The neighborhood that is the Internet long ago went bad, forcing computer users, in essence, to lock their doors.

But researchers at Carnegie Mellon University say computer security could be enhanced further if they can instill computers with the digital equivalent of street savvy.

The National Science Foundation has awarded the researchers $6.4 million over the next five years to develop this idea. They'll be modeling the normal patterns of human/computer interaction, so computers will be better able to recognize Internet traffic that is abnormal or dangerous.

"It's not a whether [this approach] will work or not," said Mike Reiter, who will lead Carnegie Mellon's new Security Through Interaction Modeling Center. "It's a question of how well. We have reason to believe it will be pretty effective."

As viruses, worms and other nasty threats have proliferated on the Internet, human users have learned through hard experience that certain actions, such as opening an e-mail attachment sent by a stranger, can lead to trouble, Reiter noted.

Today's computer security devices, such as firewalls, are designed to follow specific instructions, blocking certain kinds of known threats. But they can't learn, so they can't recognize unanticipated attacks.

Reiter hopes to incorporate models of normal Internet interactions into these defenses, so that the computer can delineate what is normal and what is not. In this way, computers should be able to recognize a wider range of possible attacks.

The funding comes from the NSF's Cyber Trust program, which promotes research into more dependable and secure computer systems.

While Carnegie Mellon studies ways for computers to stay out of trouble, researchers at the University of California, San Diego, will receive $6.2 million of Cyber Trust money to study the epidemiology of the Internet, so computers can sense spreading infections and take actions to suppress them.