Communication breakdown: On the low mutual intelligibility between human and neural captioning
This highlights a critical communication gap between humans and AI in vision-language tasks, showing that neural models may produce superficially English-like outputs that are not intelligible to humans, which is an incremental finding adding to existing evidence.
The study compared a neural caption-based image retriever's performance using human vs. neural captions on the ImageCoDe dataset, finding it performed much better with neural captions (despite their lack of awareness of hard distractors) while humans performed near chance with the same neural captions.
We compare the 0-shot performance of a neural caption-based image retriever when given as input either human-produced captions or captions generated by a neural captioner. We conduct this comparison on the recently introduced ImageCoDe data-set (Krojer et al., 2022) which contains hard distractors nearly identical to the images to be retrieved. We find that the neural retriever has much higher performance when fed neural rather than human captions, despite the fact that the former, unlike the latter, were generated without awareness of the distractors that make the task hard. Even more remarkably, when the same neural captions are given to human subjects, their retrieval performance is almost at chance level. Our results thus add to the growing body of evidence that, even when the ``language'' of neural models resembles English, this superficial resemblance might be deeply misleading.