I thought today about when AI might be better than humans at understanding us. We’re a multilingual family and we listen to music from around the world. My kids might demand a French song followed by a Hebrew song, followed by Baby Shark (of course).
Soon, the scenario might exist that AI is better at understanding us than a human could be.
“OK Google, play ‘Where Did I put That Thing?’ from Spotify?” (I’ve learned that you don’t ask Google to play the real name for this song… just like you don’t tug on Superman’s cape or mess with Jim).
Bam… bibbidi bobbidi boo.
OK, now the kids want ציגלה ביגלה — Tsigale Bigale. Oh, you haven’t heard of it? Why it’s the 1984 hit by Israeli artist Yigal Bashan:
The yardstick for AI is typically how well could a human perform at the task. Speech transcription? Sentiment analysis? Emotion detection? Intent recognition?
TURN ON THE LIGHT
What action did the person want to perform? On or off? What did the person what do turn on? Very good. You’ve identified an intent and two entities (or slots, depending on who you ask).
If I were to ask someone “play Tsigale Bigale from Spotify”, what would be there response?
Seagull beagle? Sig All Lay Big Ole? Even getting the transcription correct, they’d Google it in English and see this:
So, what would someone really considerate do to interpret the results?
Hmm… I know he wants to listen to something from Spotify. It Sig or tsig something… If he asks to listen to it, he’s probably listened to it before. Let’s start looking through.
Oh, there it is! The song he’s listened to a million times!
OK, playing Tsigaleh Bigaleh.
So, by knowing a bit more about my song listening history and that I’m not asking for something in English, the AI could perform better than a human in this task. Oh well, no big deal, Eh Beh Tsigaleh (Yiddish for the “Goat says meh”).
And what’s the connection between these songs? “Bibbidi Bobbidi Boo” reminds them of “Tsigaleh Bigaleh Boom”.