Expert systems slip into the mainstream
- By Brian Robinson
- Jul 20, 1997
In the early days of their development expert systems were touted as the answer to just about everything. By allowing computers to "think" like humans their supporters believed such systems would marry the problem-solving abilities of people with the electronic speed of computers and usher in a new age of enlightenment. That proved to be hopelessly overdrawn however and expert systems eventually seemed just one more hyped technology that would eventually wither and die.
Well not quite. They may not be called expert systems anymore but that technology's descendants - variously known as intelligent systems intelligent agents autonomous agents and knowledge advisers - have quietly become common components of many information technology systems and they are spreading.
In concept expert systems are simply a way to codify the knowledge and expertise of people so it can be used to provide guidance and solutions for given situations. Software "wizards" - those helpful little characters that are cropping up everywhere in PC software and that often seem to know what you are going to do before you do it - are just one example of the current generation of expert systems.
In government too expert systems are becoming a part of the everyday scene.
They help budget examiners analyze agency funding and write appropriations legislation they are becoming an important environmental aid to industry and the public and the military uses them for logistics and in battlefield situations. But getting to that point hasn't been easy.
"Expert systems have got a bum rap over the years " said Arline Dell a management and budget information analyst in the Office of Management and Budget. "Because of the early hype people invested in huge overhauls of their systems and consequently had huge failures. Consequently management overall got a very bad idea about expert systems."
With the development of expert systems coming so soon after movies such as "2001 " "everyone wanted to build HAL [the psychotic but very human-like computer of the Stanley Kubrick classic] and that was the criteria of success " said Dustin Huntington chief technology officer of Multilogic Corp. and the developer of EXSYS one of the leading expert-system development tools. "That was impossible then and likely will be for the foreseeable future."
But expert systems have proven to be very successful in tackling small well-defined problems Huntington said and that is where the focus is now.
In the early days of expert-system development however the concept of codifying knowledge was taken to the extreme.
Expert systems were often seen as systems that could incorporate the entire institutional knowledge of an enterprise replacing entire legions of human experts that organizations employed. Once built the theory went all you would have to do would be to plug the problem into the expert system and it would spit out an answer.
Underlying this was a belief that computers and artificial intelligence would eventually be able to mimic the human thought process with all of its subtleties and nuances. But "it turns out the brain is far more complicated than people thought " Huntington said. "Studies show that at the lower levels it may even be chaotic in nature. Right now it's impossible to accurately model the human brain in expert systems."
Instead current expert systems are used more as aids to human analysis of situations. For example the Carnegie Group's Knowledge-Based Logistics Planning Shell was used by the military in Bosnia to plan a fuel distribution network for U.S. troops there as part of the peacekeeping mission.
In the past explained Bill Elm director of advanced decision-support systems in Carnegie's government business unit this kind of planning was a "gut-check" thing if it were to be done in any timely fashion. Any kind of precise analysis could take several years using a team of expert logisticians - an approach that was feasible in the long-lived and relatively stable scenarios of the Cold War era.
"But that's not feasible today " he said. "Time constraints call for a much faster response."
It is highly complex fast-moving situations such as Bosnia that Elm believes show off the value of the current crop of expert systems. He described the situation where one field person used Carnegie's system to analyze a problem with another person at military headquarters working the same problem using traditional card-file methods. Both came up with widely differing numbers although the field officer was eventually able to convince the headquarters staff that his analysis was the correct one.
"In fact the distribution system did work as the field guy laid it out " Elm said. "The numbers came up well within 10 percent of the actual numbers. Our criterion is that if you get to plus or minus 20 percent of any planning system then you will be doing well."
In this instance Elm said the expert system is used to bolster the user's expertise because the system contains a lot of domain knowledge of various scenarios so it can quickly put together an analysis of a given problem. But there still needs to be a dialog between the system and the user Elm said and the user is still able to apply a gut-check to the final result.
"It's a collaborative way of using expert systems " Elm said "and in this sense it's a powerful tool for the knowledgeable professional."
At OMB expert systems are used extensively "and they are wonderful " Dell said.
One system in particular is used to track agency funds for budget purposes a task OMB needs to do in an exacting way to identify how the funds will be used. In this case the expert system is used to help OMB budget examiners open up new accounts through which they track these funds and see if the funds are being used according to the specifics of appropriations legislation.
"This matters because for OMB examiners setting up a new account is not a frequent thing " Dell said. "If it's more than twice a year then that's a lot so examiners have no in-depth knowledge of what such an account should be and how it should be set up."
OMB also needs to develop appropriations language for budget submission which means dealing with specific legal language. But people at OMB are not attorneys or program specialists Dell pointed out so the agency has developed an expert-system appropriations adviser that prompts OMB users for the kind of things that need to go into writing appropriations language.
"The examiners use this during the budget season when things are the most rushed and people are up against deadlines " Dell said. "It's a tool they can use at just the time they need something like that."
The use of expert systems at the Occupational Safety and Health Administration is aimed more at the outside world than internally. OSHA develops publicly available "advisers" that people can download or access through a Web page these advisers help users such as architects or construction companies to comply with building codes and environmental regulations.
Several of these expert systems were put on OSHA's Web site recently said Ed Stern director of OSHA's expert-systems project. Already the systems are getting "significant" attention with up to 500 downloads per week and around 500 hits per week for online advisers. The agency intends to develop many more of these advisers Stern said.
Not everyone has had the same success in deploying expert systems however. The Internal Revenue Service with its sprawling and Byzantine tax code and regulations would seem an obvious candidate for the use of expert systems both within and outside the agency. But the agency is involved in the tax modernization program and any sweeping introduction of expert systems would require a change in the scope of that program.
Nevertheless the IRS has managed to introduce expert systems in some small niche areas said Gary Morris who heads up the IRS' artificial-intelligence lab. The agency uses a system called Determiner for example to answer the question of just when someone is an employee or a contractor. This is a perennial bone of contention at tax time with companies wanting one decision and workers another. The IRS has to make the final decision and Determiner is used as an "expert in a can" by lower-level employees to help them recognize what questions to ask.
Bring It to Life
Actually building expert systems has become much easier than in the early days when considerable programming and systems knowledge was needed. Now the most important part of the process for agencies that want to use expert systems is deciding what the system will be used for and making a thorough analysis of the agency process that the system will be applied to.
"I would call this the bellwether step " Elm said. "We find it critical to do this because there are underlying ways in the process that will enable an intelligent system to 'come alive' when used by an intelligent person. Otherwise people tend to become enamored of the technology itself throw something up there and say they are working with intelligent systems."
After that there are a number of commercially available tools mostly in the form of expert-system "shells" in which people can build the particulars of expertise. Costs will then vary depending on whether the expert system is done from scratch for a particular application or whether it is used to add functionality to an existing application.
"Writing an expert system is a heck of a lot easier than it used to be " the IRS' Morris said. "There are many off-the-shelf tools available now. [Use of this technology is] not leading-edge anymore but expert systems plus other technologies such as natural language become an enabler."
In the case of OSHA's expert systems for example it takes about a year to produce the result - for a typical cost of $65 000 to $85 000 according to Stern. However OSHA uses outside contractors to develop and write the system software.
But the tools themselves can be relatively inexpensive. EXSYS for example only costs a couple of thousand dollars (depending on the configuration) runs on a PC and "is at the same complexity as a word processor " Huntington said. Typically he said it takes just a day or so for people to get to where they can easily input logic sequences into the system. The next generation of EXSYS due out in September will be "even easier to use " he said.
Despite all this and despite the obvious success within agencies of using these smaller modular intelligent systems many IT professionals still find it a tough haul to convince their managers and even fellow professionals of the worthiness of expert systems. Overcoming the stigma of the technology's early years has not proven easy.
But there are some observers who see difficulties in applying expert systems to government. Steve McClure director of object tools for International Data Corp. believes the diversity of organizations within the government will make it difficult to formulate a consistent approach.
"Legislation is not written by logicians but by politicians and [it] is usually left intentionally vague for later interpretation " he said. "Because of this people who are trying to automate things and put intelligence into the front line for people to use often run into contradictions about how to write rule-based systems."
Nevertheless he sees a wide opportunity in government for the use of such things as "symbiotic" systems which flag users when something needs attention and prompt for further information and for general intelligent-assistant applications.
To spread the gospel federal agency professionals together with members of academia and industry formed the Intelligent Technology Group. Chaired by OMB's Dell it's an informal group that has been meeting for the past several years to share information about expert systems and other artificial-intelligence technologies.
A more standards-based industry association is being formed though the formal announcement has not yet been made. Companies are trying to get ahead of possible future problems over conflicting agents said Peter Janca manager of IBM Corp.'s Intelligent Agent Software Center by seeing if they can write some common operating rules for agents.
And Janca warned that whatever the current ambivalence over the use of expert systems and intelligent agents people should get used to the idea of those tools being an integral part of the IT landscape. In a study he conducted several years ago one he said he still stands by Janca predicted that by 2000 "every significant mainstream data proc-essing application" will contain intelligent agents.
-- Robinson is a free-lance writer based in Arlington Va.
* * *
At A Glance
Status: Expert systems are used throughout the federal government but they are struggling to overcome a poor reputation.
Issues: Increasingly simple to create and manipulate expert systems are proving popular in situations where high volumes of complex knowledge have to be handled.
Outlook: Good. Expert systems are quickly winning converts and should become ubiquitous in a wide number of data processing applications within the next few years.