Humanly: A Configurable and Traceable Environment for Human-AI Collaborative Writing
Humanly is a configurable writing platform that transforms the writing process into verifiable evidence by recording detailed activity and AI assistance to generate sealed certificates, thereby enabling the detection of human-AI collaboration and distinguishing manual typing from automation across educational and professional scenarios.
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
The Mystery of the Invisible Ink
Imagine you are a detective trying to solve a mystery, but the only clue you have is the final note left on a desk. You can read the words, but you have no idea who wrote them, whether they typed them on a keyboard, dictated them to a robot, or copied them from a book. This is the current state of "writing authenticity" in our digital world. For a long time, we've tried to guess if a piece of writing was made by a human or an Artificial Intelligence (AI) just by looking at the finished product. It's like trying to tell if a cake was baked from scratch or bought from a store just by tasting a single bite. Sometimes, a human who writes very smoothly gets accused of using a robot, while a robot that writes a bit clumsily gets a free pass.
The problem is that the "finished text" doesn't tell the whole story. It hides the messy, creative, and sometimes chaotic process of how the words actually got there. Did the writer struggle with a sentence for an hour? Did they ask a smart computer for help to fix a grammar mistake? Or did a robot write the whole thing and the human just hit "submit"? To solve this, scientists in the field of computer science and language are building tools that don't just look at the cake; they watch the whole baking process. They want to create a system that records every keystroke, every pause, and every time a human asks an AI for help, turning the invisible process into a visible, trustworthy record.
Enter HUMANLY: The "Black Box" for Writers
This is where a new platform called HUMANLY comes in. Think of HUMANLY not just as a word processor, but as a high-tech, transparent "baking studio" for writing. Instead of just giving you a blank page and hoping for the best, HUMANLY lets the person in charge (like a teacher, a conference organizer, or even a writer themselves) set the rules of the game before anyone starts typing. They can decide: "You can use AI to polish your sentences, but you can't let it write the whole thing," or "You have 30 minutes, and you can't copy and paste from the internet."
Once the rules are set, the writer enters the workspace. As they type, HUMANLY acts like a super-observant security camera that isn't spying on what they are thinking, but how they are working. It records a detailed "activity log." It knows if you typed a word slowly, if you paused for a long time, if you highlighted text to ask the AI for help, or if you suddenly pasted a huge chunk of text. It even tracks if you looked away from the screen or if your typing rhythm suddenly became too fast for a human to keep up.
When the writing is done, HUMANLY doesn't just save the document. It packages the entire session into a sealed certificate. Imagine this certificate as a time-capsule that contains the final story, a replay of the writing process (like a "rewind" button for the whole session), and a special digital stamp that proves no one tampered with the record. If a teacher or a reader wants to check the work, they can open this certificate and see exactly how the document was made. They can see, for example, that "86% of the text was typed by the human, 3% was pasted, and 11% was helped by AI."
The "Typing Detective" and the Robot Test
One of the coolest parts of HUMANLY is its "Typing Detector." The researchers wanted to know if this system could tell the difference between a human tapping keys and a robot (or a computer agent) doing the typing for them. To test this, they played a game of "Red Teaming," which is like a friendly security drill where they tried to trick the system. They had humans write normally, and then they had a powerful AI agent try to control the computer and type the text for them.
The results were quite clear. The HUMANLY Typing Detector was incredibly good at spotting the robot. In their tests, it correctly identified the automated typing 85% of the time without ever falsely accusing a real human writer (a 0% false alarm rate for humans). It was also lightning fast, analyzing the typing patterns in less than a second, whereas other methods using giant language models took many seconds or even minutes to make a guess. The system looks at tiny details, like the gaps between keystrokes. Humans usually have tiny pauses and irregular rhythms; robots often type with perfect, machine-like speed or with gaps that are too short for a human finger to manage.
Why This Matters (And What It Can't Do)
The paper suggests that HUMANLY is a helpful tool for teachers, editors, and anyone who needs to trust a piece of writing. In their user study, people playing the roles of teachers, students, and reviewers found the system easy to use and felt it provided fair evidence. It helps clear up confusion when a student's writing looks "too perfect" or when a writer is unfairly accused of using AI.
However, the authors are very careful to say what this system cannot do. A HUMANLY certificate is not a magic mind-reader. It only records what happens inside the HUMANLY window. If a writer uses a different computer, dictates their story into a voice recorder, or copies text from a website that isn't connected to HUMANLY, the system won't know. It can't prove that a writer didn't get help from outside; it can only prove what it saw happening on the screen. It's a powerful tool for transparency, but it's not a perfect verdict on a person's honesty.
Ultimately, HUMANLY shifts the conversation from "Does this text look like AI?" to "Here is exactly how this text was made." It turns the writing process from a mystery into a story that can be read, checked, and trusted, one keystroke at a time.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.