"Excuse me, are you human?"

How do you know your new e-mail pen pal isn't an intelligent agent?

Published January 25, 2000 5:00PM (EST)

"The number you have reached is not in service at this time. Please check the number you are dialing or contact your operator for assistance. This is a recording."

Remember that message? The time was the 1970s, and Bell Telephone was in the process of upgrading phone switching systems all over the country. Ma Bell, it seems, was fearful that a technologically unsophisticated customer might mistake Bell's recorded messages for an unresponsive, unfriendly, human being. Rather than risk an upset customer, the Bell system prefaced every message with a few tones, and concluded each with those oft-parodied words, "this is a recording."

Perhaps Ma Bell was being too cautious. Today those four magic words have largely been banished from the telecom lexicon, yet there's little fear among telco executives that somebody's grandma will start e-mailing complaints about rude and insensitive operators.

Ironically, if Grandma did write an e-mail about poor service, it's increasingly likely that her message might be read and replied to by a machine -- a machine engaged in the elaborate deception of pretending to be a human being.

Already, most of the e-mail sent to President Clinton at the White House is intercepted, categorized and replied to by a sophisticated mail handling system. Originally designed by researchers at the Massachusetts Institute of Technology Artificial Intelligence Laboratory, the system determines the purpose of the e-mail that has been sent to the president and chooses a response from one of many that have been previously written by a human staffer.

According to MIT professor Tom Knight, the White House system then displays the e-mail message and the selected response to a human operator who nominally checks the machine's work before clicking the send button. The system keeps track of how many times each constituent writes; if this is the second time you have written on a particular subject, you automatically receive a different form-letter response -- one designed for people who are especially concerned about a topic. Of course, the White House system also keeps track of how many letters have been sent on each subject and their stated opinions, so that the executive branch gets some kind of feedback from the people it allegedly represents.

Automated technology for handling e-mail is rapidly moving into the world of e-commerce as well. The January issue of MIT's Technology Review magazine includes a profile of General Interactive, whose EchoMail product is currently used by Nike, J.C. Penney and other companies to automatically screen and route incoming e-mail.

Interestingly, the article credits EchoMail with J.C. Penney's rapid decision to cancel its sponsorship of the television show "Ellen" in May 1997 after actress Ellen DeGeneres' TV character came out as a lesbian. According to the article by Deborah Shapley, EchoMail analyzes each incoming e-mail to determine which product is referenced, the kind of request, the issue and the "attitude" of the person who composed the e-mail message. Back in 1997, EchoMail determined that lots of hostile e-mail messages were coming into J.C. Penney and alerted its human supervisor that they demanded immediate attention.

Unlike callers receiving those recorded telephone messages of the 1970s, most people receiving a canned response from the White House or J.C. Penney today probably don't realize that their e-mail was read and replied to by a machine.

Unlike a recorded message, there's no easy way to tell if an e-mail message is a canned response or has been personally composed and sent to the recipient. Indeed, it's reasonable to assume that, as e-mail becomes more and more pervasive, Internet-enabled organizations will go to great lengths to make sure that their automatically generated replies appear as human as possible. Perhaps they'll delay responding to a message for a few hours or pepper their missives with occasional grammar or punctuation errors.

In the world of e-commerce, little matters as much as the appearance of "customer care" and its resulting customer loyalty. Personalized e-mail encourages warm feelings on the part of a customer, whereas nothing poisons a customer as surely as bulk-reply e-mail containing canned responses.

Automated systems work because people tend to send e-mail on similar topics. Form-letter replies work great if you can pick the appropriate form response to a particular letter. But the precedent that these systems are creating -- a precedent that makes it OK for computers to imitate people -- is dangerous. It's a precedent that could eventually tear at the social fabric of the online world.

Today, exchanging e-mail is probably most people's favorite Internet activity. We use it to catch up with friends, make dinner plans, exchange ideas. E-mail is also a great way to make new friends -- I know of more than one couple who met and fell in love as the result of speaking their minds on a mailing list.

Despite all these positive associations with e-mail, though, there are times when e-mail brings the most hateful of Net experiences -- a deluge of spam. Junk e-mail turns the pleasures of e-mail inside out. Instead of a pleasing exchange with friends, a blast of junk e-mail is an attack by someone who is trying to take advantage of you. The only silver lining to spam is that you can frequently identify it -- and therefore delete it -- without even opening it.

But, what if the people who spam you were to borrow a trick from EchoMail? What if instead of receiving an e-mail message shrieking, "BUY THIS BOOK NOW!" you received an e-mail message from a woman at that company -- an employee who stumbled across your Web page and shared some of your interests. Say you engaged in an e-mail conversation for a few weeks, and then one day this woman wrote that she was reading a book and really enjoying it. Over the next week, she sends a number of e-mails, telling you how wonderful the book is. Then she insists that you get a copy of the book, so that the two of you can talk about it on a deeper level.

Technologically speaking, there's no reason why marketers can't adopt this sort of aggressive, agent-based reach-out marketing. And economically speaking, there is no reason why markets wouldn't embrace it. Computers and bandwidth are getting cheaper every day. By comparison, the number of consumers in the world is increasing only nominally. In the future, e-commerce companies will come under increasing pressure to use advanced technology to sell products; simulated human beings and electronic replicant friends will surely prove a cost-effective method.

I don't mean to suggest that companies like Nike and J.C. Penney will resort to reach-out marketing in the coming years. But some companies will, and it won't take many companies adopting these technologies before the online community is irreparably poisoned. Today, when you receive an e-mail message from a person you've never heard of, there's a reasonably good chance that, if it's not recognizable as spam, another human being is trying to make contact with you. But if reach-out marketing becomes widely used by even a few bad actors, there won't be any way to know for sure.

Technologists might argue that public key cryptography could provide an easy answer to this problem: Simply require that each person who sends e-mail have a signed public key, and then make sure that the organizations issuing the signatures do not sign keys belonging to computer programs. But even ignoring the privacy implications of such a regime, the technological solution won't work: There is no way to prevent a marketer from sharing his public key with an intelligent reach-out agent.

I think that the only real solution to the eventual problem of reach-out marketing will be legislative. One approach would be to adopt Ma Bell's strategy from the 1970s: Require computer programs that send e-mail to identify themselves as such, and require that each computer-generated e-mail include instructions for contacting a real human being. Another approach would be to simply ban the practice before it becomes widespread.

Whatever the solution, I hope we can get proactive -- and quick. The last thing I want to see is a crowd of intelligent agents causing the destruction of our online communities.

By Simson Garfinkel

"Simson Garfinkel is a frequent contributor to Salon, the Chief Technology Officer of Sandstorm Enterprises, and the Chief Scientist of Broadband2Wireless, Inc."

MORE FROM Simson Garfinkel

Related Topics ------------------------------------------

Artificial Intelligence Privacy