Jerry May, a Pitt business professor, fidgets in a crowded computer laboratory in Scaife Hall. He’s waiting for class to begin, but he’s not the teacher—he’s a student. He’s here, on his own time, to learn about a programming language used in artificial intelligence (AI).The goal of AI is to build computer systems with the reasoning capabilities of humans.
While May’s professional home is the Joseph M. Katz Graduate School of Business, it’s not so surprising that he’s spending extra time around campus learning about AI. He’s a Yale-trained researcher with a doctoral degree in operations research, which involves the use of mathematical models to analyze the nature of any process—an inventory system, a traffic pattern, an electrical circuit. This former rabbinical student, with bachelor’s degrees in mathematics and economics, knows he’ll be able to adapt what he learns to his own research, just as he did as an undergraduate at Chicago’s Roosevelt University and as a graduate student at Yale.
On the first night of class, May is surrounded by about 70 others—mostly graduate students. No one gets official credits for these informal gatherings; the reward is learning AI programming concepts. A week later, about 30 people show up. Within a month, May is among a handful of people with the tenacity to stick with these intense programming sessions on top of teaching classes, grading papers, mentoring students, conducting research, writing publications, and finding time for family life.
What he got from these sessions more than 20 years ago was an appreciation for evolving technology—how next-generation knowledge, applied creatively, produces new concepts and even new products. Today, May is a professor of decision sciences and intelligent systems in the Katz School, where he’s still looking for interesting ways to use technical expertise in solving business problems.
Recently, for instance, May and two of his graduate students began exploring an emerging phenomenon: the death of television as we know it. What they found has the potential to shape an entirely new marketplace.
With the advent of the digital personal video recorder (PVR) and related services like TiVo, the nature of television viewing has begun to change dramatically. Traditional TV viewing, with its sit-down-on-a-schedule programming, is morphing into something else, something not imagined when commercial TV was born in the 1940s.
PVR technology—from TiVo or your cable company—uses sophisticated software and an internal hard drive that give today’s TV-watchers a vast assortment of viewing options. While most of us still assemble on a particular night at a particular time to see the latest episode of, say, Commander in Chief, The Daily Show, or Larry King Live, that habit may soon be part of television history.
With any selected broadcast, services like TiVo instantly record the incoming broadcast signal. This makes it possible for people to pause their viewing during a live broadcast, make a sandwich in the kitchen, and immediately retrieve the just-missed content. Or they can delay tuning into a scheduled program by a few minutes and then begin retrieving the digitally stored content. In either case, viewers can totally avoid advertising by fast-forwarding over the commercials.
The computer-like nature of PVR devices means that viewers can also select programming based, for instance, on a favorite actor or type of show. Want to watch only comedies, or only comedies with Dave Chappelle? No problem. The shows will be waiting for you when you’re ready to watch.
So, these smart devices can be used by cable companies or other service providers to deliver viewer-requested content or to offer suggestions about what the viewer might like to watch based upon stated preferences or previous viewing habits. This is possible because providers gain information about viewing habits and preferences, much in the same way Amazon.com can offer book and music suggestions based on buyers’ browsing and purchasing patterns.
The National Association of Broadcasters estimates that, by 2010, about 55 million homes will have PVRs, up from 7 million in 2004. And, this year, two major telecommunications companies, SBC and Verizon, announced plans to compete with cable companies using Internet-based television content. Internet protocol television, or IPTV, will integrate the capabilities of the Web with television, video, music, and general computer content. There’s no doubt that the digital environment will continue to flourish, as will viewers’ ability to control and customize what they watch. PVRs and set-top boxes (the digital control box supplied by service providers) will increasingly rule digital broadcast content in the era of digital everything.
“The change in technology—with PVRs and set-top boxes—is changing the advertising model for TV,” says May. With television moving into an on-demand viewing culture, it’s now the viewer—not the network or cable company—who determines what programs to watch and when to watch them. The conventional advertising model of blasting all viewers with the same commercials throughout a TV show is becoming less effective, maybe even obsolete.
May and two of his former students—Mordechai Gal-Or (KGSB ’96) and William Spangler (KGSB ’95)—have jumped on this emerging trend. They’ve developed a tool to identify those viewers who are most likely to be interested in particular products. Their method is based on some fundamental AI concepts, such as machine learning (the development of techniques that help machines “learn”) and data mining (searching for hidden patterns and relationships in large amounts of data). The trio calls their concept “addressable advertising.”
Sending traditional commercials into every home, says May, is like blanketing a neighborhood with advertising flyers. Everyone in the neighborhood gets the same message, whether they’re interested or not. Addressable advertising theoretically makes it possible for an advertiser to send commercials primarily to those viewers who are most likely to want that advertiser’s product. “We think the advertising model is moving to one where, instead of trying to buy time on a particular show, there is the ability to buy a slot on a person’s set-top box,” says May. A particular person’s box.
Previous research by others has shown that TV viewers tend to fall into categories based on factors such as gender, age, and income level. “People in a particular segment will watch a particular mix of shows in a particular way,” says May. “The idea is that you can pick up patterns from that.”
May, Gal-Or, and Spangler obtained data about TV viewing habits from Nielsen Media Research, the source of TV’s Nielsen ratings. (The data was collected with viewers’ permission.) Then, the trio applied data mining tools, plus May’s years of experience in finding seams of raw meaning within mountains of complex data. From this, they created a method to determine a viewer’s gender, age-range, income level, and purchasing tastes based upon that person’s TV viewing preferences.
Apart from some prickly privacy issues, the trio’s method makes it possible for advertisers to target specific viewing households with messages tailored to that household’s preferences.
Some people, for instance, enjoy getting certain catalogs in the mail or welcome recommendations from Amazon.com. So, too, some viewers will elect to watch or at least tolerate TV advertising that’s in tune with their interests and purchasing preferences. Just as television viewing is becoming more personalized and customized in an on-demand environment, so too will advertising. The trio’s new tool is poised to help make that happen.
May and his two former students—who are both now professors in the business school at Pittsburgh’s Duquesne University—are pleased with the odds, which vary depending on the target segment. For instance, their model can identify women viewers, age 18-34, with 58 percent accuracy; random selection has only a 25 percent probability of success.
In other words, using the model, advertisers are at least twice as likely to hit the intended audience with their ads. May says their models don’t have to be perfect. “You don’t have to outrun the tiger,” he says. “You just have to outrun anyone who is with you when the tiger is chasing you.” In a market that generates billions of dollars in revenue annually, a two-to-one improvment in reaching the right audience translates into big money. Meanwhile, May remains intrigued by the tools of artificial intelligence, just as he was when he stuck with the night class on AI programming years ago. He continues to seek out graduate students who have business problems to solve, and he’s fervent in giving credit to Spangler and Gal-Or for the years of experience and creative ideas they brought with them to the joint project.
“Work that derives from real-world problems adds credibility in the business school world,” he says.
It’s clear that he enjoys working with students and colleagues on a variety of projects. “Usually,” he says, “somebody comes to me with a load of data and thinks there ought to be something interesting in it.”
Usually his response is: “Let’s find out.”
Cindy Gill is a senior editor of Pitt Magazine.
Links to external Web sites are offered for informational purposes only and the information there is not guaranteed or endorsed by the University of Pittsburgh or its affiliates.