Imagine if, in the future, CatGPT joined ChatGPT, allowing us to converse with our cat companions through meows, purrs, or hisses.
Many people are curious about what it would take for AI to communicate with animals now that massive language models have mastered the art of understanding human speech.
In a recent editorial that was just published in Current Biology, two neurobiologists explore the solution.
Yossi Yovel and Oded Rechavi from Tel Aviv University in Israel present the “Doctor Dolittle challenge” as a tribute to the Turing test, which establishes a benchmark for artificial intelligence.
When attempting to communicate with an animal, the task calls for an AI-based, sizable ‘language’ model to get over three major roadblocks.
It must make use of the animal’s natural communication cues. The animal must not pick up any new signals, like a dog picking up on commands from a human.
It must employ these signals in a variety of behavioral scenarios, not simply in hostile or courtship situations, as scientists have demonstrated by playing back to birds a known avian alarm sound.
Additionally, the animal must have a quantifiable response “as if it were communicating with a conspecific [an animal similar to itself] and not a machine.”
Consider the honey bee as an example. To inform the colony of the location of food, it engages in a waggle dance. With the use of this information, scientists were able to build a robotic bee that can attract other bees and direct them to a specified location.
The first and third requirements of the Doctor Dolittle challenge are met by this. The dance, however, is limited to this particular situation. Scientists are still unable to elicit a bee’s desires or emotional state.
Additionally, even if we check off all three of the aforementioned boxes, we might never be able to interact with animals as effectively as many pet owners or animal enthusiasts would like.
One day, an algorithm might be able to detect whether our pet cat is feeling affection or annoyance, but there might not be a means to elicit that information. It’s possible that human language is just distinct from that of other animals in certain respects. All that we can understand is constrained by our umwelt, or “self-centered world.”
Even if a lion could communicate, according to Ludwig Wittgenstein, “we could not understand it.”
The authors Yove and Rechavi argue that if Wittgenstein is correct, “we will never be able to ask [a cat] “how they feel” or explain that ChatGPT already means CatGPT in French (and that it might be humorous).
That bee dance we believed we could perform? According to Yovel and Rechavi, it probably carries a lot more information than we’ve discovered, “including subtle tactile and acoustic signals about the quality of the resource.”
The authors continue, “We are not even sure which other types of data would need to be recorded, but these data would also need to be collected and fed into the AI algorithm if it were asked to crack the code.”
Would we also have to capture electric fields?
On the other hand, since primate communication is most similar to our own, mastering it might be simpler. However, massive amounts of data would still need to be collected in order to train AI models, necessitating continuous monitoring of monkeys in the wild. Where would that data originate?
Even if information could be collected and used, researchers would need to observe a ‘natural response’ from monkeys, demonstrating that they had heard and understood an artificial device’s attempt to communicate with them.
While neural recordings could be useful in this regard, demonstrating objective knowledge in other circumstances might be outright impossible.
Yovel and Rechavi believe AI can be used in the future to better understand animal communication, but they acknowledge that it might not be able to assist us interact with animals in the same way that Doctor Doolittle did.
Even if artificial intelligence (AI) gains “a million fold” in strength, some of the barriers that currently prevent us from communicating with animals will still exist, according to neurobiologists.
“Even if we will never be able to talk to animals in the human way, understanding how complex animal communication is and attempting to tap into and mimic it is a fascinating scientific endeavor,” the researchers write in their findings.
We therefore urge researchers to use AI to interpret animal communication in accordance with the requirements of the Doctor Dolittle challenge…
The essay was published in Current Biology.
