Semantic E-Mail Delivery
An experimental system automatically figures out where to send e-mail.
A prototype e-mail system being tested at Stanford University later this year will radically change how users specify where their messages are supposed to be delivered. Called SEAmail, for “semantic e-mail addressing,” the system allows users to direct a message to people who fulfill certain criteria without necessarily knowing recipients’ e-mail addresses, or even their names.
E-mail addresses are an artificial way of directing messages to the right people, says Michael Genesereth, an associate professor of computer science at Stanford who works on SEAmail. “You want to send messages to people or roles, not to strings of characters,” he says. Semantic technologies are aimed at making just this sort of thing possible. The idea is to create programs that understand context, so that users can interact with the software more naturally. Technical details, such as the need to specify an e-mail address, get hidden inside the system, so that everyday users no longer have to pay attention to them.
Genesereth says that users were wildly positive about a previous prototype built by his group and used among semantic researchers. For example, people wanting to send a message to “Michael Genesereth” could simply type his name as a recipient, and his most recent e-mail address would automatically be selected. A user could also send a message to a group such as “all professors who graduated from Harvard University since 1960.” SEAmail can handle both of these examples, Genesereth explains, without requiring the user to spend time doing research or keeping an address book up to date.
In SEAmail, a user selects recipients for a message in much the way that she would set up a search query. The parameters can be as simple as a person’s name, or as complex as sets of logical requirements. But the system is limited by how much information it has about potential recipients. “To realize the full potential, we need to have rich data about the people who are sending messages to each other, their interests, and so forth,” Genesereth says. Within an organization, he says, there’s usually a lot of available data. The technical challenge is setting up an integrated version of the data that SEAmail can access easily. The data needed to fulfill the request for professors who graduated from Harvard, for example, would probably come from several databases, Genesereth says. His team is currently researching ways to pull together existing databases without affecting how they’re already being used.
But getting good data for SEAmail becomes a much harder problem on the broader Internet than it is within an organization, Genesereth says. Although there are semantic standards that can allow systems to extract information about people from Web pages, he worries that outdated information could degrade the quality of the system.
“This technology has clear benefits, but it’s also ripe for misuse,” says Oren Etzioni, director of the Turing Center at the University of Washington. “The technical issues are solvable. The tricky things are the social issues. How do we create a workable system, given the vagaries of human nature?” In particular, Etzioni worries that, if the tool were broadly available, some people would receive overwhelming amounts of mail, without a good way to limit it. While semantic tools could be used to create filters for e-mails coming in, he says that there’s no clear way to control the flow of incoming mail without also losing out on some of the serendipitous messages that make such a system useful.
Assuming that worries about spam could be properly resolved, semantic e-mail addressing might be interesting in combination with other semantic approaches, says Luke McDowell, an assistant professor of computer science at the United States Naval Academy, in Annapolis, MD. McDowell worked on a system that extracted information from the body of e-mails to simplify the process of planning parties and agreeing on meeting times. In general, he says, semantic tools could help people manage their e-mails better by using contextual knowledge to automate tasks.
SEAmail will be used at Stanford later this year as part of a larger “digital department” project that aims to introduce several semantic technologies, Genesereth says. The computer-science department will use the system first, but the plan is for the technology to spread through the university until everyone has the option of using SEAmail. He sees the technology as having a lot of potential for internal use by large businesses, for which its benefits far outweigh the potential for abuse. However, with more refinement, he says, it could eventually become a tool for the broader Internet too.