[youtube https://www.youtube.com/watch?v=E_6AZ8slivc?rel=0]
“There are more things in AI than classification… the entire paradigm of classification, which has fueled machine learning and data mining, is very limited… What we need is a process that is structured, multi-layered, knowledge-intensive, much more like kids playing Lego, instead of a karate chop that divides things into categories… Current knowledge bases are fact-rich but knowledge poor…’You can’t play 20 questions with Nature and win’ (Allen Newell, 1973)… What we need is knowledge, reasoning, and explanation.”
Slides (from KDD Keynote, scroll all the way down the page) are here
Oren Etzioni, executive director of the Allen Institute of Artificial Intelligence (formerly founder of Farecast and Decide.com), takes a contrarian view of all the deep learning hype. Essentially, he argues, while systems that are better than ever at classifying images or words are great, they’re still not “intelligent.” He describes work underway to build systems that can truly understand content, including one capable of passing fourth-grade short-answer exams.
“I think that poeple are often confusing computer autonomy with computer intelligence. computer viruses are autonomous, dangerous, but not particularly intelligent. Chess playing programs are intelligent (in a sense) but very limited. They don’t even play checkers!”
“I love star trek and particularly the star trek computer because AI is used there as a tool to help and inform Captain Kirk and the crew. That’s a much better model than fear mongering in movies like HER and Transcendence. AI can be used to help us and enhance our abilities. For example, we are all inundated with huge amounts of text, articles, technical papers, and nobody can keep up! How about if your doctor had a tool that would help him or her to figure out the latest studies and procedures relevant to your condition? Even better—what if you had a tool to help figure out what’s going on that’s much better [than] google or webmd.”
[Why do you enjoy working on AI? What first motivated you to get into the field?] “It is one of the most fundamental intellectual problems and it’s really, really hard. I find computers so rigid, so stupid that it’s infuriating. My goal is to fight “artificial stupidity” and to build AI programs that help scientists, doctors, and regular folks make sense of the world and the tsunami of information that we all face every day.”
The Turing test is about tricking someone to believe that a computer is human. At AI2 we are working on programs that will try to pass tests in science & math which requires the program to understand the questions (hard) utilize background knowledge (even harder) and accumulate that knowledge automatically (great big challenge).”
[Would the IBM computer that played Jeopardy be called intelligent by your metrics?] “Watson was an impressive demonstration but it was narrowly targeted at Jeopardy and exhibited very little semantic understanding. Now Watson has become an IBM brand for any knowledge based activity they do. The intelligence is largely in their PR department.”
