Deep Learning Machine Does Similes Like a Poet

For the first time, a machine has outperformed the average human's ability to answer verbal reasoning questions.


Until now, putting a machine up against a human in a game of analogies was like putting someone who was really bad at tennis up against some good tennis racket swinging guy — like, champion level. We can now kiss our figurative supremacy behind thanks to a program devised Chinese researchers that just outperformed humans in verbal reasoning for the first time.

With this latest news my profession inches closer to obsolescence, like a slow moving animal trapped in some, like, sticky pit that makes it move slower. Anyway, verbal reasoning is part of how we measure IQ, and the creative leaps involved have typically been the domain of humans. But using data mining to analyze massive library’s worth of text as the relationship between words and phrases has given an edge to this new machine from researchers Huazheng Wang of China’s University of Science and Technology and Bin Gao at Microsoft Research.

Here’s one of the analogy examples used in the test:

“Identify two words (one from each set of brackets) that form a connection (analogy) when paired with the words in capitals: CHAPTER (book, verse, read), ACT (stage, audience, play).”

Unfortunately, the exact answers supplied by China’s machine haven’t been reported yet, so you’ll have to guess how yours compares. Make it a game with your friend, maybe. Just don’t go on and on about it because no one likes someone taking a halfway clever idea then punching it repeatedly like it’s a horse that used to be alive but now is not.

Related Tags