RENOWNED LSD proponent John Lilly’s attempts to speak with dolphins were certainly inventive. In experiments over decades, he variously plied the animals with his favourite drug, flooded a house to allow a human to live side by side with one and even tried to commune with them telepathically.
His failure has shared the fate of most efforts to do a Dr Dolittle and talk to the animals. The orthodox position is that human language – the sort that allows us to exchange pleasantries about the weather or discuss abstract concepts such as the price of fish – is our sole preserve. If you have ever dreamed of listening to a whale’s tales of the deep ocean or asking your dog why it howls at the vacuum cleaner, dream on.
Advertisement
Or, perhaps, wake up to a coming reality. Some researchers think that soon we could finally break through the human-animal language barrier, a belief fuelled not by psychedelic optimism, but by the data-crunching smarts of artificial intelligence. Our relationship with the animal world may never be the same again.
AI is good at language. Today, our email services can complete sentences for us, our browsers automatically translate web pages and voice assistants decode our commands. Earlier this year, research company OpenAI released a system called GPT-3 that can write compelling prose from scratch.
Decoding animal communication is just a logical next step, says at Imperial College London. “I think it’s the right time, with the right data and with the right expertise, to possibly solve this problem.” The latest AIs learn linguistic patterns from huge amounts of human-supplied language data, without any clue how our languages actually work. Essentially, they create a vast, multidimensional “cloud” of words clustered according to how we use them, and this lets them decode new snippets of text.
In 2018, researchers at Facebook realised that if you twist the clouds for two languages in just the right way, you can get words with the same meaning to line up, . This was a crucial breakthrough, says Bronstein: it suggested we might be able to decipher languages with no pre-existing translations.
Bronstein’s pet project is sperm whales. Their large brains, complex clan-based social structures and intricate communication system, comprising sequences of clicks known as codas, make them a promising target for interspecies communication. Bronstein leads the machine-learning team at , also known as the Cetacean Translation Initiative, an international collaboration of researchers attempting to decode sperm whale chatter. In a paper published in Scientific Reports last year, he and his colleagues analysed around 26,000 recordings to create models that reliably segregated codas into human-defined categories based on the number, rhythm and tempo of clicks, which whale was speaking and its clan.
This is still a long way from deciphering meaning. For that, the project is embarking on what Bronstein calls an “industrial-scale” effort to capture between 400 million and 4 billion clicks a year, deploying underwater robots and buoys loaded with acoustic sensors off the coast of Dominica in the Caribbean. Simultaneously, sensor-laden tags will help identify individual whales, piece together who is talking to who and reconstruct behaviours associated with certain patterns of clicks.
“AI is good at human language. Animal language is a logical next step”
Interspecies internet
Meanwhile, the plans to apply similar techniques to primates and birds like crows and ravens. Even more ambitiously, an unlikely coalition of animal psychologist Diana Reiss, computer scientist Neil Gershenfeld, rock star Peter Gabriel and internet pioneer Vint Cerf envisages an through which animals communicate both with each other and us via interfaces such as animal-friendly video chats or underwater touchscreens.
That might hint at a return to studies along the lines of Lilly’s flooded-house experiment with dolphins. But AI isn’t a silver bullet, warns at the German Primate Center in Göttingen, who studies guinea baboon communication. It is great at detecting patterns and can neatly sort whale calls, say, into piles based on their acoustic properties, but often can’t tell you what those piles relate to. “It’s not a magic wand that gives you an answer to the biological questions or questions of meaning,” says Fischer. For that, you need to correlate calls with observations of behaviour – an incredibly challenging task when studying deep-ocean animals like whales, even now that we have high-tech robots and sensors.
AI can easily be led astray as well, says at Queen Mary University of London, who is using it to build models of birdsong to help study bird development and evolution. Without guidance, it might pick up acoustic properties irrelevant to the animals – a variation in pitch, for instance, when birds might actually be listening out for rhythm or some other feature.
The answer may be to enlist the birds themselves to help. Stowell and his colleagues train zebra finches to hop around to indicate which snippets of song they regard as most similar, then feed this extra information into an AI. “This is really a stepping stone, because we need to know which sound differences are salient to the animals,” says Stowell.
Bronstein hopes to get round this same problem by building a sperm whale chatbot that can feed learned patterns of codas back to the animals to see how they react. He says it is uncertain whether we can ever go beyond superficial interaction, particularly with species like whales whose lives are so different from ours. “It might be that it will only be a rough approximation of the true depth and meaning of what they’re saying.”
Yet while we may never discuss the weather with a whale, even small advances in eliciting meaning from animal language could be revolutionary, says , an expert on prairie dogs. “Even if they have alien concepts, if we could make at least some of our desires known to them and they could make some of their desires known to us, I think that we would have a completely different world.”
Slobodchikoff has founded a start-up called that aims to develop an AI to enable people to communicate with their pet dogs, analysing vocalisations, facial expressions and actions to help diagnose behavioural problems, among other things. Meanwhile, scientists at Georgia Tech in Atlanta have created a system that , while researchers at the University of Cambridge have built an algorithm that
If livestock could directly communicate their concerns to us, that might completely reshape animal agriculture, if not potentially render it morally untenable. Interspecies communication in general could force us to re-evaluate assumptions about our supposedly exceptional place in nature.
“Many of the problems we face now, such as the climate crisis or pandemics, follow from seeing ourselves as separated from the other animals,” says , a philosopher who studies interspecies communication. “Learning to listen to them, and viewing ourselves as part of a bigger whole, is a practice that can challenge this and open up new pathways to change.”
It is a grand vision that seems worth signing up to – especially if it has the bonus of getting the dog to stop barking at the vacuum cleaner.
