AI has a long way to go

Twitter just recommended for me a tweet that it labeled as being on the topic “filmmaking.”

It was a tweet by Ana Mardoll about a doctor behaving badly. It had nothing to do with filmmaking.

But it did include the phrase “right before he wrote the script.”

Dear Twitter: Words have multiple meanings. In this case, the word script is short for prescription.

More generally, recommended-for-you systems continue to be terrible at guessing what might be of interest to a reader, and part of the reason for that is that those systems aren’t always even correctly analyzing the topics of the recommended posts.

In this case, I was interested in the tweet in question. But not for the reasons that Twitter’s algorithm thought.

…Edited to add an even better example, an hour later:

Twitter showed me a post from someone I’ve never heard of. Its explanation/label for why it was showing it to me? “Astrology.”

I have a strong antipathy for astrology. I have less than no interest in being shown astrology tweets.

But the reason this particular tweet is such a great example is that it has nothing to do with astrology.

But it uses the phrase “three-star Michelin restaurant.”

So Twitter’s inept AI saw the word “star” in the tweet, decided it was a tweet about astrology, and showed me the tweet because it thought I was interested in astrology.

At this point I’m almost inclined to think that Twitter’s AI is intentionally trolling me.

One Response to “AI has a long way to go”

  1. Jeremy

    AI may have a long way to go, but then again, so does natural intelligence.


Join the Conversation