4.9 Intelligent Agents

4.9 Intelligent Agents

Unlike an expert system, an agent is embedded in its environment and can perceive and react to it using inputs of conditions. It can dynamically construct new rules as it works; in other words, some agents are capable of using sensors to monitor their surroundings, develop new rules, and then take action independently. For example, Doppelgaenger is an agent developed at MIT's Media Lab that uses sensors that provide many kinds of information about the user population in its computing network environment. These sensors can include active badges that provide location information about users, along with login information that detects their arrival and departure from the MIT computer network.

This agent can gauge user actions that reflect the frequency and duration of the use of applications, programs, and workstations and telephones. The system monitors what applications and data sets users tend to access in order to construct a profile of user preferences so that as it monitors their behavior, it dynamically creates new rules. Doppelgaenger uses this user information to construct profiles into statistical clusters; this type of intelligence is used by the agent to provide users with network-specific information likely to be of interest to them. It also uses this behavioral information to provide them with notification about databases and applications that meet their profile interests. It can also alert specific users about changes to data sets they have an interest in.

4.10 A Bio-Surveillance Agent: A Case Study

Through the use of these types of sensors, for example, agents could be used for the construction of a real-time bio-surveillance system for monitoring bio-terrorist attacks. DARPA recognized this and solicited applications for such an agent-based system, citing a covert release of an infectious disease as one of the most insidious threats to civilian and military personnel within the United States. DARPA believes that in addition to traditional threats to our national security, our adversaries now possess the ability to disable our nation's infrastructure and inflict casualties on U.S. citizens at home and abroad, as was made evident by the 9/11 attacks.

According to the Department of Defense (DOD), if effectively executed, such an attack could go unnoticed and infect a large number of our forces with a fatal disease. For the individuals that survive, the quality of life, burden on the medical system, and impact on the local government and economy would be immeasurable. For this reason, DARPA believes surveillance for covert biological warfare and biological terrorist activities is needed to counter this type of threat. If an event occurs, surveillance is needed to identify the presence of the pathogen or the initial indicators of disease as soon as possible so that a rapid response can be implemented.

DARPA envisioned the development of an integrated bio-surveillance system capable of very early detection of a covert release of biological agents (see Figure 4.1). In its solicitation, it called for a system that would do the following:

  • Receive and correlate data from heterogeneous databases

  • Glean applicable data from these databases, while maintaining patient privacy privileges

  • Analyze the data to discern abnormal biological events from normal epidemiology patterns

  • Classify abnormalities and identify specific pathogens, as well as determine the release event and location

  • Provide alerts to the appropriate DOD emergency response infrastructure

click to expand
Figure 4.1: Bio-terrorism system using agents with sensors.

This type of bio-terrorist system would obviously not only make use of agents equipped with an assortment of sensors to monitor various environments; the system would also make use of expert system technology, as well as data mining models constructed with neural networks and machine-learning algorithms designed to recognize the signatures of deadly viruses, all of which would need to be networked over a secured, fail-safe infrastructure, providing 24/7 support to medical, disaster recovery, and DOD personnel.

The objective of this program in bio-surveillance is to develop, test, and demonstrate the technologies necessary to provide an early alert to the appropriate continental U.S. DOD emergency response elements regarding a release of biological agents, involving both natural and unnatural pathogens, against military and civilian personnel. Specifically, the system would reduce the existing probable alert period for a nominal pathogen from four days after release to two days (see Figure 4.2).

click to expand
Figure 4.2: Agent system would serve to provide early detection.

DARPA specifically wants a bio-surveillance system using autonomous altering algorithms for monitoring multiple sources of data that would do the following:

  • Determine relevant sentinel data sources

  • Develop appropriate data correlation and analysis software for anomaly detection, pathogen classification, and epidemiology

  • Integrate information, using heterogeneous databases and traditional and nontraditional data sources

  • Detect pathogens despite masking by normal infections [similarily of symptoms of bio-warfare attack to other diseases (e.g., flu)]

  • Collect incident-specific information with adjustable privacy constraints

DARPA is calling for a system that would use agents to retrieve, monitor, and assemble the data, enabling other AI-based components to use models to conduct analyses, issuing alerts to the responsible DOD personnel. A modified system could be used to provide public health agencies and corporate offices of multifacility providers with early warning of biological or certain chemical incidents. Surveillance is accomplished through (1) broad-based, large-scale agent sensors monitoring for patient symptoms, and caseloads by type, therapy, and intervention or diagnostic procedures, and (2) the rapid retrieval of key documents and information for suspect cases from strategic clinics and hospitals.

A combination of agents and data mining models could be used to construct a bio-terrorism system so that, for example, the reporting of certain symptoms such a rashes or sore throats by regional health clinics, hospitals, and emergency-planning agencies, or certain increases in school absenteeism or frequencies of over-the-counter sales of cold medicine that exceed certain thresholds, may signal a potential attack. More sophisticated sensors might monitor water supplies or air samples. All of this in combination with data mining techniques and tools could be used to recognize an epidemic. Using agents, a computerized system could use emergency room data to monitor the frequency rates of rashes, fevers, coughs, and intestinal problems and the location of patients to identify potential biological attacks. The agents would work in unison with expert systems and other data mining tools in this anti-bio-terrorism system. Farther in the future are agents with the capability to detect bio-agents like aflatoxin in actual detection devices.