Alan Connor 

Devious humour and painful puns: will the cryptic crossword remain the last thing AI can’t conquer?

When human solvers battle artificial intelligence, who is able to think more cryptically, faster? And are some devious clues just too tough for software?
  
  

Many crosswords, post-solve
Yay, solved … the fruits of cryptic human toil. Photograph: Alan Connor

The Times hosts an annual crossword-solving competition and it remains, until such time as the Guardian has its own version, the gold standard.

This year’s competitors included a dog. Rather, an AI represented as a jolly coffee-drinking dog named Ross (a name hidden in “crossword”), and who is embedded on the Crossword Genius smartphone app.

The human competitors at the event – which took place at Times’ parent company News UK’s London headquarters, in the shadow of the Shard – were, as usual, bafflingly fast: pondering the next clue while scribbling the letters of the previous. An AI can conceivably “think” about multiple puzzles at once: so did it outwit us mortals?

For now, we humans maintain our edge. Ross “gave up” on an incomplete grid after serial champion of the contest Mark Goodliffe had raised a hand to tell the invigilators that battle was over.

This was not a done deal. I’m sure Ross had decoded one across …

1ac MP ousted by Liberal, absolutely without authority (9)

… replacing the MP of IMPLICITLY (synonym for the clue’s “absolutely”) with an L for the solution, ILLICITLY (“without authority”) while the flesh contenders were still considering that the answer might be an adjective or an adverb (or some MP). The next answer being the Finnish for Finland is something you or I might or might not know; Ross pretty much “knows” everything.

Here’s one, though, that stumped Ross:

13d Radical overhaul of motorsport’s image (9)

A radical may also be described as a FIREBRAND – or, as the setter has it, an F1 RE-BRAND. This clue isn’t quite like its gridmates. It’s closer to a joke. When you see it, you feel that it is correct, partly because of your smile: it does not play to an AI’s forte – namely, asking: have I seen something like this before?

And this was a clue from the steady-as-she-goes Times. Pity the AI confronted with this paper’s Paul when in punny mode, like “Picnicker, by the sound of it?” for ART THIEF.

For now at least, that sense of a setter – one of our fellow humans – conveying to you “yes, I went there” is something you need yourself to be a human to understand.

A suggestion: instead of identifying fire hydrants and motorcyclists, online security should involve cracking cryptic clues, ideally those with fanciful puns. The Guardian’s setters are available.

(Full disclosure: I test-solved some of the puzzles an earlier iteration of Ross was trained on. I grew fond of Ross and sometimes use “him” to see whether a clue might equally well offer two alternative entries.)

And in our cluing conference: many thanks for your clues for STOKES. The runners-up are PeterMooreFuller’s last-letters “Runs! Most to Brook, three tons in the end for the England captain” and Wellywearer2’s cheeky “Builds up the tension? It’s what Bram does!”; the winner is the ingeniously plausible “Loads Tinder, fingers swiping right”.

Kludos to Dunnart. Please leave entries below for our next challenge: how would you clue PUNNY?

 

Leave a Comment

Required fields are marked *

*

*