Recent correspondence in Nature highlights the growing integration of large language models (LLMs) in scientific work. However, the issue of addressing academic misconduct in this context demands a more critical approach. The use of LLMs to generate texts that researchers then sign with their own names, without proper disclosure, violates fundamental principles of academic integrity. This is not genuine scientific inquiry but its mere imitation, and appeals to fears of “stigmatization” should not excuse such practices. If a researcher shifts their core intellectual activities onto AI without meaningful contributions of their own, it raises a pressing question: why allocate grants or retain such individuals in academic positions when their work could just as well be performed by a machine?
Voluntary disclosure systems for LLM usage are unlikely to succeed, as those engaging in misconduct are unlikely to admit it. In this context, watermarks – such as those developed by DeepMind – are crucial tools for detecting academic dishonesty, combating misinformation, and preventing the degradation of AI models trained on machine-generated content.
Science must remain a domain of genuine intellectual discovery, not a venue for hidden manipulations. Achieving this requires mandatory mechanisms for disclosure, clear penalties for misconduct, and robust oversight by the academic community. Without these safeguards, the credibility of the scientific process itself is at stake.
Немає коментарів:
Дописати коментар