Solving mathematical puzzles: A challenging competition for AI

Abstract

Since its origin, the holy grail of artificial intelligence has been to understand the nature of intelligence and to engineer systems that exhibit such intelligence through vision, language, emotion, motion, and reasoning. In such context, AI researchers have always looked for challenges to push forward the limit of what computers can do autonomously and to measure the level of "intelligence" achieved. Competitions have been and are currently run on conversational behavior (for example, the Loebner prize1), automatic control (for example, the International Aerial Robotics Competition2 or the DARPA Grand Challenge on driverless cars3), cooperation and coordination in robotics (for example, the RoboCup4), logic reasoning and knowledge (for example, the CADE ATP System competition for theorem provers5), and natural language (for example, the EVALITA competition6 for the Italian language). Historically, also games have raised the interest of the AI community: A number of competitions are still being held nowadays (for example, the World Computer Chess Championship, 7 Mario Championship8 [Togelius et al. 2013] and its successor Platformer Competition,9 the General Game Playing competition10). For a high-level overview of the field of AI in games see the paper by Yannakakis and Togelius (2014).