Dungeons and Dragons, not chess and Go: why AI needs roleplay
aeon.co
aeon.co
No. The author is trying to be funny, but the Turing test already encompasses this. The rest of the article is meandering.
Edit: And the whole intro of "These computer scientists play chess and go, so they wrote bots to do it. They should instead write a bot that plays the game I like instead, then it would be 'real' AI". Come on. The hypocrisy is palpable.
You see similar problems with self-driving cars. They may know the rules, the physics and the street signs, but they apparently still have trouble deciding whether a shadow is an object or not, whether a plastic bag blowing across the street is something you can ignore or not, that a ball rolling onto the street can mean that a child will soon run after it, etc.
We can think outside the box, AI can't. Not yet, at least.
I'd argue that if you want to solve the ability to tell a story, D&D is a terrible stepping stone. The arguments in this article are weak (You need to play in person, and physically roll the dice? ReallY?). No serious scientist thinks that's necessary.
But D&D doesn't have to be at the bottom of the list of things to focus on to pass the Turing test. In fact, I don't consider passing the Turing test all that interesting as a goal. But being able to deal with complex, open-ended, social situations, would be an interesting goal for AI.
AI does very well in limited situations, when there's a clear box to think inside of. But there are many situations, including self-driving cars, or playing a game like D&D, where there's no real limit to the things that are relevant to the activity.
Computer Arimaa and Go have since surpassed top humans, but I suppose D&D could be on the same level as Calvinball.
You can preprogram a chat bot to pretend to be a stoic monk who says very little, and could probably convince some folks that it was a real human playing the game as long as it took it's turn as expected, and rolled dice when asked.