Marine Oil Spill Expert Systems

Gerald Graham, Ph. D.

Worldocean Consulting Ltd.

Victoria, BC, Canada

worldoceanconsulting@live.com

Ian Morrison, Ph. D.

Acquired Intelligence, Inc.

Victoria, BC, Canada

Abstract

This peer-reviewed scientific paper, originally presented at the 27th AMOP Technical Seminar in 2004 in Edmonton, Alberta, Canada, examines the general suitability of expert systems to marine oil spill response operations. Expert systems are interactive, computer-based decision support tools which capture the knowledge of so-called ‘domain’ experts. They differ from information management systems in the sense that they explicitly represent the decision-making process of experts.

Although no two spills are identical, response experts typically employ any number of ‘rules of thumb’ when formulating a response strategy for a particular spill. Once identified, these patterned responses can be automated. There may never be unanimity on the best way to respond to a spill, but unanimity is not required. For an expert system to be useful, all that is required is that the knowledge it incorporates lead to a more efficient response.

Expert systems will never replace human judgment. However, the current generation of response experts is rapidly nearing retirement after thirty years of service. With major spills occurring less often, it is important to preserve the bank of knowledge acquired by these experts, for transfer to and use by the next generation of responders, who were raised in the high-tech era.

Leading edge expert system technology allows one to simply, cheaply and effectively commit these lessons learned to the computer for easy access, retrieval and application at the time of the next spill.

1 Introduction

Expert systems are interactive computer software applications that capture the knowledge of an expert or experts in a particular domain; the persons whose knowledge is captured are usually referred to as ‘domain experts’. For their part, the computer professionals who commit this knowledge to a database are called ‘knowledge engineers’. The software they develop includes a set of ‘rules’ which are deemed by the experts to govern the field in question. The rules in question are based on the collective experience of experts over the years. Typically, expert systems also include an ‘inference engine’, which is the method for deciding which rules are to apply in the present situation.

Expert systems take the knowledge of an expert and make it available to others in an interactive format. They attempt to replicate the way an expert gathers and processes information related to a particular problem. Essentially, the expert is trying to gather as much information as is necessary to determine the nature of the problem, with a view to finding a solution to it. One characteristic of experts is that they know what questions to ask; this skill allows them to zero in on the problem. The responses their questions solicit can also lead them to skip a bunch of subsequent steps that a non-expert might waste time on. This time and effort-saving feature can be built into expert systems

A principle which underlies expert systems is that if something happened in the past, then there is a good chance that it will happen again at some point in the future. A corollary to this theorem might be that if someone is faced with a particular situation, the chances are that someone else, somewhere in the past, has encountered a similar situation. What the person faced with the challenge now wants to know is:

• What are my options for dealing with this situation?

• How have other people successfully dealt with similar events in the past?

• What are some of the pitfalls, and how can I avoid them?

2 What Makes Expert Systems So Special

Expert systems are one type of decision-support system, i.e. they are intended to assist people as they go about making decisions. An expert system differs from an information management system in that it explicitly represents the decision-making process of one or more human experts. An information management system can be an enormous aid to a person making a decision, and it may restrict the available options to such a degree that it appears to be offering a solution. But it is not following, or even attempting to follow, the same process that a person would. An expert system, in capturing a person's expertise, does try to emulate the way that that person processes information. And the source of that information, including restrictions of available options due to relationships between data, can be an information management system, just as it can be any other source of information that the expert uses. Expert systems and information management systems are not incompatible. The crucial point is, however, that an expert system actually makes a decision whereas an information management system provides information for someone else to make a decision.

An expert system can make use of the information from an information management system, just as it can raw data or some other analysis. The converse can also be true; in other words, an information management system can include rule-based decisions. What this shows is that any complex application requires a mix of technologies. In our experience, for instance, even “pure” expert system applications involve knowledge-base and inference in only thirty to seventy-five percent of the application. Thus, it is entirely possible that at some point in the future convergence will occur, such that expert systems and information management systems will combine so many of the other’s attributes that they will be virtually indistinguishable. For the time being, however, they remain separate types of applications.

It is perhaps useful to acknowledge another confusion about expert systems. An expert system is meant to capture the knowledge of one or more people for performing a particular task, or a small set of tasks. In the strictest sense, an expert system does not handle situations not previously encountered by the experts. Prospective clients sometimes ask for an expert system that can solve difficult problems for which they have no solution and are disappointed to learn that an expert system must be based on the knowledge they actually have rather than what they would like to discover. The knowledge acquisition process can include the application of an expert's knowledge to new situations by drawing the attention of the experts to the new situation and observing how they respond, but this isn't self-learning. There are artificial intelligence approaches to learning new knowledge (machine learning algorithms, neural networks, data mining, case-based learning) and there have been attempts to integrate these techniques in expert system applications with limited success. Expert systems, like other technologies, are not as flexible as people and can only incorporate some of what a person knows. Still, one shouldn't underestimate the value of just some of what an expert knows when that expert isn't there to ask.

3 A Brilliant Future Behind Them?

Expert systems once had a reputation for being big, expensive, complicated affairs, and many of them failed because they weren’t able to deliver what they promised, they took too much time and effort to maintain and keep up to date, and they were not flexible enough to adapt to changing needs and times. Most required dedicated single-user workstations, and the major development environments were very expensive; knowledge engineers were a scarce commodity, and the time required for subject matter experts and other personnel was an even more expensive contribution. All of this has changed except for the expense involved for the experts, but that has been reduced because these systems can be developed much more quickly. The change started with the appearance of simple and cheap expert system shells on PCs, and the emergence of software which allowed employees to build small, mundane but effective applications to help in their own jobs.

Although some believe that expert systems have disappeared, quite the opposite is true. Expert systems are usually embedded unnoticed within applications and several products have been embedded as part of special-purpose products. Technological developments have made it possible to build expert systems that are affordable, sleek and simple to use. Applications have also been developed that reside on web-based servers and interact with users via handheld devices. Knowledge-bases have also been embedded with small-imprint inference engines that operate in real-time as part of a device.

Expert systems can either be bought ‘off the shelf’, or they can be custom-made, to meet the client’s needs and specifications. They can be as simple or sophisticated as the situation demands and budgets allow for. They can be of general application, or relevant to a particular location or situation. The expert advice they contain can be either ‘in-house’ expertise, or it can be acquired from an external source.

Expert systems can even be built by the clients themselves, assuming they have acquired the program and technique to develop the rules governing the system. Expert systems can also be conveniently linked to existing databases or other, external applications, such as oil properties databases, spill trajectory software, oil weathering models, sensitivity atlases and electronic contingency plans. Marine oil spill expert systems will not supplant or replace these applications. Instead, a web-based system can either create a link to them, or use their information or results directly. In short, modern knowledge-based applications are rarely just expert systems; they are a coordinated symphony of many technologies.

2 Some Common Applications of Expert Systems

Expert systems are used in a wide variety of industries and sectors, including oil spill response, and they can be as simple or complicated as the situation demands. They are particularly effective at preserving the knowledge of an employee who has some specialist knowledge which risks being lost upon retirement, or change of job perhaps.

Another common application is troubleshooting. Thus, if you have ever had a problem with your computer printer, then chances are you have used an expert system (whether you knew it or not). Typically, the computer asks you to pick from a list of symptoms commonly experienced, such as “Won’t print” or “Paper Jam”, and then takes you through a type of decision tree. You are asked a series of questions, and your response will determine which screens pop up next, until, hopefully, the problem has been isolated and eventually solved before all the various options have been exhausted.

Auto mechanics also use expert systems to diagnose what is wrong with a car that has been brought in to the shop because of some sort of mechanical breakdown. Similar techniques are used for complex tasks, such as diagnosing and repairing helicopters, planes and factory equipment. Also, expert systems are frequently used in the medical field, particularly with respect to interpreting laboratory results.

Expert systems are well-suited for “configuration problems” in which a satisfactory solution is composed from many alternate, but compatible, local decisions. Sometimes a second step is applied to select the best solution from the set of satisfactory ones. Probably the best example of this was R1/XCON, which guided the assembly of Digital Equipment Corporation's VAX computers.

There are many combinatorial problems, simple in nature but complex in execution, which can benefit from a similar approach. Filtering, or screening applications are also popular uses of expert systems, whether to determine who is eligible for a bank loan, insurance policy, admission to graduate school, or worthy of special attention in the operating room. Correct sequence and application of procedures is another useful application of expert systems to ensure consistent, high-quality delivery of services. Corporate or organizational policy evolves over time and is often difficult to interpret for application to a particular situation. Expert systems can translate dry policy documents into live, effective action.

In short, expert systems have a broad application in many different fields, and are used by experts and non-experts alike. As the title of the following book would suggest– Beyond the Internet: How Expert Systems Will Truly Transform Business (Smith, 2001) would suggest, expert systems are predicted by one author at least to have a brilliant future ahead of them.

3 Some Features of Expert Systems

Expert systems can be beneficial in many ways. In contrast to a human expert, for instance, who may not be available when needed, and who eventually has to get some sleep, an expert system can be available all the time. Also, an expert system’s performance is constant, whereas human performance typically drops off after x number of hours. Moreover, an expert system can be accessible to multiple users at the same time, provided each of them has access to a computer.

A major feature of expert systems is that they can speed up decision-making. This is particularly true when they are relied upon to solve complicated problems, involving a significant amount of data or calculation. They can also make the decision-making process transparent. An example of this would be where an employee is asked to explain the rationale for a certain course of action, whether before or after the decision has been taken. He or she can walk management through the expert system, explaining the logical process by which a decision was eventually arrived at. A report can also be printed out which summarises the facts surrounding the incident in question and the recommended course of action.

By the same token, that same official could be asked to explain why he or she chose a certain course of action that ran counter to the recommendation advanced by the expert system. In this connection, there is a fear in some quarters that the expert system leaves a potentially incriminating ‘paper trail’ that could conceivably end up as evidence in court. In reality, though, there will always be a ‘paper trail’, whether or not expert systems are used. There is also a fear that expert systems could have a ‘chilling effect’ on responders, discouraging them from using their own judgment. In the final analysis, however, few, if any, expert systems will force someone into a course of conduct they don’t want to follow. The system, in other words, can normally be overridden; this can help identify the difficult decision points where the existing system needs improvement. In this respect, the situation is somewhat analogous to the ‘cruise control’ feature on some car models.

A further feature of expert systems is that they can ‘nudge’ a decision-maker to consider certain elements in the course of making a decision, things that he or she might not have considered otherwise.

One limitation of these systems is that some knowledge and expertise defies automation. This is proof enough that expert systems are no substitute for a human expert. In some instances, such systems will simply remind an expert what he or she knows already, but has forgotten, perhaps in the ‘heat of battle’.

4 Expert systems: Love’em, Hate’em

Not everyone appreciates expert systems; some people think they are wonderful, while others distrust them, to say the least. Often, though, people have mixed opinions about them. In other words, even those who like them don’t think they’re perfect, and those who distrust them admit that can be useful in some cases. This is only natural, especially with a relatively new technology. Part of the ambivalence can be explained by a misunderstanding as to what expert systems are, and what they seek to accomplish. Some specialists, for instance, see them as robots that are going to do them out of a job. Some consultants appear reluctant to contribute their experience to the knowledge base, which they view as a potential competitor. This view is short-sighted: at the time of a spill, experts themselves can benefit from expert systems and the knowledge they contain, assuming they have access to the knowledge base. They can also use the system to guide local officials through the logic behind the advice they are giving the client.

For expert systems to make inroads in decision-making, a lingering distrust of computers will have to be overcome in some quarters. Not all people who dislike computers are innumerate or technophobes; some people, especially those who work in the field, just don’t want to be burdened by ‘gadgetry’. Such people find computers to be an unreliable machine whose battery invariably runs out when it is most needed, or an expensive toy that gets short-circuited in a downpour. These same people, however, could benefit from the advice of experts, perhaps operating in the ‘back rooms’, who do use computers, and who can and do relay advice to the people on the front lines. Computers are likely to gain increasing acceptance as the younger, techno-savvy generation replaces the old guard.

5 How Spill Response Lends Itself to Expert Systems (and vice versa)

The International Tanker Owners Pollution Federation calculates that international technical experts have responded to more than four hundred and fifty ship-source spills in a total of eighty-five countries in the course of the past thirty years (International Tanker Owners Pollution Federation Limited, 2003). In addition, a significant number of government experts, spill cooperative experts and private sector consultants provide spill response advice. Considerable experience has been gained in the course of responding to these and other incidents. Valuable lessons have been learned (Salt, 2002). These lessons have been turned into ‘Rules of thumb’ (Owens, 1999). It is the contention in this paper that these ‘rules’ can be turned into expert systems.

Indeed, proprietary expert systems have been developed for marine oil spill response decision-making, such as the selection of the most appropriate shoreline cleanup and treatment methods (Lamarche et al., 1995). Marine oil spill expert systems show promise in other areas as well, such as in determining the viability of offshore response options, including mechanical recovery, in situ burning and the use of dispersants. A prototype expert system for this purpose is currently available.

Admittedly, experts do not always agree on which response options might be considered; some favour use of dispersants, for instance, while others say they do more harm than good, and advocate other techniques such as bioremediation or in situ burning, or even ‘letting the oil wash ashore’. Not surprisingly, then, on any given spill, the recommended response strategy will probably differ, depending on who you talk to, the local regulations, the availability of equipment, etc. The expert system cannot overcome these deficiencies; all it can do is recommend best practice, or best response, with the requisite caveats attached. The authorities then have a choice vis a vis the advice proffered; they can take it or leave it, or they can adapt it to their own perceived needs.

The advice offered by an expert system does not have to be perfect; the real assessment as to its value will come after the fact, when a review of the operation is undertaken and the question is asked: “Did the knowledge lead to a more efficient response?” Following the advice contained in an expert system can never guarantee a successful outcome. What it can say, however, is that when faced with a given situation, this is probably the best thing you can do, based on previous experience under similar (but not identical) circumstances.

Why, one might ask, is an expert system needed when the experience from previous spills already exists in so many other forms, such as books, manuals, CDs and videos? The answer to this question is twofold:

• The information you require is seldom available when you need it, or in the form in which you need it, or the data is out of date, whereas an expert system can be conveniently, even automatically updated, just like most other computer applications.

• The expert system will actually make a decision for you, based on information you have provided, whereas the other types of information require you, the user, to choose the advice most appropriate to your situation.

Why not just rely on the advice of a human expert when a spill occurs, whether that person is on-the-scene or communicating from some remote location perhaps halfway around the world? For, when a major spill does occur, there will be no shortage of experts offering helpful advice, willy nilly. Most of the advice you will get from that expert is in his or her head; in other words, it has not been formalised or codified into any single source. The stated reason for this practice is that each spill is unique, and no two spills are alike. This is true, but it is also true that advice will be proffered no matter how unique the spill is. An expert system would merely codify what human experts are doing already, which is generalising; plus, it would make the advice available to you on your own time, at your own speed.

There are other advantages to expert systems as well. If the spill occurs in a very remote location, communications links with the outside world may be poor or non-existent. A local On-Scene Commander, possibly with no experience whatsoever, may have to improvise and make snap decisions that could be very costly in more ways than one. An expert system could help the responder determine in advance whether the proposed measures are likely to produce a net environmental benefit.

The absence of appropriate expertise at the scene of a major spill can result in a failed response operation. Sometimes, when a major new spill occurs, it is as if one is starting all over again. The same mistakes are made. The wrong equipment is used, or too much of it ends up being bought and put on standby. Faulty strategies for combating the spill are adopted. Why is this so? Partly it is because catastrophic spills do not happen all that frequently – perhaps every three to four years. Thus, the people who were around to manage the response to the last one may not be around when the next one hits. Another reason for failure is that when a spill occurs there is often considerable political pressure to adopt wasteful, non-scientific response strategies. Vessels and booms are deployed where and when they do not work or are not effective, or simply are not needed. Often, equipment is purchased at great expense and then never even used. There can be tremendous pressure to ‘do something’ or ‘look busy’, especially if the spill has garnered the watchful attention of the media. There is often a tendency to ‘over respond’, just to be on the safe side, or to avoid potential liability problems.

Even if the response option chosen is correct, and the equipment appropriate, the operation may not be conducted according to ‘best practices’; this can lead to a very inefficient allocation of resources. Sometimes crew have received inadequate training, or haven’t been sufficiently drilled. Meanwhile, the responsible party could have been looking at other options, such as use of dispersants, in situ burning and bioremediation, all of which offer some hope of preventing economic and environmental loss. The trick is to take advantage of a typically narrow ‘window of opportunity’ for a particular at-sea response option, before the oil spreads and weathers to the point where the battle is lost before it even begins.

Expert systems can assist in this effort by potentially gaining precious time, by helping to ascertain in advance whether a particular option will or will not work for a particular spill, by rejecting those options which are bound to fail, and by suggesting instead those that are likely to succeed. Admittedly, most, if not all of these calculations can be made ‘off the cuff’. However, spill response is an increasingly complex, multi-disciplinary affair; the different types of crude oil and product alone means that on all but the simplest of spills, few people can be expected to possess all of the specialist knowledge required to mount an effective response operation. Undoubtedly, teams of specialists can supply this expertise, but these can take precious time to assemble, and meanwhile the oil is spreading, increasing in volume, and changing in character. Moreover, each spill will require its own specific skill sets, and some teams will not possess the knowledge that the spill response demands. An expert system can help to fill these knowledge gaps.

The fact that expert systems use simple rules does not mean they are incapable of handling complex oil expertise. Complex decisions can be composed of simple components, and more complex mechanisms are available in the inference engines of most expert system tools, e.g. preferences, biases, agendas and other conflict resolution strategies. The systems can also be augmented by other software development tools.

Experience has shown that there is a more or less direct correlation between response time, effort and success. That is to say, generally speaking, the quicker a credible response can be mounted, the more successful the operation is likely to be. Conversely, the slower the response, the greater the effort that will be required, even with a low success rate. As with fighting forest fires, speed is a key ingredient in successful oil spill response; the sooner one can stop oil from leaking out of a vessel, from spreading on the ocean, or from impacting sensitive resources, the less the environmental damage is likely to be. If the spill can be quickly contained, response costs could also be lower, and cleanup costs should definitely be lower. The relationship between time and cost also makes it all the more important to ‘get it right the first time’, choosing the right combination of response options. To do otherwise is to waste precious time (and money).

In addition to being useful in an operational setting, expert systems could be used for several other purposes. For example, they could serve as a planning tool, to indicate the risks or implications of adopting a particular response option or strategy. Expert systems could also be used to train marine oil spill responders, or indeed anyone connected with marine spill response operations. They could also be used as an educational tool, available free of charge to the public at large via the World Wide Web.

In conclusion, one must recognize that in times of crisis even an expert can fumble and make mistakes. An expert system can act as a reminder to the responder, or it can offer a second opinion. The responder is not bound by such advice, but it could prove useful to him or her in the heat of battle. The bottom line is that if the expertise exists, then it can usually be captured, stored and used on a computer.

6 A Case Study: How an Expert System Might Have Improved Response

On or about 26 December, 2002, an oil spill occurred in Yap State, Micronesia, involving the non-tank vessel Kyowa Violet. An undetermined amount of oil spilled in or around reef waters, shoreline and mangroves. An unspecified area was reportedly contaminated with oil.

Sometime between the time of the spill and December 31, 2002 at 6:08 AM, a request was forwarded by the responsible parties to the Yap State Environmental Protection Agency (EPA) to use a commercial brand of dispersant on the spilled oil. It was a this point, i.e. approximately five days after the spill was initially reported, that an official with the Yap State EPA posted a message on the Yahoo Oil Spill Responders Discussion Group web site (Yahoo Oil Spill Responders Discussion Group, 2002) asking for advice regarding the dispersant in question, including its use and effects.

Over the course of the next seven days, i.e. between December 31 and January 6, a total of six replies to the original enquiry were posted on the Yahoo website. These ‘helpful’ tips came from different parts of the world: four from the United States, one from New Zealand and one from Saudi Arabia. The first message suggested that the dispersant in question was approved for use in several countries, which is undoubtedly true, but begged the question as to whether the product was registered for use in Micronesia. The second message stated that the product was not listed for use in the United States; Micronesia is a protectorate of the United States. The third message urged caution with respect to the use of any and all dispersants, while the fourth recommended a net environmental benefit analysis be undertaken. The fifth asked for some basic information as to such things as TDS of seawater, current temperature and crude oil API. Finally, the sixth response suggested another commercial dispersant for possible use.

What is one to make of this discussion, which started five days after the spill began, and terminated six days later? First, few details are available regarding the specifics of the spill. Thus, it is impossible to say whether the spill was an instantaneous, batch spill, or rather a continuous release; the answer to this question could affect the efficacy of the dispersant application. It is also unknown as to whether the initial query was answered by some medium other than the discussion group; it is quite possible, for instance, that someone simply picked up the phone and told the enquirer directly what to do concerning the request for permission to use the dispersant. Nevertheless, it seems clear that the enquirer, and the responsible authorities, could have benefited from a speedier process for gathering expertise; eleven days is an ‘eternity’ in marine oil spill circles! Second, the advice given would doubtless have carried far more weight if it had been consistent and from a single, authoritative source. Selection guides for dispersant use exist in various countries. The knowledge contained in these guides provides excellent material for web-based expert systems that would be available to the user at any time of the day or night, including Boxing Day.

7 Summary and Conclusion

Expert systems represent a widely-used, proven technology. They are an excellent vehicle for capture and storage of knowledge on the computer, for rapid retrieval whenever and wherever it is needed. They are particularly useful where time is of the essence, and decision-making is complex. Small, discreet, user-friendly systems can now be developed relatively easily and inexpensively, and readily updated. Expert systems have been and are being used as a decision support tool for marine oil spill response operations. However, there potential for application in this area remains largely untapped, partly because they are misunderstood.

The marine oil spill response community is very international by nature; with more than two thirds of the world’s oil supply plying the worldocean, a spill could occur virtually anywhere, anytime. Advances in information technology permit almost instantaneous retrieval of information and knowledge that can be crucial in terms of formulating a response. The use of computers is now all-pervasive: their compact size, increased battery length, high speed and ability to store and process large amounts of data make them an indispensable tool on location. They can also be operated off-site, with information being sent in from the spill site to the place where the computer is located.

Expert systems could be a potent, cost-effective weapon in the arsenal of response tools. Spill cleanup costs can easily run into the hundreds of millions of dollars. Some of this expenditure ends up being wasted or not recovered because it was neither reasonable nor justifiable. Hundreds of millions of dollars have been invested in spill research and development in the course of the past thirty years. Much of this has gone into the testing and development of hardware, such as booms, skimmers, absorbents, dispersants, as well as a variety of response technologies, such as remote sensing, burning and bioremediation. Comparatively little time and effort has been devoted to improving the decision-making part of the equation, the Achilles heel of spill response. It is perhaps time to take greater advantage of these tools.

8 References

International Tanker Owners Pollution Federation Limited, ITOPF Handbook 2003/2004, 2003, 48 p.

Lamarche, A., C. Black, A-P Varanda, and E. H. Owens, “The Use of Knowledge-Based Software to Identify Shore Line Treatment Options”, in Proceedings International Oil Spill Conference, American Petroleum Institute, Washington, DC, pp. 55-60, 1995.

Owens, E.H., “Practical Guidelines or ‘Rules of Thumb’ for Spill Response Activities”. Proceedings of the Twenty Second Arctic and Marine Oilspill Programme (AMOP) Technical Seminar, Environment Canada, Ottawa, ON, pp. 695-704, 1999.

Salt, D., “Learning the Lessons of History”, Spillcon 2002, Sydney, Australia, 3 p. http://www.spillcon.com/2002/Papers/final/Salt,D.pdf

Smith, L., Beyond the Internet: How Expert Systems Will Truly Transform Business, Stoddart Publishing Company Ltd; Toronto, ON, and Niagara Falls, NY, 246 p., 2001.

Yahoo Oil Spill Discussion Group, Dec. 31, 2002 through January 6, 2003, http://groups.yahoo.com/group/oil-spill-responders/message/346, plus messages 349, 350, 351, 352, 353, 354.