ÐÓ°ÉÔ­´´

THE task of managing the environment is far from straightforward. Such
simple-seeming jobs as treating waste water from the homes in a town require
constant monitoring. When it comes to deciding how to treat toxic wastes,
an expert has to take into account various elements specific to both the
site and the chemicals, as well as weigh up the method of treatment against
cost, effectiveness and potential side effects. Difficulties are multiplied
when managing oil spills or fires in chemical plants, when fast reaction
and correct procedures are essential to limit damage.

‘Expert systems’ are now available to help cope with these and other
problems in environmental management. Expert systems are computer programs
which can make decisions based on a set of ‘rules’ drawn up by human experts;
they are among the most important practical uses of artificial intelligence
research. Simple systems, running on inexpensive personal computers, can
monitor the treatment of waste water, and automatically compensate for fluctuations
in the flow or quality of water that might otherwise disrupt processing.
Unlike human operators, they are not bored by such dull tasks.

Expert systems offer an important advantage over automated controls:
they can make decisions based on intricate criteria or incomplete information.
In a crisis, an expert system offers the emergency crew immediate access
to data on hazardous materials, and can help in selecting the most effective
course of action to take before human experts are available. Other systems
can carry the complex rules that allow planners to make accurate estimates
of the cost of cleaning up toxic wastes, or to judge whether a proposal
for treatment is the best one for the geology of the site and the chemicals
involved.

Artificial intelligence researchers started to develop expert systems
in the late 1960s, but it was almost two decades before the first environmental
applications emerged. The reason it took so long, according to Judith Hushon,
vice-president of Roy F. Weston, an engineering firm in Washington which
is a leading developer of such expert systems, is that environmental management
is a complex field.

The science is not well understood, and there are few absolute methods
for tackling pollution control. An added problem is that to identify the
optimum solution may need the combined efforts of environmental, civil and
chemical engineers, environmental chemists as well as toxicologists. It
can be hard to link such diverse fields into a single system.

On the other hand, the complexity of environmental management means
there is a real need for expert systems. The basis of many decisions in
pollution control, according to Daniel Greathouse of the US Environmental
Protection Agency’s Risk Reduction Laboratory in Cincinnati, Ohio, is ‘typically
a large array of interrelated information, some of which may be incomplete
or inaccurate’. Persons with the necessary training and expertise, he says,
‘are in short supply. Many of the decisions are made by recent graduates
of technical academic programs with two or less years of job experience.’

He concedes that expert system technology ‘is far from the stage where
the human thought process can be cloned’, but says that a good expert system
can reach conclusions in routine cases, freeing human experts for more difficult
problems.

Because of these advantages, the number of expert systems has grown
rapidly. Hushon counts about 80 for environmental applications, and admits
she may have missed some outside North America. Most of the systems on Hushon’s
list run on microcomputers, but more than a dozen of the more complex ones
run on minicomputers, and a handful run on mainframes. Developers are moving
up to more powerful computers because the most common microcomputers, clones
of the IBM PC/AT, can address only 640,000 bytes of memory. Many programmers
can squeeze their systems into the limited space, but more room is needed
for complex tasks.

The Environmental Protection Agency does not have a major program to
develop expert systems per se, but ‘we do have a lot of individual projects’,
says Darwin Wright of the EPA’s headquarters in Washington. The agency’s
interest in expert systems arises partly from its decentralised structure.

The EPA is charged with enforcing federal environmental laws, supervising
clean-ups of toxic waste sites and granting permits for industrial operations
affected by federal environmental laws. Most decisions on how toxic waste
sites should be cleaned up are made by 10 regional EPA offices, but in the
lawsuit-happy US, it is important that the decisions be consistent. Officials
at the EPA headquarters complain that the regional staff lack time to read
books containing the information they need. Moreover, high turnover of personnel
makes it important to transfer knowledge.

Many of the EPA’s systems are designed to cope with paperwork. One,
called Permit Writer’s Assistant and based on a microcomputer, helps to
issue permits to factories applying, for example, to discharge waste into
a river. Another reviews plans submitted by contractors for disposing of
hazardous wastes. It checks that the proposed confinement systems are adequate
to withstand chemical attack from the wastes present, and that wastes put
into the same site are compatible. Greathouse is developing a system that
will also review requests from contractors who, having received approval
for plans to treat waste at a particular site, wish to modify the terms
of the contract because conditions turn out to be different than expected.

Another area that the EPA is working on is leakage of petrol from underground
tanks. William Foskett of the EPA headquarters says the agency has only
40 people to monitor two million underground tanks, leaks from which can
contaminate wells and aquifers. He is working on the Soil Treatment Technology
Adviser, an expert system to identify the best technology for treating a
particular leakage. ‘My systems are not intended to do engineering,’ he
says, but rather to help state regulators judge the effectiveness of a proposed
treatment.

Foskett is also working on a system to help regulators determine the
urgency with which problems need to be tackled. The hodge-podge of local
regulations can make this complex. His policy adviser will help regulators
give consistent interpretations, and automatically produce documents explaining
why certain clean-up standards must be met. ‘It’s not really intellectually
high-powered stuff, but it’s a place where people can make mistakes,’ he
says.

The EPA develops some of its own systems, and commissions others from
outside contractors such as Roy F. Weston and Arthur D. Little, the international
engineering firm based in Cambridge, Massachusetts. One request for expert
system research drew more than a dozen responses. Companies are realising
there is a commercial market for such systems.

Some developers use special computer languages marketed expressly for
writing expert systems. Known as ‘shells’, these save work by letting people
‘write programs in English-like grammars’, says Hushon. The shells also
handle screen management, compute uncertainty, and devise strategies for
problem solving. As with other high-level computer languages, those advantages
come at the cost of flexibility and speed – limitations particularly severe
on microcomputers. Because of those restrictions, shells are more useful
for developing prototypes than for writing programs that are intended for
regular use.

Chemical companies often build expert systems around commercial shells.
Both DuPont and Monsanto have developed decision support systems for operators,
engineers and maintenance workers at their plants, using the G2 Real-Time
Expert System offered by the Gensym Corporation of Cambridge, Massachusetts.
Gensym has installed more than 300 expert systems, although only a few are
specifically for environment management. Perhaps the most intriguing is
the one that controls the environment of Biosphere II, a man-made ecosystem
in the Arizona desert (‘Biosphere II: a world apart’, New ÐÓ°ÉÔ­´´, 18
March 1989).

Biosphere II is a privately funded program, masterminded by Space Biospheres
Ventures, to study an artificial self-contained environment in order to
learn more about the ecosystem of the Earth, which the developers call ‘Biosphere
I’ Since November, the expert system has monitored the concentration of
gases in a test module of 500 cubic metres.

‘It alleviates the need to keep someone full-time on the system to monitor
it,’ says Norberto-Alvarez Romo, a project engineer who will live in Biosphere
II when the 200,000 cubic-metre structure begins operation in September.
Romo, who will join seven other humans and 3800 species of plants and animals,
in addition to countless bacteria and other microscopic life, will stay
inside the self-sufficient Biosphere II for two years. By September there
will be a more elaborate expert system in place to ‘monitor everything –
temperature, water, water flow, air and nutrients in the water – and also
to run checks through the data base for analytic lab results,’ says Romo.

Many expert systems do much more mundane jobs. There are 10 systems
available to manage waste water treatment, a vital job in every town with
a sewer system. Most run on microcomputers, and use commercial sensors.
Treatment plants operate continuously, but inflow varies widely during the
day. Variations in the rate or quality of input, or accidental or intentional
discharge of toxic wastes, can disrupt growth of the bacteria which decompose
sewage.

After such events, operators need to determine what happened, then readjust
conditions to the proper levels. Human operators have long done the work,
but as Hushon notes, ‘it’s not a very nifty job’. The expert system always
pays attention, and can automatically compensate for disruptions to regenerate
the bacteria.

In the UK, NNC (the former National Nuclear Corporation) uses an expert
system to analyse hazards in large industrial plants, which guides designers,
no matter how limited their knowledge of handling hazardous materials, to
produce safe designs. Among the tasks the expert system handles is checking
hazard classifications of different parts of the plant, so that explosives
are not sent into areas containing dangerous chemicals. Barry Keen, an engineer
with the NNC, says that they are also working on expert systems to aid in
managing the operation of industrial plants, but details are proprietary.

One unusual expert system for plant management is being developed by
Peter Henderson of the Fawley Marine Laboratories. The laboratories are
adjacent to the Fawley Power Station which overlooks Southampton Water.
Coastal power stations have long suffered from fish clogging the intake
valves for their cooling water, a problem that has shut down the nuclear
plants at Dungeness and Hartlepool in the past. Henderson’s system, which
incorporates data from 10 years of research on fish movement, predicts the
total annual ‘catch’ of various species that would be trapped on intake
screens at any coastal location.

Hushon sees much interest in expert systems to aid emergency response
crews, who ‘have to know so much to do a good job’. She has developed FRES,
a prototype First Responder’s Expert System, for chemical spills. Users
answer the computer’s questions about the immediate environment and the
nature of the incident. The system analyses that data to determine chemical
toxicity, first aid requirements, protective gear recommendations, potential
chemical interactions, need for evacuation and clean-up recommendations.
FRES runs on a microcomputer, so that it can be taken to the scene of the
emergency.

The biggest limitation on FRES, according to Hushon, is the present
generation of data bases. What is needed, she explains, is a data base that
stores values – such as boiling point, flammability, vapour point and toxicity
– in numeric form to permit comparisons and evaluations. Existing chemical
data bases store the values along with other information in text fields,
a much less useful form.

The Alberta Research Council in Calgary is working on Hermes, or the
Heuristic Emergency Response Management Expert System. Like FRES, it prompts
users for vital data and then gives a hazard assessment and advises about
protective clothing, evacuation and other potential responses. The screen
displays a map and graphic plots of hazard levels, as well as text. Future
versions will be connected to computerised map data bases. Hermes also can
serve as a training tool, says Chris Lumb, industrial liaison manager for
the research council. That capability is important; emergency crews now
have little chance to test their skills except in crises.

Lumb says Hermes has passed preliminary tests, but much work remains.
It provides very general responses, for a limited number of chemicals, including
propane, nitrogen and anhydrous ammonia. It also considers only surface
transportation of chemicals in an incident, and does not model dispersion
into water or the ground. The current version runs on microcomputers or
special-purpose computers developed by the Symbolics Corporation for artificial
intelligence applications; future, more powerful, versions will run on Sun
workstations.

Oil spills in water can cause pollution over a widespread area. Engineers
at BP Exploration in Aberdeen are modelling the spread of an oil slick with
a computer, but are sceptical about expert systems. To calculate the drift
of an oil spill, the company is developing Scoop, a model which incorporates
data on wind and tides.

Chris Brandon, one of the BP team, believes that an expert system telling
you how to treat a spill would be counter-productive in some cases. The
present computer model can predict where the oil is going and how much should
remain on the surface, but Brandon says that deciding how to deal with the
oil ‘is more a matter for the trained operator’.

Despite their promise, the applications of expert systems remain limited.
The most widely accepted ones, such as the CORA system that the EPA uses
to estimate costs for cleaning up a waste site, address small, definable
areas of expertise. Although such systems may assist regulators, engineers
need a much broader range of expertise to mastermind, for example, a clean-up
operation. The hardware limitations of current microcomputers pose a barrier
to building more complex systems. Hushon says some systems failed because
they ‘were too naive or only walked you through a decision tree of regulations.

Establishment problems also have slowed the spread of expert systems.
Many have been developed as student projects in universities, where little
effort is given to providing the friendly interfaces needed by users. Some
students may take their systems to industry when they graduate, but others
abandon them. Concern about legal liability is another impediment in the
litigation-happy United States. Although no one has yet been sued for offering
incorrect data in a data base, Hushon says potential liability is a psychological
impediment to raising development money. Because of that concern, many developers
have relied on government funding, with the government assuming a greater
part of the liability.

Lumb says that one emerging trend is a growing integration of expert
systems with other systems, such as chemical and environmental data bases.
Expert systems that work with real-time sensor inputs, like those used for
waste water treatment, could be integrated into systems that automate chemical
manufacturing.

As microcomputers grow more powerful, developers hope that the advantages
of expert systems will lead to an ever-wider range of applications.