ARTIFICIAL INTELLIGENCE TOPICS

Reliable and relevant information about AI

AI in the News: The Top-Ranked Stories This Week These are a few of the stories that have been selected by the AAAI's NewsFinder program. It is still learning what our viewers think is interesting, so please provide your own ranking for any or all stories that appear here. For all recent stories see AI in the News. 1. <<Date>> <<Headline>> <<Source>> <<Brief Extract>> SLIDER BAR 2. <<Date>> <<Headline>> <<Source>> <<Brief Extract>> SLIDER BAR ... 5. <<Date>> <<Headline>> <<Source>> <<Brief Extract>> SLIDER BAR <<SLIDER BAR WITH EVERY STORY with points 0, +1, +2 marked. Mouseover shows text>> 0 = Not interesting to most viewers interested in AI +1 = Somewhat interesting to most viewers interested in AI +2 = Very interesting to most viewers interested in AI

Early work in AI, in areas such as story understanding and commonsense reasoning, tried to tackle the problem head on, but ultimately failed for three main reasons. First, methods for representing and reasoning with uncertain information were not well understood; second, systems could not be grounded in real experience, without first solving AI-complete problems of vision or language understanding; and third, there were no well-defined, meaningful tasks against which to measure progress. ...we are now at a time when we are well-poised to make serious progress on the goal of building systems that understand human experience. Each of the previous barriers is weakened ... [This problem] will be a driving challenge for work in AI in the years to come, and results from the work will profoundly impact our knowledge of how we live and interact with the world and with each other. Henry Kautz, "Understanding Human Experience". Position paper for the article, "Artificial Intelligence: The Next Twenty-Five Years", Matthew Stone and Haym Hirsh, editors, AI Magazine, 26(4): Winter 2005, 85–97.

Henry Kautz President of the Association for the Advancement of Artificial Intelligence

AITopics Editorial Board

• Harry Barrow
• Bruce Buchanan
• Jon Glick
• Alan Mackworth
• Deborah McGuinness
• Peter Norvig
• Jonathan Schaeffer
• Alan Schultz
• Aaron Sloman
• Reid Smith
• Ramasamy (Samy) Uthurusamy
• Bonnie Webber

Assistant Editor

• Colleen van Lent

Student Intern

• Liang Dong

For more information about this site, please see About Us.

We gratefully acknowledge past funding for this project from the NSF (Award #0738341).

AI Topics Donor
AAAI sponsors AI Topics as a free resource for students and others interested in learning about AI. An all-volunteer editorial board, together with the AAAI members, maintains the content, but there are costs associated with maintaining the code and adding new tools to the site. Please consider a contribution to continue this educational effort to explain the science of AI and issues surrounding it.

The NewsFinder Program

The NewsFinder program was developed to replace the labor-intensive process of finding and formatting current news stories of interest to readers of AI in the News. The program was initially prototyped by Tom Charytoniuk (Rice University) and developed by Liang Dong (Clemson University) under the supervision of Drs. Reid Smith and Bruce Buchanan.

The program first collects RSS feeds from Google News on the topics "artificial intelligence", "robots", "machine learning", "data mining", "computer vision". It discards blogs in order to have news stories from stable sources that appear elsewhere in print. The individual words and the number of occurrences of each word in each story are given to a support vector machine (SVM) that has been trained to score the likely degree of interest of a story on a three-point (0, +1, +2) scale.

Some phrases have also been added to the individual words in each story so that extra training data would not be required to learn them reliably. These include terms that are likely to indicate interest, such as "artificial intelligence", "intelligent machine", "autonomous vehicle", <<list here>>. Additional terms have been included in the first layer to indicate a story is not likely to be interesting to readers, such as "press release", "company's financials", "stock symbol" <<list others>>.

Additional terms associated with each story include the name of the source and metadata about the story (e.g., "press release"). The credibility and availability of sources are considered relevant to the degree of interest since readers are more likely to be expected to know about items that appear in the New York Times, USA Today, CNN News, or the Wall Street Journal than items appearing in trade publications or small, specialized magazines.

Initial training of the SVM was on approximately 100 stories that we scored manually. Feedback from readers provides additional training data, so the program should be adapting to the interests of the readers.

A very readable description of support vector machines can be found in the SVM documentation for php.