Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Not automatically. AI can raise the judged creativity of one story or concept while making a group of AI-assisted works more alike. In some ideation tasks, it also narrows the number and variety of ideas people generate. The answer therefore depends on what “original” means, what task people perform, when AI enters the process, and whether you are judging one finished work or the diversity of many works.
For game creators, that distinction matters. An AI-assisted quest, character or visual pitch may be stronger than a first human draft, yet a whole slate of games can drift toward the same familiar structures if everyone starts from similar machine suggestions. The studies below show how to gain useful assistance without treating an AI output as the creative direction.
As an Amazon Associate I earn from qualifying purchases.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →What does “less original” mean in creative work?
Originality is not one measurement. Researchers use several outcomes that can move in opposite directions:
- Individual novelty: how unusual or new one story, image, mechanic or concept appears to evaluators.
- Quality and enjoyment: whether people rate that single work as well written, compelling or creative.
- Idea variety: how many distinct directions a person explores during ideation.
- Collective diversity: how different a set of works is from one another.
- Perceived creator creativity: what evaluators infer about the person who made the work, which can differ from the work’s score.
A system can improve the first two measures while reducing the latter ones. Calling every result “less original” hides the trade-off. A designer may produce a more polished character concept with help, while ten designers using similar prompts converge on similar characters.
#1 Best Overall
Similarity also is not the same as plagiarism or proof that a creator had no ideas of their own. The experiments summarized here measured ratings, semantic distance, idea counts, fixation or perceived creativity under controlled conditions. They do not establish a philosophical definition of authorship.
Does AI make one finished creative work less original?
In a 2024 experiment, online writers who could view up to five AI-generated story ideas produced short stories that evaluators rated as more creative, better written and more enjoyable than stories made without AI access. The improvement was especially pronounced among writers who scored lower on a pre-existing creativity measure. The study therefore found an increase, not a decrease, in the judged quality of an individual story.
The same experiment reported a measured novelty increase of 5.4% for writers given one AI-generated idea and 8.1% for writers allowed as many as five ideas, compared with writers without AI access. Those percentages describe evaluator-rated story novelty in that bounded experiment; they are not a universal boost for every medium or every prompting strategy. The authors, Anil R. Doshi and Oliver P. Hauser, summarized the trade-off as “an increase in individual creativity at the risk of losing collective novelty” (Science Advances, 12 July 2024).
There are important boundaries. Participants were typical online writers creating short fiction, not professional novelists, game studios or unusually creative specialists. They received ideas from the model rather than delegating the entire story, and the test did not cover long-form production, interactive systems or visual assets. The result supports a narrow conclusion: in that setting, access to AI suggestions helped many individuals make a story that evaluators liked more.
Can AI-assisted works become more alike?
Yes. In the same short-story experiment, AI-assisted stories were more similar to one another than stories written by participants who had no AI access. A single story could score higher while the collection lost variety. This is a population-level effect, so it matters when a publisher, studio or class compares a portfolio rather than one submission.
A separate 2025 study examined 2,200 college-admissions essays across three preregistered studies. Each additional human-written essay contributed more new ideas to the collection than each additional GPT-4 essay. The difference remained when the researchers changed prompts and model parameters. That finding is evidence about essay diversity, not proof that every genre, model or creative workflow homogenizes in the same way (Computers in Human Behavior: Artificial Humans).
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteFor games, collective similarity can appear in a pitch deck before it is visible to players: the same quest motivations, color scripts, enemy archetypes or progression loops recur because teams begin with comparable machine-generated suggestions. The risk is greatest when a team accepts the first plausible direction and uses AI to elaborate it repeatedly, rather than using the system to open several genuinely different branches.
What happens during visual ideation?
The result changes with the task. In a 2024 visual-ideation experiment involving N=60 participants, support from an AI image generator was associated with greater fixation on an initial example. Participants generated fewer ideas, with less variety and lower originality than the baseline condition (CHI 2024 paper).
Fixation means that an early image becomes an anchor: later concepts stay near its composition, subject or visual language. That pattern does not contradict the short-story result. The studies used different media, controls and outcome measures. An image reference can make one direction easier to elaborate while simultaneously suppressing exploration of alternatives.
A 2026 systematic review and meta-analysis covering 19 studies and 61 effect sizes found a small but statistically significant overall homogenization effect in human–AI co-creation. The effect was stronger in semantically constrained ideation—tasks that already specify a topic, format or meaning—than in minimally constrained divergent-thinking tasks (2026 review). Its pooled estimate is a pattern across studies, not a claim that every AI session reduces creativity.
Does the timing of AI assistance change perceived creativity?
Timing can affect how observers judge the creator, even when the finished artifact is usable. In three preregistered experiments with N=940, a 2026 proceedings report found that evaluators perceived creators as less creative when AI entered early rather than late in the process. Elaborating substantially on AI suggestions reduced that penalty (2026 proceedings report).
“Early” and “late” are process descriptions, not fixed software settings. Early use can mean asking for the initial premise, visual reference or mechanic before a human direction exists. Late use can mean testing edge cases, tightening prose, debugging a script or generating alternatives after the creator has chosen the core idea. The finding concerns evaluator perception in controlled experiments; it does not show that late use always produces better games.
How do the main studies differ?
Comparing the evidence requires keeping the unit of analysis and task visible. This table places the results side by side.
| Study | Participants or material | AI role and task | What changed | What it does not establish |
|---|---|---|---|---|
| Doshi and Hauser, 2024 | Online writers making short stories | Access to one or up to five AI story ideas | Higher evaluator-rated creativity, writing quality, enjoyment and novelty; AI-assisted stories were more alike | Effects for professional writers, games or all creative media |
| Wadinambiarachchi and colleagues, 2024 | N=60 participants | AI image-generator support during visual ideation | More fixation, fewer ideas, less variety and lower originality than baseline | Every visual workflow or model |
| Essay-diversity study, 2025 | 2,200 college-admissions essays in three preregistered studies | Human-written essays compared with GPT-4 essays under varied prompts and parameters | Each additional human essay added more new ideas than each additional GPT-4 essay | Interactive entertainment, fiction quality or individual authorship |
| Systematic review, 2026 | 19 studies; 61 effect sizes | Human–AI co-creation across tasks | Small overall homogenization effect, stronger in semantically constrained ideation | A universal reduction in creativity |
| Mahaney, Majumder and Martuza, 2026 | N=940 across three preregistered experiments | AI introduced early or late; evaluators judged creators | Early entry lowered perceived creativity; elaboration on suggestions reduced the effect | Objective quality of every finished artifact |
What does this mean for game design?
Game development combines several creative tasks, so the answer can change within one project. A team might use a language model for quest premises, an image model for mood-board references and code assistance for implementation. Each stage has a different exposure to fixation and convergence.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Writing and narrative design
AI can help a writer move past a blank page or strengthen a weak draft. To protect variety, generate directions only after the team has listed its own premises, and ask for contrasts rather than ten variations on one premise. Keep discarded human ideas visible so the group can compare the final choice with paths the model never proposed.
Mechanics and systems
A model can restate familiar mechanics effectively, which is useful for prototyping but dangerous as a source of novelty. Define the player experience, constraints and non-negotiable differences first. Treat machine suggestions as a challenge list—“what would make this loop unlike the obvious genre version?”—rather than as the design brief.
Visual development
The N=60 visual study makes early reference images a particular caution. Start with thumbnails, written adjectives, physical sketches or deliberately incompatible references before asking an image tool to elaborate. If one generated image becomes the anchor, pause and produce alternatives with changed silhouette, camera, palette and material logic instead of polishing the anchor indefinitely.
Implementation and iteration
Late-stage assistance is less likely to determine the game’s identity. Code review, test-case generation, localization drafts and accessibility checks can reduce routine workload while leaving the core creative decisions with the team. Review generated changes for hidden assumptions and keep a human-authored specification as the source of truth.
How can creators use AI without narrowing their options?
The following workflow is a practical response to the measured risks; it is not a guarantee of originality.
- Make a human first pass. Write a premise, mechanic or visual brief before requesting machine ideas. Record the constraints that make the project distinctive.
- Ask for dissimilarity. Request mutually incompatible directions and specify what each must avoid. Reject outputs that merely swap nouns in the same template.
- Separate exploration from selection. Generate a broad set, then step away before choosing. This reduces the chance that the first plausible suggestion becomes the default.
- Change the frame, not just the wording. For a quest, alter the player goal, power relationship, time scale and failure condition. For art, alter silhouette, viewpoint, material and cultural reference.
- Have humans recombine. Combine fragments from different directions and add details that came from playtesting, team experience or the project’s fiction bible.
- Use AI later for refinement. Editing, debugging, consistency checks and edge-case generation are less likely to dictate the central concept than asking for the first idea.
- Run a convergence review. Compare the team’s concepts side by side. Flag repeated themes, layouts, verbs, color relationships or progression patterns before production locks them in.
- Preserve provenance. Keep prompts, source sketches, revisions and human decisions. A clear record helps the team explain which choices are original direction and which are machine-assisted elaboration.
What are the evidence’s limits?
- The short-story experiment was bounded: online participants, short fiction and up to five ideas. It did not test professional game teams or long production cycles.
- The visual result came from a specific task with 60 participants. It should not be generalized to every image model or art pipeline.
- The essay study measured idea contribution and diversity in admissions writing, not whether a single essay was enjoyable or artistically successful.
- The 2026 meta-analysis pools heterogeneous studies. Its small average effect can coexist with positive results in particular tasks.
- Evaluator ratings and semantic similarity are observable measures, not final judgments about artistic value, intent or moral authorship.
- Models, interfaces and user habits change. A result tied to one model version or interface may not transfer unchanged to another.
These limits are why a responsible claim is conditional: AI can improve an individual output, yet certain workflows make outputs or ideas converge. The evidence supports measuring both the quality of each artifact and the diversity of the set.
What should a studio measure in an AI-assisted project?
Do not rely on a single “creativity score.” At milestone reviews, track a small dashboard:
- Individual evaluation: blind ratings for novelty, coherence, usefulness and enjoyment.
- Portfolio diversity: semantic similarity or a structured comparison of themes, mechanics, silhouettes and progression patterns.
- Exploration breadth: number of genuinely distinct concepts before selection, not the number of near-duplicates generated.
- Process timing: whether AI supplied the premise, an intermediate option or a late refinement.
- Human contribution: which decisions came from playtests, sketches, writing rooms and design constraints.
If one metric rises while another falls, report the trade-off instead of averaging it away. A polished prototype and a diverse slate are both valuable, but they answer different production questions.
So, does using AI make creative work less original?
Sometimes, under specific conditions—but not by default. Individual works can receive higher creativity and novelty ratings with AI assistance, while collections of AI-assisted works can become more similar. Visual ideation can suffer from fixation, and early AI involvement can lower perceived creator creativity. The practical test is not whether AI was used; it is whether the workflow preserved independent exploration and added human decisions that materially changed the result.
Frequently Asked Questions
Is a more similar set of AI-assisted works automatically bad?
Not always. A shared visual language or mechanic can be an intentional franchise or genre choice; the problem is unintended convergence that removes meaningful alternatives. Teams should decide how much variation they need before judging similarity.
How can a small team check for convergence without specialized software?
Place concepts on one page and annotate repeated verbs, themes, silhouettes, camera angles, color relationships and progression loops. Ask each creator to explain the idea’s origin and identify one deliberate departure from the group’s most common pattern.
Does late-stage AI use guarantee that a creator will be seen as original?
No. The 2026 timing experiments found a perception pattern, not a guarantee. A late edit can still erase distinctive choices, while an early suggestion can become original after substantial human transformation.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchShould a game team ban AI during brainstorming?
A blanket ban is not required by the evidence. A staged policy—human-first concepting, deliberately diverse alternatives, documented decisions and later assistance for refinement—addresses fixation and convergence while retaining potential productivity gains.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
