alike.
That is the AI creativity paradox.
Why this matters beyond essays
This is not just about college admissions essays.
The same problem can show up anywhere people use AI to generate ideas:
article angles
business strategies
product concepts
lesson plans
research questions
marketing copy
political messaging
personal statements
coaching exercises
video titles
design concepts
If many people use the same tool in the same way, the outputs may start to converge.
The problem is not that every AI answer is identical. The problem is that AI often pulls ideas towards familiar regions of the possibility space. It is trained on existing patterns. It predicts likely continuations. It is often tuned to be helpful, safe, coherent and acceptable.
Those are useful properties. But they can also reduce surprise.
They can make unusual ideas less likely to appear.
They can produce what I think of as the polished average: content that sounds good, but does not add much new, original structure to the world.
The hidden danger: fluency feels like creativity
One reason this matters is that fluent AI output can feel more creative than it really is.
A polished answer gives the brain a sense of completion. It reduces effort. It feels organised. It looks like thinking has happened.
But fluency is not the same as originality.
A good creative process needs more than smooth output. It needs:
divergence
reframing
personal experience
odd associations
constraint shifts
useful weirdness
selection
testing
revision
The danger is that AI can give us the feeling of having explored a wide space when we have actually explored a narrow one very efficiently.
That is why we need better AI habits.
Not “never use AI”.
Not “let AI do the thinking”.
Something more disciplined:
Generate first. Probe with AI. Check for convergence. Re-enter constraints. Test one idea.
That is the habit Creative Diversity Gym is designed to train.
The H-AGI principle: AI extends cognition, but the human governs cognition
Creative Diversity Gym is part of IQ Mindware’s H-AGI approach: Human-led Adaptive Generative Intelligence.
The basic principle is simple:
AI should extend human cognition, not replace human judgement.
Used well, AI can act as an external cognitive workspace. It can help us simulate alternatives, challenge assumptions, generate analogies, identify blind spots and reframe problems.
But the human still needs to govern the process.
That means you should not begin creative work by asking:
What are the best ideas?
You should begin by asking:
What do I think before AI enters the loop?
This matters because your first ideas contain information that the model does not automatically have: your experience, your context, your taste, your frustrations, your memories, your audience, your constraints and your half-formed intuitions.
Those things may not be polished.
But they are often where the non-generic material lives.
So the first move in Creative Diversity Gym is a Human-First Lock.
Before you can use AI, you have to write your own ideas.
The Creative Diversity Gym loop
The tool takes you through a short sequence:
human-first ideation
AI probe
diversity audit
novelty/usefulness selection
causal probe
transfer challenge
The whole flow can be completed in around 20–30 minutes.
The goal is not to produce hundreds of ideas.
The goal is to train one repeatable creative habit.
Step 1: Frame the task
You start by writing what you are trying to create.
This could be:
a Substack article
a lesson activity
a product feature
a business offer
a research hypothesis
a YouTube title
a coaching exercise
a positioning statement
You also rate how important originality is, how much time you have and what state you are in.
That last part matters.
If you are clear, you can use the full flow.
If you are overloaded, the tool simplifies the task.
If you are rushing, it forces a slow-check before you commit.
If you are stuck, it asks you to write bad ideas on purpose, because sometimes the fastest way out of creative paralysis is to lower the threat level.
Step 2: Generate human-first ideas
Before any AI prompt appears, you write:
obvious ideas
personal or experience-based ideas
strange or high-variance ideas
one idea you almost dismissed
The “almost dismissed” idea is important.
Many original ideas are initially rejected too quickly because they feel awkward, impractical, embarrassing, too specific or too strange. But that is often where the human signal is.
AI is very good at producing acceptable ideas.
The almost-dismissed idea is where you may find something less average.
Step 3: Use AI as a probe, not replacement
Only after you have generated your own ideas does the tool generate prompts for AI.
But these prompts are not asking the model to “give me the answer”.
They are probe prompts.
You can choose modes such as:
Reframe: give me alternative ways to understand this problem
Contradict: assume my current frame is wrong
Cross-domain: use analogies from games, architecture, education, law or other domains
Minority path: give me a non-mainstream but plausible route
Constraint inversion: turn my main limitation into an advantage
Anti-cliché: identify generic ideas and replace them with stranger but still useful versions
This is the key shift.
AI is not the author of your thinking.
AI is a probe engine.
It shakes the frame.
You still decide what matters.
Step 4: Audit the diversity of the idea set
After you paste back AI suggestions, the tool asks you to tag your ideas.
For each idea, you mark:
whether it came from you, AI or both
what kind of frame it uses
how novel it is
how useful it is
The tool then gives you simple diversity proxies.
It does not use embeddings or claim to measure true semantic diversity. The point is not to automate judgement. The point is to train your judgement.
The key question is:
Does each new idea add a new region, or just polish an old one?
That question is the practical version of the research problem.
If ten ideas all use the same frame, you do not yet have diversity.
You have variations.
If AI gives you six elegant versions of the same basic idea, you do not yet have creative expansion.
You have fluent convergence.
Step 5: Re-enter constraints
Divergence is not enough.
A strange idea is not automatically a good idea.
So the next step is to place ideas into a simple grid:
high novelty / high usefulness: develop now
high novelty / low usefulness: save as inspiration
low novelty / high usefulness: reliable baseline
low novelty / low usefulness: drop
This matters because the answer to AI homogenisation is not random weirdness.
The goal is useful originality.
You want ideas that open the search space, but can still survive contact with constraints.
Step 6: Turn one idea into a causal probe
The final step is where the H-AGI habit becomes practical.
You choose one idea and convert it into a small test:
Why might this idea work?
What do I predict will happen?
What is the smallest safe test?
What would count as success?
What would count as failure?
What could mislead me?
When will I update?
This is important because creative work is not complete when an idea sounds good.
The world still has to test it.
A title can be tested.
A lesson activity can be tested.
A product angle can be tested.
A coaching exercise can be tested.
A research framing can be tested.
AI can generate possibilities, but reality supplies the feedback.
The transfer challenge
At the end, the tool gives you a training process for real life transfer.
The next day, without AI, you return to the same method and generate five new ideas.
Then you ask:
Did I recover the human-first categories?
Did I avoid the same clichés?
Did I add at least one new frame?
Did I select by novelty and usefulness?
Did I create a small causal probe?
This is the training part.
The goal is not just to use the tool once.
The goal is to internalise the method.
Over time, the habit becomes:
I do not ask AI to replace my creative search.
I use AI to expand, probe and test it.
What this tool is for
Use Creative Diversity Gym when you want to create something where originality matters.
It is especially useful for:
article ideas
product concepts
educational activities
business positioning
research questions
creative projects
content strategy
audience tests
offer design
It is not designed for routine tasks where generic output is fine.
Sometimes you just need a quick summary, a checklist or a clean draft. That is fine.
But when originality matters, you need a different workflow.
You need to protect the search space.
The deeper point
The future of AI-assisted creativity will not be decided only by model capability.
It will also be decided by human habits.
If we all ask the same tools the same kinds of questions in the same kinds of ways, we should not be surprised if more of our work starts to feel the same.
But AI does not have to make us more average.
Used well, it can help us explore more frames, test more assumptions, generate more contrasts and build more useful ideas.
The difference is whether we use AI passively or actively.
Creative Diversity Gym is built around one simple rule:
Human first. AI second. Judgement always.
Try the tool with one creative task today.
Do not start by asking AI for ideas.
Start by writing your own.
Then use AI to probe the space.
Then ask the question that matters:
Is this adding a new region, or just polishing an old one?
Access the creativity diversity gym web app here. This app helps with prompt generation for use with your favourite AI llms!
Try it out and share your feedback. Remember, it’s built as a learning tool, to both protect and augment your own creativity.
Looks great - hope to try over the weekend