Chapter 17. Agent Safety and Security
Abstract
Automobiles have proven to be a wonderful invention, and people all over the world depend on them. But people also recognize that cars used improperly can cause injuries. Cars can also serve as hiding places for car bombs. It's hard to imagine a software agent that could cause physical harm to anyone -- it's only software, after all. But what if that software is controlling an electrical appliance, say a coffee maker? Could a control failure cause the coffee maker to overheat and start a fire? We hope not; the coffee maker should in any case have passed an Underwriters' Laboratory test to assure that it won't start a fire even if it fails catastrophically. It's easy to imagine a software agent that damages data stored in computers, however: usually, it's called a virus. Of course, agents are supposed to be friendly and useful, not malicious and destructive. And they presumably operate in a constrained environment of some sort. But to be useful, an agent must be able to operate flexibly and dynamically: it may, for example, need to determine where to go next in search of some particular piece of data. It may need to store results or to send messages back to its initiator. It certainly will require some computing cycles on every site it visits. (We use the term applet for agents that are imported from a remote site for strictly local execution; agent encompasses both applets and mobile agents.)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1996
- Accession Number
- ADA465298
Entities
People
- Carl Landwehr
- David Goldschlag
- Michael Reed
Organizations
- United States Naval Research Laboratory