Terry Prone: Badger-licking test proves AI is not as smart as it’s cracked up to be

'AI has no oil on its feathers, no capacity to laugh at you or itself'

Greg Jenner tested Google’s AI capacity by typing in a nonsense phrase he made up plus the word ‘meaning’, and it treated ‘you can’t lick a badger twice’ as if it were a wise old proverb or seanfhocal. File picture

The only badger I ever met was named Bill, a great friend of Rupert Bear, and you can figure that our encounter might have been a while back. He did leave a good impression, though, did Bill Badger, who was well brought up, reliable, and a total gent. So it came as a surprise to know that licking a badger is a thing. Indeed, licking a badger twice could be described as a thingX2.

The actual phrase — the sort of thing you’d expect to find in folklore because folklore is full of that kind of bilge — goes like this: “You can’t lick a badger twice.” 

This cute concept was dreamed up to establish that if you trick someone once, they are unlikely to fall for the same device a second time. (Now, it might be argued that recent American presidential elections disprove this, but that would be to distract.)

The badger-licking emerged into public discourse courtesy of a man named Greg Jenner, who noticed that: “You can type any random sentence into Google, then add ‘meaning’ afterwards, and you’ll get an AI explanation of a famous idiom or phrase you just made up.”

Mr Jenner, who describes himself as a “public historian”, hosts a podcast about history which matches comedians and academics. He didn’t get the badger statement from something like Peig. He made it up, as a kind of clinical trial.

Once he posted up his badger-licking content, it created a stir, appealing first to people who fear and dread artificial intelligence, but even more so to people like me who have too little in their day and so wanted to try this particular manifestation of AI for themselves in the name of research.

Now, the one commonality guaranteed among people who went through the Irish education system in the last century is that their heads are part-filled with seanfhocail — maxims from the great oral tradition which are supposed to have wider symbolic application.

Today’s students may not believe it, but the fact is that we were forced to learn this twaddle off by heart and encouraged to deploy it in essays and in oral Irish exams. So, when I decided to try out this AI for myself, up popped a phrase Bill Badger would have known, if he was brought up in Clontarf, rather than Nutwood: 

“Aithníonn ciaróg ciaróg eile.” Which I always understood to mean “one earwig knows another earwig,” which might be interpreted as: “It takes one to know one.”

AI told me I had been wrong, these many decades, about mutual recognition in the earwig community and that the phrase I had fed into the system actually referred to beetles. If I wanted to go with earwigs, the phrase to be learned off was: “A gailseog a bhfuil a fhios ag gailseog eile.”

But the thing about AI is that there’s no stopping it, as anybody drafting a document will confirm. AI always wants to correct, to reduce, to herd a writer into safe cliché territory. 

At this stage, most of us receive or send at least one message a day followed by an embarrassed apology for some idiotic transposition of meaning inflicted on the draft, unasked, by a linguistic robot. And you know what? The apology and correction are rarely needed, because those of us spending a lot of time on computers have developed an automatic capacity to correct the correction.

AI corrected me away from ‘ciaróg’ to ‘gailseog’. But it went further. OMG did it go further.

It told me, in a fairminded thoughtful tone: “While earwigs might not have sophisticated social structures like bees or termites, they do recognize each other and engage in interactions, including mating and competition. They may also emit scents, some detectable to humans and others that attract other ear-wigs.”

'It came as a surprise to know that licking a badger is a thing. Indeed, licking a badger twice could be described as a thingX2.' File picture: iStock

You’d nearly feel sorry for earwigs because of them lacking sophisticated social structures, but then you get to the bit that enlightens you to the fact that male earwigs have oversized genitalia. 

Bully for them, you think, until you are told by AI what the earwig lads do with their oversized earwighood, which is too disgusting for repetition in a decent newspaper. (Go look it up. AI is eager to serve you.)

At this point, you’re shuddering away from earwigs even more than you were at the outset — and, let’s face it, none of us knows a human being who actually likes earwigs. Killer sharks have a bigger following. But at least you know what AI will do with a genuine old phrase.

The next step is to try it with a new proverb, which reminds me of an encounter between my first boss, the late TV host Bunny Carr, and a participant on a communications course Bunny was running.

The participant, a bookish type, got high-minded about cliché-use on foot of another course participant who had articulated a cliché.

“Anybody could think that up,” the bookish type claimed.

“OK,” said Bunny equably. “Think up a cliché now.”

I had to sympathise with the poor man when, this week, I tried to think up such a phrase. Because of a tendency to pomposity, what I eventually came up with was: “The benefits of wild swimming are unevidenced.” Look, I was trying my best. I failed.

I put the sentence in, typed ‘meaning’ in afterwards, and AI basically told me to get stuffed. I was offered a myriad, a multiplicity of useful studies about the good wild swimming does for your brain, body, and spirituality.

I tried “restless republics eschew carpet”, and, while this wasn’t productive, it did provoke the system into sympathising with me for the deficit of results. Maybe, I thought, I need to keep it simple. 

So I typed in “crosswords rot your brain”, on the basis that any good robot would react to the proposition. AI got almost hysterical about the slur, saying it was definitely not true.

“In fact,” it said, “studies suggest that regularly doing crosswords, along with other word and number puzzles, can actually improve brain function and may even delay the onset of cognitive decline.”

Then it got quite huffy about my proposition, summing the situation up thus: “Engaging in activities like crossword puzzles can actually be beneficial for brain health, and the claim that they ‘rot’ your brain is not supported by scientific evidence.”

Which brought me, albeit by a circuitous route, back to the badgers, because the badger-licking statement is not an old saw. It’s a freshly made-up new saw. And AI doesn’t have enough intelligence, artificial or otherwise, to spot it for what it is. AI can’t go: “Ah, here.” It has no useful cynicism. It has no oil on its feathers, no capacity to laugh at you or itself. It takes everything seriously.

AI goes earnestly to work on topics like badger-licking and seeks to fit them into what AI already knows. Which is scarily like what us humans do every day.