Human-Plus-Machine Co-Creation Loop
Use AI to generate variations, then let human judgment direct each iteration
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 96%
The creator begins with an intention and supplies original work, constraints, or source material to an AI system. The model generates variations that expose combinations the creator might not have considered. A human then judges those outputs, selects promising elements, and revises the work using personal taste and expertise. That revision becomes the input for another round, producing an alternating loop of machine expansion and human selection. The machine contributes speed and breadth; the human supplies meaning, standards, context, and final responsibility. This approach avoids the low-quality flood likely to result from fully autonomous generation while using AI to reach creative destinations that would have been difficult to discover unaided.
Origin
Extracted from Masters in Business through Jim O'Shaughnessy's explanation of AI-assisted artists using Wand to generate and refine variations of their own work.
Core principles
- 01Let machines expand the option set
- 02Keep human taste and intent in control
- 03Use personal work as the creative starting point
- 04Iterate between generation and judgment
- 05Treat AI as a collaborator rather than an autonomous author
How to run it
- 1
Set the creative intent
State what the work should accomplish, evoke, or communicate before generating anything.
Pro tip Include non-negotiable stylistic or factual constraints.
Watch out A vague objective encourages generic output.
- 2
Supply authentic material
Give the model the creator’s own draft, artwork, references, or research as the foundation for exploration.
Pro tip Use representative examples that reveal the creator’s actual voice or style.
Watch out Do not provide material the creator lacks permission to use.
- 3
Generate variations
Ask for multiple distinct interpretations, combinations, or extensions rather than one supposedly final answer.
Pro tip Request deliberate diversity across composition, tone, or approach.
Watch out Quantity without meaningful variation creates review noise.
- 4
Apply human judgment
Evaluate each option against intent, quality, truth, and taste. Select useful ideas rather than accepting an output wholesale.
Pro tip Look for directions that are both surprising and recognizably compatible with the creator.
Watch out Novelty alone is not quality.
- 5
Revise and return
Modify the selected direction yourself and submit the revision for another focused round of generation.
Pro tip Narrow the prompt as the work converges.
Watch out Repeatedly regenerating without human revision weakens authorship.
- 6
Own the final decision
Stop iterating when the work satisfies the creator’s standards and intended audience.
Pro tip Perform a final factual, ethical, and rights review.
Watch out The human remains accountable for the published result.
In the wild
A graphic artist feeds original work into Wand and requests variations. One output suggests a treatment the artist had never considered. The artist selects it, revises it through personal judgment, and sends the revision back for another round until the final piece remains distinctly theirs but reaches an unexpected destination.
→ The artist expands the creative search space while retaining authorship.
A writer supplies a draft chapter and asks an AI for openings that emphasize different tensions. The writer rejects generic options, combines two useful ideas, rewrites the passage in their own voice, and asks the model to identify remaining clarity gaps.
→ The chapter improves through accelerated exploration and human-controlled revision.
Common mistakes
Publishing the first generation
Initial AI output is an option to evaluate, not a finished artifact with guaranteed quality.
Removing the human loop
Without expert selection and revision, generated content tends toward generic or poor results.
Letting the tool erase the creator
The process should amplify the creator’s intent and source material rather than substitute a model’s defaults.
Is it for you?
Best for
It is best for creators who possess clear taste, domain knowledge, and original material but want to widen their range of options.
Not ideal for
It is not ideal for generating finished work without human review, direction, or accountability.
From the transcript
“The magic comes when you add a human in the loop.”
“The magic comes by being able to partner with that and co create”
“this is an iterative process.”
From the episode
Jim O'Shaughnessy on How AI Will Change Everything From Arts to Stocks