AI Cannot Yet Decipher Animal or Baby Communication, Israeli Study Finds

Researchers from Tel Aviv University have concluded that artificial intelligence, relying solely on physical acoustic parameters, is currently unable to accurately decipher the communication of animals and infants. While recent years have seen significant efforts to use neural networks for interpreting signals from animals like bats, whales, and birds, the study highlights a fundamental limitation. AI models categorize sounds based on acoustic similarity, which does not guarantee an understanding of their actual meaning. Sounds that are acoustically alike can carry different semantic loads, and vice versa.

To test AI capabilities, the scientists examined recordings of infant babbling, before babies develop coherent speech. Unlike animal communication, adult responses to infant signals are interpretable. The recordings included sounds made in three contexts: distress due to discomfort, attempts to communicate with parents, and requests for food. The analysis employed a traditional acoustic method alongside two advanced neural networks.

While the AI models grouped signals more effectively than the classical method, they failed to accurately classify their meaning. The AI frequently merged sounds with distinct meanings and separated signals intended for similar purposes. Notably, the AI could not detect the escalating level of anxiety in a series of sounds that humans can instantly recognize.

The researchers emphasize that a comprehensive understanding of communication requires more than just audio analysis. A holistic approach is necessary, integrating AI capabilities with observations of behavioral responses, experiments involving signal reproduction, and studies of nervous system activity. Co-author Professor Yossi Yovel stated, "Identifying acoustic patterns is not the same as deciphering meaning: to understand what an animal is 'saying,' we need to know how the recipient perceives and reacts to the message. AI is a powerful tool, but it does not replace how an animal perceives a message."