AI Grand Challenges: Story and Creativity
These are the AI challenges that, if solved, would demonstrate the power of intelligent systems that possess creativity and/or the ability to reason about and manipulate stories.
- The idea of enumerating AI Grand Challenges that involve storytelling or creativity started when my post-doc Steve Lee-Urban pointed me to a book titled "Mysteries of Harris Burdick". The book is a set of illustrations that are scenes from a coherent story, but the story is never told. The reader is invited to imagine what that story could be. This is a living document.
- Could a computational system look at a set of pictures and discover the objects and relations? Could it read the captions that go with the pictures? Could it then "fill the gaps" between these scenes to create a coherent story?
- The following is a story about an AI agent that lost to humans at crossword puzzles. Why would a computer lose if it could crunch through words really fast? Humans must be doing something cognitive that enables them to zero in on the best words fast.
- Santi Ontanon points out that AI systems have surpassed humans at "classic" problems such as chess (and checkers), and recently beat human champions as Jeopardy. AI systems are beginning to become competitive at human-levels at the game of Go. Starcraft is less of a certainty, but Santi optimistically wonders what might be next for computers to beat humans at.
- Having recently speculated on the challenge of beating Colin Mochrie at "Whose Line is it Anyway?", I propose Improv comedy sports as the next frontier of game-playing agents.
- Of course this is somewhat tongue-in-cheek since improv comedy sports is not a real sport. Case and point, on "Whose Line is it Anyway?" the scoring doesn't matter. To me that makes the problem all the more compelling, because the actual metric for whether one is doing well at the "game" is whether the audience is being entertained. Unfortunately, that is not easy to measure.
- AI Magazine's article on General Artificial Intelligence focuses in part on ways of assessing human-level AI. I pulled out the two that I thought were most interesting: general video game learning and story comprehension.
- The next challenge has been around for a while, but I thought I would bring it up again.
- Aerith, a character from the Final Fantasy series, is famously cited as achieving this. However, Aerith's death scene was a scripted cut-scene and therefore doesn't count. The criteria for this challenge are: (a) the character must autonomously make decisions with the explicit intention of bringing about an emotional response in human players, and (b) the emotion cannot be fear, which is deemed too easy to achieve in an intentional manner.
- In all actuality, the challenge to emotionally move players is a story generation challenge, as the means to create emotion is often more about what happens to a character--thus the story--than what the character decides to do at any given moment.
- The next if for roboticist:
- If we can build machines that demonstrate the agility, coordination, and timing of Cirque du Soleil performers, I would believe anything could be accomplished. I think this makes autonomous driving seem easy. The first step might be to have a robot balance on a bongo board: bit.ly/T5Zmfr. Yes, this extends to clowns as well: if machines can perform comedic routines, then we may also have solved many of the challenges associated with Human-Robot Interaction (HRI).
Did you find this story interesting? Be the first to like or comment.