AI Fails to Decipher Animal Language in Israeli Study

Researchers at Tel Aviv University have investigated whether artificial intelligence can accurately interpret the meaning behind animal vocalizations, a study detailed in the journal Current Biology and reported by Ynet. The team, led by Professor Yossi Yovel of Tel Aviv University's School of Zoology, collaborated with scientists from Germany and the Hebrew University of Jerusalem.

To overcome the difficulty of definitively knowing what animal sounds mean, the researchers used recordings of infants' vocalizations in three contexts: distress, calling for a parent, and requesting food. These sounds were analyzed using traditional acoustic methods and two deep neural networks, one trained on animal sounds and the other on adult human speech.

The AI models were tasked with grouping similar sounds. The expectation was that if the AI understood meaning, it would group different sounds related to requesting food together, separate from calls to parents or distress signals. However, this did not occur. While the neural networks identified acoustic patterns better than traditional methods, they frequently grouped sounds with different meanings and separated sounds with similar meanings.

Furthermore, the AI failed to recognize increasing urgency in sequences of signals, a nuance humans perceive intuitively. The study's main conclusion is that current AI primarily analyzes the sound's characteristics rather than its intended meaning, indicating that similar-sounding vocalizations do not necessarily convey similar messages. This suggests advanced AI cannot yet reliably "decode" animal language.

The researchers noted that while AI is increasingly used to study animal communication, existing models focus on acoustic properties, missing the animals' intended meaning. They propose that effective animal communication research requires combining AI with behavioral observation, playback experiments, and potentially brain activity studies. Professor Yovel emphasized that AI is a valuable tool but cannot replace direct animal observation, stating, "Identifying acoustic patterns does not necessarily mean deciphering meaning. To understand what an animal is 'saying,' we must know how the receiving animal perceives and reacts to the message."