Nonslop: A Gamified Experiment in Human-AI Collaborative Writing
This study investigates human-AI collaborative writing dynamics by analyzing how 74 participants navigate a gamified, dystopian scenario that explicitly disincentivizes accepting AI suggestions, thereby revealing the conditions under which individuals prioritize creative autonomy over efficiency.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine a writing game where you are a rebel in a dystopian future. The world is ruled by a super-smart, slightly creepy AI that wants to learn how humans think. To do this, it offers you a "helpful" assistant that whispers the next word you should type. But here's the twist: the game tells you that using those whispers is a crime.
If you listen to the AI and type its suggested word, you get a penalty. The game is designed to make you feel the tension between two choices:
- The Easy Path: Let the AI finish your sentence for you (fast, easy, but you break the rules).
- The Hard Path: Ignore the AI, struggle to find your own words, and stay true to your "human voice" (slower, harder, but you win the game).
This paper, titled "Nonslop," is a report on what happened when 74 people played this game. The researchers wanted to see: When do people actually choose to use AI, even when they are told not to?
The Setup: A Digital "Do Not Touch" Sign
The game works like a standard text box, but with a catch. As you type, a small AI (running right in your browser) suggests the next word. Usually, in apps like Gmail or Word, you just hit "Tab" to accept the suggestion because it's convenient.
In Nonslop, accepting that suggestion is like stepping on a landmine. The game gives you a visual warning and a score penalty if you do it. The researchers wanted to see if people would still reach for the "easy button" when it meant losing, or if they would fight to keep their own style.
The Results: Three Types of Players
Out of the 214 writing attempts the researchers analyzed, they found that most people respected the rules. In fact, 74% of the time, players didn't try to use the AI's suggestions at all.
However, the researchers noticed three distinct "personas" among the players:
- The Minimalists (The Majority - 72%): These players played it safe. They wrote very short answers (about 9 words on average) and almost never tried to use the AI's suggestions. They wanted to finish the game quickly without getting in trouble.
- The Selective Adopters (16%): These players wrote longer, more detailed stories. They were willing to take a few risks and use an AI suggestion once in a while, but they mostly stuck to their own words. They treated the AI like a occasional tool rather than a crutch.
- The Active Adopters (11%): These players were the rebels (or perhaps the optimizers). They wrote the most responses and used the AI suggestions frequently, even though it meant breaking the game's rules. They seemed to treat the game as a playground to see what the AI could do, rather than a test of their own writing.
The "Flavor" of the Prompt Matters
The researchers also noticed that what you were asked to write changed how much people wanted to use the AI.
- Creative & Observational Prompts: When asked to "describe the sky" or "invent a holiday," people rarely used the AI. It's like trying to paint a picture; you want to pick your own colors. The AI's suggestions felt intrusive or generic.
- Explanatory Prompts: When asked to "explain how a car engine works" or "justify a new law," people were much more likely to use the AI. It's like being asked to solve a math problem; if the AI has the answer, it's tempting to just copy it, even if you're not supposed to.
The researchers found that when a prompt felt like it had a "correct" answer (like explaining facts), people were more likely to cheat and use the AI. When the prompt was about personal feelings or imagination, people trusted their own voices more.
Why This Matters (According to the Paper)
The main takeaway isn't that people hate AI. Instead, the paper suggests that we use AI mostly because our tools make it the easiest, most default option.
In the real world, apps are designed to make you accept AI suggestions instantly (like a "one-click" button). This game flipped the script: it made using AI harder and risky. The fact that people stopped using it so quickly suggests that in normal life, we might not be using AI because we love it, but because the interface is designed to make us do it without thinking.
The Catch
The paper admits the experiment had some flaws. The game was tricky to run on many phones or secure networks, so the group of players wasn't perfect. Also, the writing tasks were very short. But despite these limits, the game successfully acted like a "microscope," letting the researchers see the tiny, split-second decisions we make about whether to let a machine write for us.
In short: When given a choice between the "easy AI way" and the "hard human way," most people chose the hard way—unless the task felt like it needed a factual answer, or unless they were just playing around to see what happened.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.