A Japanese startup has launched Puddin AI, a system designed to distinguish human-written content from AI-generated text by analysing the writing process itself. The tool is already being tested at universities and could reshape how academic originality is assessed.
TOKYO: Can artificial intelligence really tell whether a piece of writing was created by a human or generated by AI?
A Japanese technology startup believes it can. Osaka-based Valar Intelligence has developed a new platform called Puddin AI, designed to identify whether academic papers and other written documents were produced by a person, an AI system, or a combination of both.
As the use of generative AI tools continues to grow in classrooms, workplaces and research institutions, concerns about originality and authorship have become increasingly common. The launch of Puddin AI reflects a broader effort to address those concerns while promoting transparency in academic work.
Unlike traditional AI-detection software that analyses only the final text, Puddin AI focuses on the entire writing journey. Users must write and submit their work directly through the platform, allowing the system to observe how the document is created from start to finish.
The platform records details such as when a user begins writing, how quickly they type, the sequence of edits they make and the time spent developing the document. These behavioural patterns are then analysed using approximately 200 indicators designed to measure what the company describes as the “humanness” of a text.
According to Valar Intelligence, common human characteristics such as spelling mistakes, pauses during writing and the estimated time typically required to complete a document are all taken into account. These factors help the system determine whether a piece of work was written naturally by a person.
Puddin AI can also detect situations where content may have been generated by an AI tool and pasted into the platform. Because such actions dramatically shorten the writing process, the system can identify unusual patterns that differ from normal human writing behaviour.
The results are classified into three categories: AI, AI-supported and Human. This approach allows educators and organisations to gain a more nuanced understanding of how a document was created rather than relying on a simple pass-or-fail assessment.
The system currently supports Japanese, English and four additional languages. Its developers say it can be used for university assignments, academic papers and corporate documents.
Interest in the technology is already growing. Kyushu University has begun experimenting with Puddin AI in classroom settings, while several dozen universities are reportedly considering adopting the platform.
Andrew John Chapman, an associate professor of energy economics at Kyushu University, welcomed the initiative. He said he wants students to spend time producing their own work and believes the system can help verify originality while supporting fair evaluation.
As educational institutions worldwide continue to adapt to the rapid rise of generative AI, tools such as Puddin AI may become increasingly important. By focusing on how content is written rather than simply analysing the finished product, the technology offers a new approach to balancing innovation with academic integrity.
