CEDMO Game is an innovative educational tool that allows users to interactively explore the strengths and weaknesses of artificial intelligence models in detecting false information. Developed by researchers at the Kempelen Institute of Intelligent Technologies (KInIT) as part of the CEDMO consortium, it challenges players to rephrase texts to alter an AI's assessment while maintaining the original meaning, thereby enhancing media literacy and critical thinking. The game provides insights into how AI reacts to linguistic nuances.
The CEDMO Game's core mechanism is straightforward: players are presented with a short text and simultaneously see an artificial intelligence model's evaluation regarding potential disinformation. The player's objective is to rephrase the given text, ensuring its original meaning is preserved, yet causing the AI model to change its initial assessment. This interactive process is designed for educational use, allowing teachers to create sessions, select texts, and monitor student attempts, fostering a deeper understanding of AI limitations through practical demonstration.
This section delves into the critical insight offered by the game: how artificial intelligence systems can be surprisingly sensitive to minor linguistic variations. Users directly experience that even slight alterations in wording, without changing the fundamental message, can significantly influence an AI's decision-making. By observing how easily the model can be misled, players gain firsthand knowledge of the nuances that impact AI outputs, promoting a more critical and informed perspective when interacting with AI technologies.
The game structures the learning experience into three distinct steps for players to master the art of challenging AI. Firstly, players must carefully familiarize themselves with the provided text and the AI's verdict on whether it contains misinformation. Secondly, the task is to skillfully rewrite the text, ensuring its core meaning remains intact, but strategically altering it to provoke a change in the AI's assessment. The third and final step involves revealing when and why the AI model was misled, which serves to practically demonstrate and underscore the inherent limits of current artificial intelligence methodologies.
CEDMO Game is presented as both an engaging challenge and a powerful educational instrument. It effectively illustrates that despite their advanced capabilities, contemporary AI models are not flawless. Their decision-making processes can be swayed by seemingly minor linguistic adjustments or rephrasing of information. Developing an understanding of these limitations is deemed essential for the responsible integration and use of artificial intelligence in daily life. Through gameplay, users will learn how AI classification models function, identify their strengths and weaknesses, enhance their critical thinking skills when evaluating information, comprehend the profound influence of linguistic nuances on AI judgments, and acquire practical experience regarding the boundaries of current AI technologies. The game supports user interfaces in Czech, Slovak, Polish, and English, although the underlying AI model operates exclusively in English, making it widely accessible for anyone interested in exploring the complexities of artificial intelligence.