Making Systems Come to Us

When I begin a document with Microsoft Word 2000, menus like Edit and Format are mercifully short. Commands that I rarely use dont appear unless I summon them—or briefly wait for the full-length list. But if I dig down to an obscure command, it will be on the short list thereafter: The software smooths the path I choose to take.

This general notion—that systems should adapt to their users, instead of the other way around—is long overdue but finally becoming affordable in contexts that are far more strategic than word processing. Internet retail sites, for example, will reduce transaction overhead per customer if they can recognize and initiate patterns such as offering spare batteries along with new digital cameras. Content sites will be more useful (and therefore more used) if they tailor their presentation to time and day of the week: At 6 a.m. on a Monday, I want weather and traffic upfront, but at 9:50 p.m. on Thursday, I want to know if “ER” is a repeat.

“User modeling” sounds like a fearsome artificial intelligence task, but in most contexts, it doesnt need complex methods. “Cheap user modeling” techniques, so labeled by MIT Media Lab researcher (and Perl guru) Jon Orwant, include simple Markov chains (if Im in my car on a weekday morning, Ill probably next retrieve my e-mail from my office) and linear predictions (if I read Chapter 1 of a document on Monday and Chapter 2 on Tuesday, Ill probably access Chapter 3 on Wednesday).

Some people would rather make their own choices: On roads that I know well, I prefer a stick shift myself. But the route through cyber-space changes every day: Im ready for the convenience of putting my life in drive and worrying only about how fast I can take the curves.