For most people, artificial intelligence means C-3PO, the Jetsons’ robotic maid, that Steven Spielberg movie or maybe IBM’s Watson, the computer system that beat the two most successful contestants on “Jeopardy!”
For Larry Birnbaum, an associate professor in the Department of Electrical Engineering and Computer Science who also works with the Knight News Innovation Lab, artificial intelligence often means something much less grand. “One of the things about A.I. is that it’s a very romantic field,” he says. “But when you look at the individual components, it’s a lot more technical work. Taken individually none of that seems very romantic, and yet the result is pretty impressive.”
Birnbaum and his colleague, Kris Hammond, run the Intelligent Information Laboratory, which focuses on decoding and replicating the way humans deal with information. That might mean creating a program that can perform the tasks of a human researcher, such as predicting what news articles might be interesting to an Internet user. Birnbaum works in the lab with colleagues in Medill and is developing a program that can identify all of the “pork” in a legislative bill.
The hurdles most challenging to an artificial intelligence programmer can seem trivial to a human brain. Understanding that recommended articles need to be similar to the original seems obvious, but understanding what key differences make a new article worth someone’s time is not. A low-salt version of a recipe might be useful, but a political article wouldn’t have a low-salt version; it would have conservative or a liberal version.
Patrick McNally, a fourth-year Ph.D. student in Birnbaum’s lab, sums up the disparity between what is trivial to a human and what is trivial for a computer as he’s explaining “Manatee,” a system he created to automatically generate webcomics. “This is a more ‘artificial intelligence-y’ project,” he says, as the program takes a noun and searches the Internet for something often held in opposition. The results are put into expressions like “Whatever absorbs your oil spills” or “I like thrills like I like flights – cheap” which are placed over a Google image of the noun. “Depending on your perspective it’s a little more hardcore or a little more frivolous.”
One of his newer projects merged the three panel comic style of “Manatee” with a more news-oriented bent, relying on Twitter data instead of user-requested words. Using Google, the system determines what people in the world are talking about, like sports events, political speeches or natural disasters, and overlays tweets about those subjects on a Google image that’s related. It’s not perfect — sometimes the Google image is of an NHL game while the tweet is about the NFL, but McNally says sometimes that makes people pay more attention to what they’re seeing. “I’ll often hear people walking down the hall and they’ll stop, you know, for 30 seconds and then they’ll continue walking down the hall and I’ll be like ‘Yeah! It worked.’”
His office is filled with monitors which flash streams of tweets involving the words ‘I love,’ from “I love Cheetos” to “I love boobs” to “I love my dad.” According to the volume of data he gets per day, he says people use the phrase ‘I love’ four times a second. “Twitter’s like a miracle to me. I can’t believe so many people are willing to pour out so much stuff every day.”
And in a field that sometimes requires huge amounts of data to be successful, Twitter can be a blessing. Shawn O’Banion, another Ph.D. student in Birnbaum’s lab, uses data from Twitter to create personal news recommendations. Most of the “recommended for you” stories on websites are generated by analyzing the things you’ve previously clicked on, but his project “Twitter Profiling” combs through your Twitter feed and matches articles to what it perceives as your interests. If you tweet about fall, for instance, it might suggest you try a new pumpkin recipe, because so many stories that mention the fall season also include the word pumpkin. “I don’t actually assign importance to the word ‘pumpkin,’” O’Banion explains.
“The system parses through these thousands of stories and it has an algorithm that calculates the importance of it.”
The large amount of data he receives from Twitter allows him to train his system and fine-tune its algorithms. “Using Twitter is a good source of information because even though people use it in different ways, it usually says something about your personal interests,” he says. “It’s not just this combination of your personal life, your daily activities but also a platform for sharing information.” If you tweet about a baseball game, the system will recognize that and suggest sports stories to you, but if you tweet a news story about a new Android tablet, it will recognize your interest in technology.
Though it seems hard to bare your soul in 140 characters, it turns out Twitter can be an incredibly revealing platform. “One thing I love about Twitter is it’s real people saying things,” McNally says. “So it’s wildly inappropriate most of the time, it’s often quite racist. It’s pretty real compared to other things.”