AI Can Write a Perfectly Reasonable Scientific Sentence, and That’s the Problem

When working with AI-generated scientific content, it's crucial to remain vigilant for potential inaccuracies that may arise from the model's inference process. In one instance, an AI model was asked to continue a sentence about Ce³⁺-doped perovskite microlasers. The resulting statement, "The incorporation of Ce³⁺ introduces intermediate energy states that facilitate more efficient carrier relaxation, thereby reducing non-radiative energy losses and promoting population inversion, ultimately lowering the lasing threshold," appeared to be a logical continuation of the initial information.

However, the model inserted an assumption about the introduction of intermediate energy states without explicitly citing evidence from the manuscript or relevant literature to support this claim. This subtle shift from a supported statement to an unverified assumption can be easily overlooked, as the sentence maintains a fluent and coherent structure.

To mitigate this issue, researchers should carefully review AI-generated content, paying particular attention to any new assumptions or claims that do not have a clear basis in the available information. By cross-referencing the model's output with the original sources and supplementary references, researchers can better ensure the accuracy and credibility of AI-assisted scientific writing.

It's essential to recognize that AI models can provide valuable assistance in developing scientific arguments, but they should not be relied upon to generate entirely new, unsupported assertions without human oversight.