Nonslop: A Gamified Experiment in Human-AI Collaborative Writing

Julian Togelius; Maria Edwards

arxiv: 2606.12350 · v1 · pith:WVGB2N3Gnew · submitted 2026-06-10 · 💻 cs.AI

Nonslop: A Gamified Experiment in Human-AI Collaborative Writing

Maria Edwards , Julian Togelius This is my paper

Pith reviewed 2026-06-27 10:01 UTC · model grok-4.3

classification 💻 cs.AI

keywords gamified writinghuman-AI interactioncreative autonomyAI suggestionsdystopian narrativeLLM collaborationindividual expressionauthenticity

0 comments

The pith

A dystopian gamified writing task reveals when people choose creative autonomy over AI suggestions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a writing game that places AI word suggestions in front of users but frames the entire exercise as a dystopian scenario in which an AI is trying to absorb what remains of human individuality. By explicitly penalizing AI-like output through the narrative and rules, the design inverts the usual helpful-assistant pattern to reduce default acceptance of suggestions. Data from 74 participants and 214 responses are then examined for patterns in when users maintain their own voice versus accepting assistance, across different prompts and response styles. The resulting observations are meant to illuminate genuine preferences rather than convenience-driven behavior in AI-augmented creativity.

Core claim

The central claim is that embedding AI suggestions inside a game whose story and scoring system actively discourage their use creates observable conditions under which participants reveal authentic choices between preserving individual expression and accepting machine assistance, thereby supplying a method for studying human-AI creative interaction that is less contaminated by the default helpfulness of ordinary interfaces.

What carries the argument

The dystopian narrative and rule set that presents AI suggestions while explicitly disincentivizing their adoption to protect human individuality.

If this is right

Behavior patterns will vary systematically with task type, showing which prompts most strongly elicit resistance to AI input.
Response characteristics such as length, originality, and deviation from suggestions will correlate with whether users obey or break the game rules.
The setup supplies a reusable framework for collecting data on human-AI creative choices that is distinct from typical assistant interactions.
The experiment directly tests the tension between efficiency gains from AI and the desire to retain personal voice.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same narrative-framing technique could be tested in non-writing domains such as image editing or music composition to measure autonomy preferences.
If the method succeeds, tool designers might deliberately add friction or narrative context to AI interfaces when user agency is the priority.
Longer-term use of such games might reveal whether repeated exposure changes participants' baseline willingness to accept AI help outside the game.

Load-bearing premise

The game narrative and penalties successfully make participants act on their real preference for autonomy rather than simply complying with or rebelling against the artificial constraints.

What would settle it

If acceptance rates of AI suggestions remain as high as those observed in standard non-gamified writing interfaces, the dystopian framing has not produced the intended shift toward authentic preferences.

Figures

Figures reproduced from arXiv: 2606.12350 by Julian Togelius, Maria Edwards.

**Figure 2.** Figure 2: Main game interface for Nonslop. Users respond to a prompt by entering free-text into the input field. Live word statistics track original and AI-suggested words, and the interface displays level, mode, and a timer. Nextword suggestions from the local LLM appear as autocomplete while typing (not shown here). Data stored included the OS, browser, time of access, and submission content of each user who acce… view at source ↗

**Figure 4.** Figure 4: Prompt-level comparison of the number of submissions per prompt [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

read the original abstract

The rapid proliferation of large language models (LLMs) raises critical questions about human creativity and individual expression in an era of AI-assisted creation. When do humans adopt AI suggestions, and what are the implications for individual voice? This study examines these questions through a gamified writing exercise where 74 participants (214 responses) replied to prompts while AI-generated word suggestions were available as they wrote. The game simulates a dystopian future in which an AI is attempting to learn from what remains of human individuality, and disincentivizes AI-like writing. In doing so, it attempts to create conditions that reveal authentic user preferences rather than default behaviors, such as accepting a readily available AI-generated suggestion. Note that this is a deliberate inversion of the "helpful assistant" design pattern; the system is explicitly forbidding you from accepting AI suggestions. We analyze user behavior patterns across different task types, user behaviors, and response characteristics to understand the factors influencing human-AI interaction in creative tasks. The study focuses on when users choose to maintain creative autonomy versus violating the rules of the game and accepting AI assistance. It also explores how these choices relate to response patterns, task characteristics, and user behavior. This gamified approach offers both a framework for studying authentic human-AI interaction and a provocative lens for understanding the tension between efficiency and authenticity in AI-augmented creativity.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper describes a game setup to discourage AI suggestions in writing via a dystopian story but reports no results or validation that the framing works as intended.

read the letter

The core of this paper is a gamified writing task with 74 participants producing 214 responses. The setup presents AI word suggestions but frames the whole thing as a dystopian scenario where an AI is trying to absorb human individuality, and the rules explicitly forbid taking the suggestions. The goal is to see when people stick to their own words anyway.

What stands out is the deliberate flip of the usual helpful-AI pattern. Most work tries to make suggestions easy to accept; this one makes acceptance a rule violation. That inversion is a clean way to look at the efficiency-versus-authenticity tension, and the participant count is reasonable for an initial behavioral probe. The task types and response characteristics they track also line up with standard HCI questions in this area.

The main gaps are straightforward. The description stops at the design and the plan to analyze patterns; there are no numbers on acceptance rates, no breakdowns by task, and no statistical tests. Without those, it's impossible to tell what the data actually showed. The stress-test concern also holds up: the abstract claims the framing reveals authentic preferences, but there's no control condition, no pre/post check, and no self-report data to show that refusals aren't just people following the explicit rule. That leaves the central claim about authenticity untested.

This is the kind of thing that might interest someone running their own user studies on AI writing tools, especially if they want ideas for disincentivizing conditions. It doesn't add a new framework or a verified finding yet. I wouldn't bring it to a reading group in its current form, and it doesn't look ready for peer review until the results and some validation of the game mechanics are added.

Referee Report

2 major / 2 minor

Summary. The paper describes a gamified writing experiment ('Nonslop') with 74 participants (214 responses) who completed prompts while AI-generated word suggestions were available. The setup uses a dystopian narrative in which an AI seeks to learn from human individuality and explicitly prohibits accepting suggestions, framed as a deliberate inversion of typical helpful AI designs. The study claims to analyze when users maintain creative autonomy versus accepting AI assistance, relating these choices to task types, response characteristics, and user behaviors, and positions the approach as both a framework for studying authentic human-AI interaction and a lens on the tension between efficiency and authenticity.

Significance. If the empirical results were to show that the dystopian framing and explicit prohibition successfully surface genuine preferences for autonomy (rather than rule compliance), the work would offer a novel methodological contribution to empirical studies of human-AI collaboration in creative writing by providing a controlled inversion of standard assistant paradigms and generating falsifiable observations about adoption decisions.

major comments (2)

[Abstract] Abstract: The manuscript states that 'We analyze user behavior patterns across different task types, user behaviors, and response characteristics' and that 'It also explores how these choices relate to response patterns, task characteristics, and user behavior,' yet the provided text contains no results, statistical analyses, tables, figures, or quantitative breakdowns of acceptance rates, task variations, or correlations. This absence makes it impossible to assess the central claims about factors influencing autonomy choices or the proposed framework.
[Abstract] Abstract, paragraph 2: The design is presented as creating 'conditions that reveal authentic user preferences rather than default behaviors' via the dystopian narrative and explicit prohibition. However, no control condition, baseline neutral-instruction comparison, pre/post framing measures, or validation (such as self-report motivation data or differential refusal rates) is described to demonstrate that observed refusals reflect underlying preferences for autonomy instead of compliance with the game rules. This undercuts the interpretation that the setup isolates authentic preferences.

minor comments (2)

[Abstract] Abstract: The parenthetical note 'Note that this is a deliberate inversion of the "helpful assistant" design pattern; the system is explicitly forbidding you from accepting AI suggestions' is written in second person and disrupts the formal tone; it should be rephrased or integrated into the third-person description of the system.
[Abstract] Abstract: The claim that the approach offers 'a provocative lens for understanding the tension between efficiency and authenticity' is asserted without any reported data linking observed behaviors to efficiency/authenticity trade-offs; if results are added, this interpretive claim should be tied directly to specific findings.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We appreciate the referee's constructive feedback on our manuscript. We address each major comment below and outline our planned revisions.

read point-by-point responses

Referee: [Abstract] Abstract: The manuscript states that 'We analyze user behavior patterns across different task types, user behaviors, and response characteristics' and that 'It also explores how these choices relate to response patterns, task characteristics, and user behavior,' yet the provided text contains no results, statistical analyses, tables, figures, or quantitative breakdowns of acceptance rates, task variations, or correlations. This absence makes it impossible to assess the central claims about factors influencing autonomy choices or the proposed framework.

Authors: The current version of the manuscript primarily describes the experimental design and framework. While the abstract summarizes the intended analyses based on the collected data from 74 participants and 214 responses, the detailed results, statistical analyses, tables, and figures are not included in the provided text. We will revise the manuscript to include a dedicated results section with quantitative breakdowns, acceptance rates by task type, correlations with response characteristics, and supporting visualizations to substantiate the claims. revision: yes
Referee: [Abstract] Abstract, paragraph 2: The design is presented as creating 'conditions that reveal authentic user preferences rather than default behaviors' via the dystopian narrative and explicit prohibition. However, no control condition, baseline neutral-instruction comparison, pre/post framing measures, or validation (such as self-report motivation data or differential refusal rates) is described to demonstrate that observed refusals reflect underlying preferences for autonomy instead of compliance with the game rules. This undercuts the interpretation that the setup isolates authentic preferences.

Authors: We agree that the absence of a control condition or additional validation measures makes it challenging to conclusively attribute refusals to authentic preferences for autonomy rather than adherence to the game's explicit rules. The dystopian framing and prohibition are central to the inversion of standard AI designs, but to strengthen the interpretation, we will revise the abstract and discussion sections to more cautiously describe the setup as creating conditions that discourage AI use, without overclaiming isolation of authentic preferences. If possible with the existing dataset, we will explore adding any available self-report data or behavioral indicators. revision: partial

Circularity Check

0 steps flagged

Empirical user study contains no derivations or self-referential reductions

full rationale

The paper describes a gamified writing experiment with 74 participants and reports observed behaviors across task types and response characteristics. No equations, fitted parameters, predictions, or derivation chains appear in the abstract or described structure. Claims about revealing 'authentic user preferences' rest on the game design and data collection rather than any mathematical step that reduces to its own inputs by construction. The work is self-contained as an empirical report with no load-bearing self-citations or ansatzes that could trigger circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper is an empirical behavioral study with no mathematical axioms, free parameters, or invented entities.

pith-pipeline@v0.9.1-grok · 5769 in / 1032 out tokens · 31820 ms · 2026-06-27T10:01:06.865404+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

13 extracted references · 5 canonical work pages

[1]

, title =

Photon Storm Ltd. , title =. 2025 , url =

2025
[2]

Co-Writing with

Mohi Reza and Jeb Thomas-Mitchell and Peter Dushniku and Nathan Laundry and Joseph Jay Williams and Anastasia Kuzminykh , year=. Co-Writing with. doi:https://doi.org/10.1145/3757566 , url=. 2504.12488 , archivePrefix=

work page doi:10.1145/3757566
[3]

International Conference on Computational Creativity , year=

Deeptingle , author=. International Conference on Computational Creativity , year=
[4]

From game design elements to gamefulness:

Deterding, Sebastian and Dixon, Dan and Khaled, Rilla and Nacke, Lennart , booktitle=. From game design elements to gamefulness:
[5]

Dhillon and Somayeh Molaei and Jiaqi Li and Maximilian Golub and Shaochun Zheng and Lionel P

Paramveer S. Dhillon and Somayeh Molaei and Jiaqi Li and Maximilian Golub and Shaochun Zheng and Lionel P. Robert , year=. Shaping. 2402.11723 , archivePrefix=

work page arXiv
[6]

Arnold and Jiho Kim , year=

Kenneth C. Arnold and Jiho Kim , year=. Interaction-Required Suggestions for Control, Ownership, and Awareness in. 2504.08726 , archivePrefix=

work page arXiv
[7]

2024 , eprint=

Inspo: Writing with Crowds Alongside AI , author=. 2024 , eprint=

2024
[8]

PLOS ONE , year=

Adoption of AI writing tools among academic researchers: A Theory of Reasoned Action approach , author=. PLOS ONE , year=
[9]

In: Proc

Lee, Hao-Ping (Hank) and Sarkar, Advait and Tankelevitch, Lev and Drosos, Ian and Rintel, Sean and Banks, Richard and Wilson, Nicholas , title =. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , articleno =. 2025 , isbn =. doi:10.1145/3706598.3713778 , abstract =

work page doi:10.1145/3706598.3713778 2025
[10]

Encyclopedia , VOLUME =

Christopoulos, Athanasios and Mystakidis, Stylianos , TITLE =. Encyclopedia , VOLUME =. 2023 , NUMBER =

2023
[11]

von Ahn, Luis and Dabbish, Laura , title =. Commun. ACM , month = aug, pages =. 2008 , issue_date =. doi:10.1145/1378704.1378719 , abstract =

work page doi:10.1145/1378704.1378719 2008
[12]

2024 , howpublished =

Neal Agarwal , title =. 2024 , howpublished =

2024
[13]

2021 , howpublished =

Parrish, Allison , title =. 2021 , howpublished =

2021

[1] [1]

, title =

Photon Storm Ltd. , title =. 2025 , url =

2025

[2] [2]

Co-Writing with

Mohi Reza and Jeb Thomas-Mitchell and Peter Dushniku and Nathan Laundry and Joseph Jay Williams and Anastasia Kuzminykh , year=. Co-Writing with. doi:https://doi.org/10.1145/3757566 , url=. 2504.12488 , archivePrefix=

work page doi:10.1145/3757566

[3] [3]

International Conference on Computational Creativity , year=

Deeptingle , author=. International Conference on Computational Creativity , year=

[4] [4]

From game design elements to gamefulness:

Deterding, Sebastian and Dixon, Dan and Khaled, Rilla and Nacke, Lennart , booktitle=. From game design elements to gamefulness:

[5] [5]

Dhillon and Somayeh Molaei and Jiaqi Li and Maximilian Golub and Shaochun Zheng and Lionel P

Paramveer S. Dhillon and Somayeh Molaei and Jiaqi Li and Maximilian Golub and Shaochun Zheng and Lionel P. Robert , year=. Shaping. 2402.11723 , archivePrefix=

work page arXiv

[6] [6]

Arnold and Jiho Kim , year=

Kenneth C. Arnold and Jiho Kim , year=. Interaction-Required Suggestions for Control, Ownership, and Awareness in. 2504.08726 , archivePrefix=

work page arXiv

[7] [7]

2024 , eprint=

Inspo: Writing with Crowds Alongside AI , author=. 2024 , eprint=

2024

[8] [8]

PLOS ONE , year=

Adoption of AI writing tools among academic researchers: A Theory of Reasoned Action approach , author=. PLOS ONE , year=

[9] [9]

In: Proc

Lee, Hao-Ping (Hank) and Sarkar, Advait and Tankelevitch, Lev and Drosos, Ian and Rintel, Sean and Banks, Richard and Wilson, Nicholas , title =. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , articleno =. 2025 , isbn =. doi:10.1145/3706598.3713778 , abstract =

work page doi:10.1145/3706598.3713778 2025

[10] [10]

Encyclopedia , VOLUME =

Christopoulos, Athanasios and Mystakidis, Stylianos , TITLE =. Encyclopedia , VOLUME =. 2023 , NUMBER =

2023

[11] [11]

von Ahn, Luis and Dabbish, Laura , title =. Commun. ACM , month = aug, pages =. 2008 , issue_date =. doi:10.1145/1378704.1378719 , abstract =

work page doi:10.1145/1378704.1378719 2008

[12] [12]

2024 , howpublished =

Neal Agarwal , title =. 2024 , howpublished =

2024

[13] [13]

2021 , howpublished =

Parrish, Allison , title =. 2021 , howpublished =

2021