Artificial intelligence is helping us talk to animals (yes, really) –

0 5

Marc Aspinall

Each time any of us uses a tool, such as Gmail, where there’s a powerful agent to help correct our spellings, and suggest sentence endings, there’s an AI machine in the background, steadily getting better and better at understanding language. Sentence structures are parsed, word choices understood, idioms recognised.

That exact capability could, in 2020, grant the ability to speak with other large animals. Really. Maybe even faster than brain-computer interfaces will take the stage.

Our AI-enhanced abilities to decode languages have reached a point where they could start to parse languages not spoken by anyone alive. Recently, researchers from MIT and Google applied these abilities to ancient scripts – Linear B and Ugaritic (a precursor of Hebrew) – with reasonable success (no luck so far with the older, and as-yet undeciphered Linear A).

First, word-to-word relations for a specific language are mapped, using vast databases of text. The system searches texts to see how often each word appears next to every other word. This pattern of appearances is a unique signature that defines the word in a multidimensional parameter space. Researchers estimate that languages – all languages – can be best described as having 600 independent dimensions of relationships, where each word-word relationship can be seen as a vector in this space. This vector acts as a powerful constraint on how the word can appear in any translation the machine comes up with.

These vectors obey some simple rules. For example: king – man + woman = queen. Any sentence can be described as a set of vectors that in turn form a trajectory through the word space.

These relationships persist even when a language has multiple words for related concepts: the famed near-100 words Inuits have for snow will all be in similar dimensional spaces – each time someone talks about snow, it will always be in a similar linguistic context.

Take a leap. Imagine that whale songs are communicating in a word-like structure. Then, what if the relationships that whales have for their ideas have dimensional relationships similar to those we see in human languages?

That means we should be able to map key elements of whale songs to dimensional spaces, and thus to comprehend what whales are talking about and perhaps to talk to and hear back from them. Remember: some whales have brain volumes three times larger than adult humans, larger cortical areas, and lower – but comparable – neuron counts. African elephants have three times as many neurons as humans, but in very different distributions than are seen in our own brains. It seems reasonable to assume that the other large mammals on earth, at the very least, have thinking and communicating and learning attributes we can connect with.

What are the key elements of whale songs and of elephant sounds? Phonemes? Blocks of repeated sounds? Tones? Nobody knows, yet, but at least the journey has begun. Projects such as the Earth Species Project aim to put the tools of our time – particularly artificial intelligence, and all that we have learned in using computers to understand our own languages – to the awesome task of hearing what animals have to say to each other, and to us.

There is something deeply comforting to think that AI language tools could do something so beautiful, going beyond completing our emails and putting ads in front of us, to knitting together all thinking species. That, we perhaps can all agree, is a better – and perhaps nearer-term – ideal to reach than brain-computer communications. The beauty of communicating with them will then be joined to the market ideal of talking to our pet dogs. (Cats may remain beyond reach.)

Mary Lou Jepsen is the founder and CEO of Openwater. John Ryan, her husband, is a former partner at Monitor Group

More great stories from WIRED

🐘 The illegal trade of Siberian mammoth tusks revealed

🙈 I ditched Google for DuckDuckGo. Here’s why you should too

📧 How to use psychology to get people to answer your emails

💸 The WIRED Recommends guide to the best Black Friday deals

Leave A Reply

Your email address will not be published.