For those who aren’t on the case, Dr. John Dolittle is a fictional character created by British writer and illustrator Hugh Lofting who represents a veterinarian who has learned to talk to animals. Lofting created Dr. Dolittle to amuse his children, in the letters he sent them from the front lines in World War I, and in 1920 he published the first of his children’s novels featuring this curious protagonist. Over the past few years, dozens of versions of the Dr. Dolittle stories have been published, including several films.
Now, scientists Yossi Yovel and Oded RechavÃ, from Tel Aviv University (Israel) use the fiction of Dr. Dolittle to wonder about the possibilities that new artificial intelligence (AI) algorithms and systems can be used to make communication between humans and other animals.
“Talking to animals is a fundamental human desire. The emergence of powerful AI algorithms, and specifically generative language models, has led many to suggest that we are about to fulfill this desire,” Yovel and Rechavà write in an article. published in the journal Current Biology titled AI and Dr. Dolittle’s Challenge.
These two experts in animal biology and neurology recall that in recent months some large scientific consortia have been formed around this issue and several commercial entities even offer such services.
The authors warn that the commercial offer in this regard is, for the moment, overrated, and they frame the future challenge of what they call ‘Doctor Dolittle’s challenge’ based on three main obstacles on the way to achieving it.
The first problem with talking to other animals is that “although generative AI models [like ChatGPT] can create new samples of animal communication, it is very difficult to determine their context, and we will always be biased by our human umwelt [in semiotic theory, problem to be part of the same environment that you want to study]”.
Second, the authors note in their study abstract, “The use of AI to extract context without supervision needs to be validated by controlled experiments aimed at measuring animal response.” This type of control is technically difficult, and furthermore, “AI algorithms tend to latch on to whatever information is available and are therefore prone to finding false correlations.”
Finally, Yovel and Rechavà recall that, from what we humans know so far, “animal communication [at least the one we would be able to collect to feed AI systems] focuses on a restricted set of contexts, such as alarm and courtship, which greatly limits our ability to communicate with respect to other contexts.
The long and detailed analysis of these two scientists is certainly attractive and provides knowledge that can be of great help to computer scientists but, as they acknowledge, it is only a first approximation in a field of communication that is still a long way from becoming a reality.
The challenge is set and “using the tremendous power of new AI methods to decipher and imitate animal communication is fascinating and important,” the authors of the scientific article point out.
Once these first criteria and problems to be solved have been defined, Yovel and Rechavà invite the world scientific community to join in finding the solution to Doctor Dolittle’s challenge.
