Co-authors Carlos Guestrin and Eric Horvitz discuss "Information Cartography" (cacm.acm.org/magazines/2015/11/193323), a Contributed Article in the November 2015 CACM.
00:00 In a wide-open world, you can go anywhere. But direct routes are better if you want to go somewhere.
00:15 Routes on the city metro join and branch off and join again, each time giving you the opportunity to follow a new path -- and discover new things. Wouldn't it be great if information research worked the same way?
00:30 Join us as we talk with researchers who aim to bring this model to the internet with Information Cartography.
00:44 [Intro graphics/music]
00:53 Eric Horvitz is known the world over for his expertise in artificial intelligence. But it was a poster hanging on an intern's wall that inspired this ACM Fellow to join her quest for a better way to present information.
01:10 DR. HORVITZ: So Dafna Shahaf was an intern with us, and I got to know her as a really broad thinker, someone who combines formal methods with large-scale, ill-defined problems, and... she was looking at these information maps that had been done by hand, of sort of how a field evolves, what's the history of machine intelligence. ... And she said, "I wonder if I could do this automatically. I wonder if I can create storylines automatically to explain hard, challenging development, developing areas to people in a clear way.
01:43 The result is Information Cartography, a model that brings together big data, optimization, and visualization.
01:51 DR. GUESTRIN: Information Cartography is about providing a guide, a map, to finding the information around big data problems, and bringing humans into the loop where massive amounts of information make it really hard for a human to navigate that space.
02:09 In one implementation, it starts with keywords, just like a web search. But there the similarities end.
02:18 DR. GUESTRIN: In Information Cartography you can imagine putting a keyword that's of interest and getting a map back. ... Another way to get started could be, "Oh, here's one article that I'm reading today, give me a map around this article." Another way to get started could be, "Here's two endpoints -- Greek debt crisis, and migration crisis -- Give me a path between them, or give me a map around these." ... From the science perspective you could say, "Here is the bibliography of the papers I've written so far. Build me a map around them. And by the way, tell me papers I should read or should have cited, that I haven't cited yet."
02:52 DR. HORVITZ: And we focused on the ACM Digital Library and looked at reinforcement learning as an example. And we show a visualization of that in our publication.
03:02 Most importantly, metromaps don't have to be static.
03:07 DR. GUESTRIN: What's really exciting for me about Dafna's vision is the ability to do this kind of an interactive way for any query that you might have. ... Take a starting grad student who wants to learn about machine learning, a field that I'm passionate about. That person probably needs to learn about the basic fundamentals of machine learning. ... Now, if you take a senior grad student, then they already know their foundation. What they really want to know about is: How is my Ph.D. thesis being impacted by new publications out there, and perhaps old ones that I don't know about? And ones in different fields?
03:44 Drs. Horvitz and Guestrin both believe that such an easy-to-navigate model goes beyond simple information management, and could benefit humanity in real and profound ways.
03:55 DR. HORVITZ: There's the prospect of using these tools to get into the minds of other people, to get other perspectives on news stories. ... And the idea is, can we come up with technologies -- can we harness computer science, can we harness optimization, information retrieval, and visualization -- to combat the prospect that we're getting only tiny keyhole views of a situation?
04:24 DR. GUESTRIN: And we've only scratched the surface of the possibilities here. If you can do this well, we can have a more informed world, faster scientific discoveries, you can have better political discourse. It's endless opportunity.
04:42 Find out more in this month's Communications of the ACM, in the contributed article, "Information Cartography".
04:50 [Outro and credits]