I love the Royal Institution Christmas Lectures.  Since 1825, they have been designed to introduce a general audience, most recently mainly kids, to some potentially pretty heavy-weight scientific topics––Sue Black and the Secrets of Forensic Science; Hannah Fry and the Hidden Powers of Maths; Alice Roberts asking What Makes Me Human?  They are pitched at a perfect level for the lay-person to tune in, not feel too stupid, and actually learn something new and interesting.  This year, the topic was The Truth about AI, and the lecturer was Michael Wooldridge, Professor of Computer Science at the University of Oxford.

Avid readers of the Mudskipper blog will have realised that I have something of an interest in AI.  Albeit, a rather critical interest.  So, the theme of this year’s Christmas Lectures was right up my neural network.

Mike’s easy, informative approach engaged me from the start, but it was a guest speaker, Tom Mustill, a director and producer of nature documentaries, who really made me sit up and take notice.

The topic, which caught my attention was about Google Translate, and how it went about translating from one language to another.  Never previously having given the subject much thought, I just presumed that Google had access to a vast database of online language dictionaries and, laboriously, translated word-for-word, whatever text was presented to it, from one language to another.  Not a bit of it.  Using machine learning, Google Translate compares a vast number of whole sentences from different languages and then ‘shapes’ how different words in these sentences relate to one another.  This geometric shaping identifies similar patterns across languages, and translates between languages on this basis.  This method opens up the possibility of potentially understanding hitherto incomprehensible languages; even animal languages.

We are suddenly in Dr Dolittle territory.  For me, it was amazing enough to be able to translate an English sentence into its German equivalent; or Arabic; or Japanese; but what if I could now find a translation into Dog; or Cat; or Whale; or Worm?  It was a use for AI that I had never previously imagined; an extension of a current technology into a realm, which seemed previously unobtainable. 

To talk to the animals?  Wouldn’t that be amazing?  As Rex Harrison sung: imagine talking to a tiger; chatting to a cheetah.  It was the kind of thing that my 7-year old self might have dreamed of; but my adult self dismissed as impossible.  And now within the realms of scientific possibility?

Much of the remainder of Mike’s lecture passed me by in something of a blur as I was still reeling from the possibilities of being able to talk to my cat.

But then, at some point, my habitual prejudice against AI reasserted itself.  Talk to the animals?  Well, really, would that be such a good thing?

Take my cat.  I know my cat.  It is a miserable, whiny bastard.  While, at the moment, I might pretend to myself that its unintelligible mewing is a sign of undying affection, if I could actually understand what it was saying, what would I really hear?  Replace the word ‘lager’ in Underworld’s song Born Slippy with the words ‘cat food’ and I think that it might be quite a close approximation: “Cat food, cat food, cat food, cat food, cat food, cat food.”  Or, more likely, since my cat is something of a connoisseur, the constant, critical complaint of “What’s this crap cat food you keep serving me?  Where’s the fresh salmon, salmon, salmon, salmon, salmon, salmon?”

Dogs’ barks; birds’ tweets; cows’ moos; sheep’s baas: if we could really understand the animals, we would be surrounded by an endless, barracking tirade of complaint and abuse on all sides of us.  It would be exhausting, demoralising; engulfing.

Sometimes, ignorance truly can be bliss.

© Simon Turner-Tree

Simon Turner-Tree is ignorant but happy.  Although not so ignorant not to ™ the word ‘Dolittle’ for some future online translator.