The mobile phone has long ceased being a simple two-way communication device: today’s handheld is a mini personal computer, complete with multimedia players, maps, and Web browsers. Now researchers at Palo Alto Research Center (PARC) want to push the phone farther. They have developed software that turns a phone into a thoughtful personal assistant, one that helps people find fun things to do. The software, called Magitti, uses a combination of cues–including the time of day, a person’s location, her past behaviors, and even her text messages–to infer her interests. It then shows a helpful list of suggestions, including concerts, movies, bookstores, and restaurants.

When a person first opens a phone that has Magitti software, she will instantly see a list of recommendations. If it’s noon, the software might suggest local restaurants. If it’s 3 P.M., it might recommend a nearby boutique for shopping. If it’s 9 P.M., a list of pubs might appear. Over time, these recommendations will change as Magitti learns more about the user’s behaviors and preferences. The software employs artificial-intelligence algorithms that have traditionally been used in research to make tailored recommendations. If, for instance, a person prefers to eat inexpensive lunches and more-expensive dinners, Magitti will pick up on this (by comparing the GPS location of the restaurant with a database of establishments) and offer up corresponding recommendations.

There are products available today that take advantage of GPS for friend finding, such as Loopt, and location-based search is available through Web portals Google, Microsoft, and Yahoo. But Magitti aims to do something different. “What’s unique is that we’ve tried to build awareness of different kinds of activities,” says Victoria Bellotti, senior researcher at PARC. “We want to find what kind of mode [the user is in]: if they’re hungry, if they’re interested in being entertained … And we’re trying to make this a relaxing interactive experience rather than being bothersome with alerts or requiring you to do searches.”

PARC’s software, which was developed for the Japanese company Dai Nippon Printing, is an example of a burgeoning trend to add more intelligence to handheld devices. And in many cases, this intelligence gives gadgets the ability to learn more about the person who operates them. As phones become more powerful, and more acquire sensors such as accelerometers and GPS, researchers are looking to artificial-intelligence algorithms to make sense of the data. Microsoft, Intel, Nokia, and universities such as MIT have groups that are exploring the applications for this type of software. (See “The iPhone’s Untapped Potential” and “Making Phones Polite.”)

Kurt Partridge, a researcher at PARC, would not go into the technical details of Magitti, but he did explain that software on the server constructs models from these sets of data that predict where a user is likely to go and what she will likely want to do based on her past behavior. Magitti pulls GPS data from her phone, as well as text messages and information about events saved in the phone’s calendar, and uploads it to a server, along with the user’s search terms, Partridge says. Text messages are important bits of information, he notes, because they often include information about future plans. If, for instance, a person is using Magitti to find a restaurant for dinner, and she gets a text message from a friend suggesting sushi, the software will put recommendations for sushi and Japanese restaurants higher on the list.

The idea of storing personal information as specific as location raises privacy concerns. But Bellotti says that this is something PARC considered when developing its system. This is why text messages are only kept for a short amount of time. But ultimately, there’s a trade-off between privacy and convenience, especially with this new breed of context-aware, location-based technologies. “I think people will initially accept these location-finding models when there is a big benefit to them,” Bellotti says. “Once they realize that nothing bad is happening, then they may become even more comfortable with it.” She likens the situation to the fact that people use credit cards for convenience even though their personal information is accessed each time they use one, and they are, in essence, leaving a digital trail behind them.

But there is still a question of how much benefit this software can actually provide. Partridge admits that the technical problems aren’t completely solved yet, and there is still work to do to make the software more accurate. One of the problems, he says, is that some of the categories that people use for activities are somewhat ambiguous. For instance, “shopping” could mean going to a farmer’s market, or it could mean going to Macy’s. “Eating” could mean sitting down at a restaurant, grabbing a sandwich at a grocery store, or enjoying a meal at home. Partridge says that there is still work to be done to make the categories more clear, so that the recommendations will be more accurate.

Magitti raises a number of questions about how people interact with recommendation systems, says Mor Naaman, a research scientist at Yahoo. “We know people do well with recommendations from Netflix or Amazon,” he says. “When you turn on a computer and it knows exactly what you’re trying to do, and it gives you good information, that’s the best thing in the world. It’s almost like magic. But I think a lot of the trick will be in the user interface and how users perceive and interact with it. That remains to be seen in a wide deployment.”

Magitti will go through public trials with young adults in Tokyo in the spring of 2008. Depending on the feedback, it might be released more broadly. The United States mobile market is another challenge entirely, says Partridge. Due to the splintered mobile market that includes various carriers and device makers, it’s much more difficult to deploy a service such as Magitti here.