Lecture: Grand Challenges in AI: Unfinished Agenda
Raj Reddy
Abstract:
In this talk I will discuss some Grand Challenges of AI, successes of the past 30 years, and the unfinished agenda for the 21st century. In 1988, I presented a list of unsolved open Grand Challenge Problems in AI, as part of the Presidential Address of American Association for AI (https://www.aaai.org/ojs/index.php/aimagazine/article/view/950). Since then some of the problems have been solved. The World Champion Chess Machine challenge was settled in 1996 when IBM Deep Blue (developed by Hsu, Anantharaman, Campbell, Hoane et al) defeated the then reigning World Champion of Chess, Boris Kasparov (https://en.wikipedia.org/wiki/Deep_Blue_(chess_computer). The Accident Avoiding Car (Driverless Car) challenge was decided in 2005 when Stanford’s Stanley headed by Sebastian Thrun and CMU’s Sandstorm headed by Red Whittaker were among the five cars to successfully complete the challenge (https://en.wikipedia.org/wiki/DARPA_Grand_Challenge_(2005). The Challenge to demonstrate Understanding of Science Textbooks by taking an Exam was successfully demonstrated recently in 2019. https://www.wired.com/2016/02/the-best-ai-still-flunks-8th-grade-science/ and http://bit.ly/aristo90. Starting from a failing performance in 2016, Peter Clark, Oren Etzioni and team at Allen Institute for AI developed a system that answered over 90% of the questions correctly in the NY Regents Science Exams.
The Unfinished Agenda for 21st Century includes: Discovery of a Major Mathematical Result by AI; Summarization of Media (Books, Talks, Music and Movies); Remote Repair in Space; Encyclopedia on Demand; Provide the Right Information to the Right People at the Right Time in the Right Language; Self-Reproducing Robots; Any Language to Any Language Translation among the top 100 languages with less than 5% error; and Any Spoken Language to Any Spoken Language (Speech To Speech) Translation among the top 100 languages with less than 5% error. Each of the above seemingly realistic problems would require significant breakthroughs and fundamental advances in AI and all other subfields of Computer Science and Technology.
Download link to the slides of Reddy's lecture
