Word games show the limits of AI



It’s one thing to play chess against a computer – you’ll lose – but it’s another to play a collaborative word game. It pushes the boundaries of today’s AI.

What is happening: The game geeks are trying to create robots that can play Codenames, the hugely popular word guessing game.

  • It is played with two teams and a 25 word board, as in the photo above.
  • On each turn, one person tries to find a one-word clue connecting as many of the 25 words as possible; then that person’s teammates try to guess as many words as possible.

Give a good hint is quite easy for computers, using basic open source machine learning tools for understanding languages.

  • “I was surprised at how well it worked,” said David Kirkby, astrophysicist at UC Irvine who programmed a Codenames bot for fun. “He provided clues that weren’t obvious to me – but they made sense.”
  • Kirkby’s bot once gave the clue “Wrestlemania” to connect the words “undertaker” and “match”.
  • Another bot coded by Jeremy Neiman, engineer at Alphabet’s Sidewalk Labs, used “telemedicine” to connect “the ambulance,” “the hospital,” “the link,” and “the web.”

What is really hard is to guess if your teammates will understand your clue. This is one area humans are great at – whether you’re playing with a sibling or a friend, you can draw on shared experiences to find the perfect word.

  • The next big step is to create robots that develop an understanding of their teammates over the course of multiple games, says Adam Summerville, professor at CalPoly Pomona who runs AI codename competitions.
  • Achieving this goal is essential for creating robots that communicate better with people to accomplish a shared task.



Leave A Reply