A study led by Tel Aviv University highlights an important limitation in attempts to use AI to "decode" animal communication.
Researchers used toddler vocalizations (where at least some intended meaning can be understood) to test whether models recognizing acoustic similarity also recognize meaning.
According to the university, advanced neural networks performed better than a classical acoustic method, but still sometimes grouped sounds with different meanings and separated sounds conveying a similar message.
The conclusion is not that AI "cannot understand animals", but that analyzing acoustic patterns alone is insufficient to prove understanding of meaning.
The study was published in Current Biology.