By Michael Guth⁽¹⁾

¶ 1 During the week of June 22 - June 26, 1998, many states in the East and Midwest experienced unusually hot temperatures, and electricity prices skyrocketed. Temperatures were in the upper 90s to 100s from Nevada to South Carolina. In the Southeast, hourly and next day electricity prices rose to more than $5,000/MWh during certain hours. These prices were an order of magnitude 100 - 200 times greater than the $30-$50/MWh average price in previous summers.⁽²⁾ Prices exceeded $7,500/MWh in the Midwest, which experienced an acute shortage of supply due to transmission line unavailability.⁽³⁾ Utilities in the Midwest asked residential and industrial consumers to conserve power.⁽⁴⁾ At one point during the heat wave and power supply crisis, it seemed that Wisconsin and Ohio would experience black outs or brown outs.⁽⁵⁾ Some power marketers defaulted on their power sales contracts.⁽⁶⁾ Those who defaulted will either have to pay contractual damages or seek bankruptcy protection.⁽⁷⁾ The high volatility of electricity prices and scarce supplies during peak consumption periods has focused attention on utilities' contingent plans to curtail power when required.

¶ 2 Hot temperatures and high demand are not the only reasons utilities curtail power. Electric utilities sometimes interrupt or curtail power on a sales contract due to an unforeseen loss of generating capacity or transmission lines. But in these circumstances, the utilities' marketing and trading staff will attempt to mitigate disruptions and continue with the electricity sale.⁽⁸⁾ For example, suppose the transaction involved the sale of a 100 MW flow of power during the 21 business days of a given month, and the supplier was unable to meet its obligation under the contract on two business days. The transaction could be restructured in several different ways.

¶ 3 First, the parties might agree to make up the difference by extending the contract beyond the end of the month and into the first two business days of the next month. Second, the supplier might agree to sell the supplier some larger quantity of weekend or off-peak power that would equal the value of 100 MW during the two business days. Third, if the purchaser was able to buy the electricity on the open market at a price equal to or less than the contract price, the parties might agree to adjust the total amount due under the contract but otherwise leave its provisions in tact.

¶ 4 The deal could be restructured in many other ways, limited only by the parties' creativity. In order to mitigate damages and restructure the transaction, the business staff must understand the legal consequences of various terms concerning power disruptions in their sales contracts.

¶ 5 Legally, utilities may curtail so-called nonfirm power without incurring liability for damages. In contrast, sales of firm power imply the seller will pay liquidated damages in the event of an inability to supply. Usually the liquidated damages will cover the cost of purchasing electricity, or reselling it in the event of a buyer's default, on the open market. The U.S. Supreme Court described the distinction between firm and nonfirm power as follows:

¶ 6 Because the amount of power generated by BPA [the Bonneville Power Administration] depends on streamflow in the Columbia River system, BPA cannot predict with accuracy the amount of power that it can generate. Accordingly, BPA historically has sold two types of power. "Firm power" is energy that BPA expects to produce under predictable streamflow conditions. "Nonfirm" power is energy in excess of firm power, and is provided only when such excess exists.⁽⁹⁾

¶ 7 At present, utilities purchase both firm and nonfirm power from each other. The utilities plan for possible interruptions in supply of the nonfirm product by having one or more of their generating plants on standby status. But even with the best of plans, supplies of firm power will sometimes become unavailable and lead to a breach of a power sales contract.

¶ 8 In principle, the rules used to determine liability for breaches in electricity sales contracts could be programmed into an expert system,⁽¹⁰⁾ which would advise end users about possible consequences of a power disruption. This article describes the development of such a system.

¶ 9 The expert system takes the user through approximately twenty provisions pertaining to energy curtailment drawn from electricity sales contracts. Each of these provisions comprises either an actual breach or a potential breach of a power sales contract. The expert system offers some insights into both the buyer's and the seller's rights and responsibilities.

¶ 10 The expert system was designed to answer business staff questions about electricity curtailment and thereby may reduce the volume of inquiries directed to a utility's in-house counsel. The expert system's knowledge base contained sufficient rules to permit the system to analyze conditions and draw a legal conclusion. The system then conveyed this conclusion in an easily understood format to end users who typically have years of industry experience but no legal training.

¶ 11 Another goal of this research was to examine the suitability of this subject matter as a domain for expert systems.⁽¹¹⁾ Unlike other areas of the law where the knowledge base could draw upon statutes, regulations, and case law for its rules and fact patterns, the business transactions conducted by electric utilities have practically no case law and no statutory law for guidance.

¶ 12 The lack of case law stems in part from the utilities' former role as heavily regulated monopolies in their respective geographic areas. The utilities' transactions were subject to review by state and federal regulators, who ensured that electricity was continuously supplied to the public despite curtailments, disruptions, or other outages that the utilities experienced. If a contract breach forced a utility to purchase electricity at a higher price from another vendor, these costs could generally be passed on to rate payers as part of the ordinary costs of doing business. Few disputes concerning the terms of electricity sales contracts and the effect of a breach ever rose to the point of litigation.

¶ 13 The lack of litigation on power sales in the formerly regulated power industry can be seen in the results of an on-line legal database search of all state and federal cases using the keywords "electric util!", "sale!", and "breach" in the same paragraph. From 1945 to 1975, no cases turned up in the search. In 1976, one case turned up.⁽¹²⁾ The next case occurred thirteen years later in 1989. In all, this search yielded approximately 12 power sales cases out of scores of thousands of power transactions over the period 1945 - 1997.⁽¹³⁾ Admittedly, this search might have failed to pick up all cases relating to power sales disputes due to the keywords used in the search,⁽¹⁴⁾ but it does reflect the reality that regulated electricity sales were seldom the object of litigation.

¶ 14 In the absence of statutory and case law guidelines,⁽¹⁵⁾ utilities apply standard principles of contract law to formulate terms for their power sales contracts. Their transactions also follow utility industry norms and practices, which meant an expert system developer (also known as a knowledge engineer) had to learn these unwritten norms and standards before he could understand the physical and legal consequences of a curtailment.

¶ 15 The expert system was designed to function as a stand-alone system or as a subroutine of a large system. The system was encoded in the expert system shell Natural Language Expert System Builder (NLESB),⁽¹⁶⁾ which was chosen for the following reasons:

¶ 16 The program files are sufficiently small to load onto a laptop or other computer with limited memory.

¶ 17 The compiled, executable code requires no supporting software or licensing agreements.

¶ 18 Multiple databases can be easily accessed, manipulated, and maintained.

¶ 19 The program would accept propositions, rules, and notes written in ordinary English, and even accept verbatim excerpts from the contracts when desired.

¶ 20 Unlike other expert system shells, NLESB requires that propositions entered into the system be written in a "normalized" form.⁽¹⁷⁾ Therefore, the first step in the expert system development was to recast the contract excerpts into the appropriate form. The normalized form places ordinary text into an outline in which the causal relationships between propositions are highlighted and clarified through the use of logical operators such as IF/THEN, OR, AND, NOT, and IFF (If and only if). To transform ordinary text this way, first the Boolean algebra operators (AND, OR, NOT) and logical expressions (IF, THEN, IFF) should be capitalized as they appear in the text, and propositions that attach to those expressions should be set apart using some convenient notation. Highlighting key terms this way will aid the knowledge engineer in finding the causal relationships between propositions. For example, the contract excerpt,

¶ 21 Each arrangement, if any, that seller makes to resell energy during a Breach Period may not exceed 48 hours in duration, unless buyer indicates that it will not resume taking energy after the 48 hour period has expired.

¶ 22 would be initially edited as follows:

¶ 23 IF | [strikeout: Each arrangement, if any, that] [the] seller [arranges] to resell energy during a Breach Period [caused by the buyer] | AND | [strikeout: may not exceed 48 hours in duration, unless] [the] buyer [has NOT] indicate[d] that [the buyer] will NOT resume taking energy after the 48 hour period has expired | THEN | [the seller's arrangement to resell energy may NOT exceed 48 hours in duration.]

¶ 24 This edited excerpt is then placed into outline form using particular spacing, indentation, and numerical conventions⁽¹⁸⁾ of normalization as follows:

¶ 25 RULE 15

IF

(1) the seller arranges to resell energy during a Breach Period caused by the buyer, AND

(2) the buyer has NOT indicated that the buyer will NOT resume taking energy after the 48 hour period has expired,

THEN

(3) the seller's arrangement to resell energy may NOT exceed 48 hours in duration.

¶ 26 This rule is listed as Rule 15 in the Appendix. Several comments are now required about Rule 15. Scholars in the artificial intelligence and law field have found that normalization often eliminates certain logical ambiguities in the wording of statutes or legal documents that the parties never intended.⁽¹⁹⁾ For example, what happens under Rule 15 if the buyer does indeed indicate that it will not resume taking electricity after the 48-hour Breach Period? The contract is silent on this condition. However, industry norms and practices say that the seller may arrange, obviously, to resell the electricity for more than 48-hours duration.

¶ 27 To clarify the consequence of the buyer's decision not to resume purchasing after 48-hours, an additional rule was added to the knowledge base:

¶ 28 Rule 17

IF

(1) the seller arranges to resell energy during a Breach Period caused by the buyer, AND

(2) the buyer has indicated that the buyer will NOT resume taking energy after the 48 hour period has expired,

THEN

(3) the seller's arrangement to resell energy may exceed 48 hours in duration.

¶ 29 A rule added by the knowledge engineer, but not contained in the original source material, is called a supplemental rule. Humans with years of industry experience would know how to interpret the contract clause relating to the 48-hour limitations on resale arrangements, but the expert system had no way of knowing the consequence of the buyer's decision not to accept delivery until Rule 17 was added.

¶ 30 Furthermore, the contract terms and Rule 15 only applied to a breach caused by the buyer. An ambiguity remained as to the buyer's rights and responsibilities for a breach caused by the seller. It turns out the buyer would not face a 48-hour limitation on his alternative arrangements, as explained in supplementary Rule 16.

¶ 31 Rule 16

IF

(1) the buyer arranges to purchase energy during a Breach Period caused by the seller, AND

(2) the seller has NOT indicated that the seller will NOT resume supplying energy after the 48 hour period has expired,

THEN

(3) the buyer's arrangement to purchase energy may NOT exceed the agreed upon damages for a seller breach.

¶ 32 The buyer faces a limitation based on the dollar amount of the agreed upon damages, but not a time-duration constraint. Some people find normalization makes statutes and other legal writing, such as power sales contracts, easier to understand.⁽²⁰⁾

¶ 33 The supplementary rules emerging from Rule 15 merely underscore the fact that the learning process in building an expert system is not confined to a consultation of the system by an end user. The domain experts learn about new ambiguities in an area of the law already familiar to them, after they see logical syntax rigorously applied to the causal relationships in these sales contracts.

¶ 34 Finally, a good English grammar teacher would want to eliminate the double negative contained in condition (2) of Rule 15. The grammar teacher might reword the sentence to read "the buyer has indicated that the buyer will resume taking energy after the 48 hour period has expired." However, diction and syntax matters. The revised wording has placed the burden on the buyer to reaffirm that it will accept delivery of power, rather than correctly placing that burden on the seller to obtain that decision from the buyer before it makes an arrangement to resell power longer than 48-hours duration.

¶ 35 Supplementary rules were also used to express ongoing duties by the contracting parties. For example, Rule 14 indicates that the buyer and seller will consult with each other about the duration of any breach period. That duty to consult applies to the parties regardless of who (buyer or seller) or what (force majeure event, negligence, skyrocketing prices) caused the breach to occur. It looked awkward to repeat the duty to consult as a consequence in multiple rules that mentioned a breach. Usually, if a consequence must be repeated in many rules, then a supplementary rule is needed.

¶ 36 In addition, it would be both logically and legally incorrect to include that ongoing duty in only one or two rules in the knowledge base. For example, if the duty to consult were only listed for Rule 15, an end user might erroneously infer that he or she had a duty to consult with the counterparty only when the conditions in the antecedent portion of Rule 15 were met. Yet the end user could have a duty to consult even if the premises of Rule 15 were not met. The simplest way to record this very broad duty to consult in the expert system was to create (supplementary) Rule 14, which is triggered by the occurrence of any Breach Period.⁽²¹⁾

¶ 37 A Breach Period occurs whenever a party causes a breach in the contract. Some of the rules in the knowledge base use the phrase "Breach Period," while other rules might use a similar phrase such as "the buyer causes a breach." The expert system had to inform an end user that these two phrases had the same meaning. In NLESB, this process of determining equivalence relations among propositions and rules in the knowledge base is known as unification. NLESB assists the knowledge engineer with the unification task by providing lists of possible propositions that match a preset percentage of words for a given proposition.

¶ 38 Unifying two propositions provides the expert system with an extra piece of information about two or more separate propositions; however, unification does not mean that separate rules would be combined into one. For example, the consequence of Rule 10 is essentially the same as the antecedent proposition for Rule 9. Thus, these rules illustrate a causal pattern of the form "If Fact 1, then Fact 2" and "If Fact 2, then Fact 3." Should the two rules be combined into one, such as "If Fact 1, then Fact 2 AND Fact 3"? In general, two such rules with overlapping content should not be combined, because the combined rule masks the causal connection between Fact 2 and Fact 3, which may be important in other contexts. That connection could become important for other rules or statutes not yet included in the knowledge base and may affect the timing of consequences. Therefore, I kept Rules 9 and 10 separate, but included an equivalence statement that Fact 2 in Rule 9 was the same as Fact 2 in Rule 10.

¶ 39 A knowledge engineer will find that some of the equivalence relations and supplementary rules derive from extra conditions that describe what happens if various rights/privileges/duties are invoked or exercised. Another technique that will often lead to new supplementary rules is posing hypotheticals and discovering that the expert system could not solve them. We saw an example of that when a hypothetical question about the duty to consult with a counterparty was posed in relation to Rule 15. Supplemental Rule 14 was added to cover the ongoing duties of the parties to consult. Similarly, once Rule 15 was coded into the system, a natural hypothetical question would be "what if the seller, instead of the buyer, is in breach?" Answering that hypothetical led to the creation of Rule 16 on the seller's liabilities for breach.

¶ 40 Adding supplementary rules, placing text in normalized form, using supplementary rules to fill the gaps in contract language, and the process of noting equivalence relations or unification are all techniques internal to the firm that constructed the expert system. These supplementary rules and information on equivalence relations do not form part of the contract for the sale of power. Thus the counterparty need not agree to or even be aware that the firm chooses to analyze the contract this way. These steps were taken solely to help the firm understand its legal rights and liabilities in curtailment situations, taking the actual wording of the contracts as given. Thus, changes in the expert system knowledge base are akin to changes in the in-house counsel's work product. They are not meant to be legal modifications to existing contracts but instead to be techniques for understanding those contracts better.

¶ 41 To construct rules that bridged the gap, where necessary, between two other rules in the knowledge base, the knowledge engineer had to know something about the background of the users and what questions they would likely pose to the system. In formulating their questions for the expert system, the end users of this system were assumed to know the difference between various defined terms and electricity products, e.g., "non-firm energy" versus "firm energy."⁽²²⁾

¶ 42 Similarly, one of the contract excerpts contains the proposition "the failure was not excused by Uncontrollable Forces." A user unfamiliar with that defined term might believe "uncontrollable forces" were limited to tornadoes, floods, and other events beyond the control of man. However, an end user working in the utility industry would interpret that term to include events beyond the control of a party, such as a break in a transmission line or even a decision by one utility to deny another utility use of its transmission lines. I assumed that the end users had several years of utility industry experience and knew terminology commonly employed in power sales contracts.

¶ 43 Finally, the software was tested with end users. The end users (business and marketing staff) are generally not domain experts, so their suggestions to change the wording of propositions or the logical structure of the rules must be verified with the domain expert (lawyers). Assuming the domain expert agrees with the change, then the knowledge engineer should implement the change to increase the usefulness of the expert system.

¶ 44 In summary, the process of building an expert system for legal reasoning involved several discrete steps, which can be enumerated as follows: (1) select the domain, goals, and purpose for the application; (2) obtain the relevant statutes, regulations, contract terms, or other legal material to serve as the source of rules for the knowledge base; (3) edit the legal source to highlight connections in logic and causal relationships; (4) recast the edited text into the normalized form and load the rules into the expert system; (5) decide on the intended end users of the system and what can be assumed about the end users; (6) check for equivalence relations between the propositions in the knowledge base; (7) ask end users to test the system, and record their suggestions to improve the system; and (8) modify the knowledge base through several iterations designed to clarify the content of the rules and eliminate extraneous rules that are never used.

¶ 45 The rules in the knowledge base, which are listed in the Appendix, cover many different facets of a power sales transaction. Some rules establish conditions under which the supply of power shall be deemed "unreliable," and thus the buyer then has the right to terminate the contract. The original contractual language called for the supplier to be deemed "unreliable." Yet two in-house counsel at my utility client requested that the conclusion be reworded to show that the supply, and not the supplier, would be deemed unreliable.

¶ 46 Other rules indicate that supply at one regional border delivery point may be substituted for supply at another, provided that the utility agrees to accept delivery anywhere on its regional border. Such rules give the supplier some flexibility in choosing delivery points for its supply, which might vary during the term of the contract.

¶ 47 Some rules impose asymmetric rights and responsibilities during a breach. For example, under one contract form, when any disruption occurs, the seller still has the legal obligation to supply the buyer up to the maximum amount of the contract, but the buyer's obligation to purchase may depend on the nature of the disruption. These contract terms could have been written to give the supplier an equal amount of flexibility. They represent part of the bargaining at arm's length for the terms of the power sales contract. As a general rule, regardless of who causes a breach in the contract, the buyer and the seller must still consult with one another to assist the nondefaulting party in mitigating its damages.

¶ 48 Other rules cover situations when a seller intends to perform under the contract by purchasing electricity from another vendor, rather than generating the electricity itself. If the seller's purchase is reduced or interrupted, then the interruption may be an excused event beyond the seller's control. In that case, the seller may in turn curtail or reduce, as the case may be, the amount of electricity supplied to the buyer.

¶ 49 Using the contract excerpts contained in our knowledge base, the buyer bears the risk of any reduction or interruption in the supply. In principle, the seller could bear that risk. However, the seller has bargained in this contract to include a provision that allows it to shift the risk of non-supply to the buyer, which presumably receives some type of price incentive for this nonfirm energy product. If the buyer fails to perform, and this failure is not excused by events beyond the control of the buyer, then the buyer must pay the seller the difference between the contract sale price and the price at which the seller was able to resell the energy. This consequence is a standard remedy in breach of contract cases. Similarly, in firm power sales when the parties have not agreed in advance to a liquidated damages provision, the buyer's damages will be limited to various additional costs it incurred as a result of having to purchase electricity from an alternative source.

¶ 50 Finally, the expert system covers contract language in which the defaulting party has the right to request an independent auditing agency verify the nondefaulting party's claims for damages. This audit verification feature is merely one of many terms in the contract over which the parties bargained and reached agreement.

¶ 51 Users begin a consultation with the expert system by asking questions (1) in standard (unspecified) form, (2) as a "how" question, and (3) as a "what-if" question. In this latter option, NLESB proceeds with "forward-chaining logic" to search antecedents of rules that share propositions in common with the user's "what-if" scenario.⁽²³⁾ NLESB also provides an index function, which lists key words in the knowledge base and indicates which rules contain each word.

¶ 52 At the start of a consultation, an end user can also enter facts into the system. These facts represent information that particular propositions in the knowledge base are true or false. The system also obtains facts in the form of answers to questions in previous consultations. Once these facts have been entered, the system will ask the user in subsequent consultations whether the system should continue to assume the facts from the previous session are still correct for the current session.

¶ 53 The knowledge engineer can include notes that define terms or explain the content of rules or propositions. For example, notes were written to explain that "Points of Delivery" was a defined term within the contract. Similarly, notes clarify that neither the breaching nor the non-breaching party would be bound by the results of the independent auditor's examination of costs. Regardless of the auditing agency's findings, both parties to the contract retained their rights and defenses with respect to claims for breach of contract. The Notes feature is also a convenient place to store information on the statutory or regulatory citations that might provide some additional information on the rules in the knowledge base.

¶ 54 As to lessons learned, English is an imprecise language compared to the mathematical operators and symbols of most programming languages. When an end user inputs information in ordinary, conversational English, many expert systems will have difficulty interpreting the input. The end user may believe that he or she has asked a precise question, e.g., "What are my options when a counterparty fails to supply energy as promised?" However, the expert system is looking for particular subjects, e.g., buyer or seller, and specific actions, e.g., breach or curtail. None of these words appear in the question posed by the end user in this example. As the expert system tried to clarify what the user had written, the user can become annoyed with the system that seems to ask a series of stupid questions such as, "Is the phrase `fails to supply' related to any of the following subjects? ...." To the extent that the end users and the knowledge base use a common terminology, then the precision problem is alleviated.

¶ 55 The fact that an expert system and the end users sometimes used different terminology merely illustrates a more "general problem with the rule-based approach: it does not give enough prominence to the adversarial nature of legal reasoning, where the opposing sides seek to establish different, and often contradictory, conclusions. Perhaps more fundamentally, the rule-based approach assumes that the set of rules has no inherent difficulties, like ambiguities, gaps, and conflicts. To make a rule-based system work, the programmer must usually eliminate these problems and make the rules appear more consistent and complete than they are."⁽²⁴⁾

¶ 56 A second lesson learned is that knowledge engineers should avoid using the passive voice in constructing rules for the knowledge base. The passive voice permits the subject of the sentence to remain hidden, yet the expert system is generally searching for this subject and trying to match it with other facts and conditions it has received.

¶ 57 Third, power sales contracts form a suitable domain for an expert system. The contracts and other legal documents pertaining to a transaction contain the rules for the knowledge base. These rules must be interpreted in a manner consistent with business trade and practices, because these business practices play a more important role in the law governing transactions than do statutes, regulations, and case law.

¶ 58 In the next development phase, the expert system will contain deeper information in a hierarchical structure. At the first level, the system will prompt the user for the type of electricity product: Firm Energy, Nonfirm Energy, or Capacity. Distinguishing between the types of energy product sold will enable the expert system to eliminate rules that apply to other products. In terms of the order for curtailing power, purchasers of nonfirm energy would be cut first. Then would come purchasers of firm energy. Finally, the purchasers of capacity can be curtailed only when the transmission line or the generating plant itself, to which the capacity rights attach, is involuntarily shut down.

¶ 59 The second level of the new program will be the period of supply, e.g., one month, one day, or two years. The consequences of curtailing power on two days in a two-year contract may be less severe than curtailing power on two days of a three-day contract. The parties would be expected to have detailed knowledge of plant availability for a contract covering a sale two or three days from the present. However, a two-year transaction beginning at some distant date in the future would have much more inherent uncertainty and risks, which one of the parties would have to assume. Third, the user will furnish information about the price (fixed or variable), the points of delivery, and any scheduling constraints written into the contract. These details will again permit an expert system to remove from consideration rules that are specific to other delivery points, or rules that apply to fixed-price transactions if the sales price is defined as a variable index.

¶ 60 Fourth, the system will seek information on preferences in the event of a curtailment. Usually, a utility has a hierarchy of customers. Beyond this hierarchy, a utility will tend to prefer longer duration customers over short term customers. However, a short-term customer with a significant market share might be elevated on its preference list over a long-term customer with a small market share. Fifth, the system will sort through information on transmission line availability: hourly, weekly, and monthly users, firm energy versus nonfirm energy users, and the electrical flows from each interface. The new expert system can perform a valuable service by alerting the business staff to transmission line unavailability. The business staff need not worry about the legal consequences of a proposed deal if it is impossible to execute using the existing transmission line system.

¶ 61 Sixth, and finally, the system will take into account any liquidated damages provision written into the contract. The current expert system already explains the consequences of most types of liquidated damages provisions. However, as transactions become more competitive and structured for individual sales, the counterparties are becoming more creative at identifying conditions that would preclude the payment of liquidated damages or, for example, require that liquidated damages be paid in energy rather than cash. This six-level, multistep process should give the enhanced expert system the necessary pieces of information to advise end users about their options for curtailing supply.

¶ 62 Expert system shells must be designed to retrieve information from a knowledge base efficiently. The shell must obtain the knowledge that the user needs and present it in a readily accessible form. However, the shell should not inundate the user with unwanted information. End users of the proof of principle expert system objected to the way Natural Language Expert System Builder (NLESB) displayed a list of propositions that matched terms in their inquiry. As a general rule, they were impatient to sift through more than three or four propositions to find information that might answer their questions. NLESB began its intelligent inferencing with just the information supplied by one end user question. NLESB needs to elicit more information from the end user at the outset so that it can begin its intelligent inferencing with a stronger foundation.

¶ 63To achieve a more complete description of the user's status and problem at the outset of a consultation, the expert system shell should have an initial screen in which the knowledge engineer defines several key questions and the user supplies answers before a consultation begins. The user's answers would help narrow the applicable rules to certain branches of the logic tree embedded in the knowledge base. For example, for this application, NLESB might ask the users to identify themselves as a buyer or seller or third party, and users would find they had a pull down menu from which to select a response.

¶ 64 As a generic expert system shell, NLESB functions reasonably well. It can accept any material input in normalized form; the subject matter of the expert system could be anything. No knowledge of computer programming languages is required to build an expert system in NLESB. If NLESB provided a screen with text explaining how to start a typical consultation, it would be more immediately accessible. Given the enormity of the task for a piece of software (provide a generic shell that takes natural language text and requires no programming), NLESB appears to have done a good job. The difference between a specific expert system and a generic shell is that the specific expert system often has an interface tailored to that application and the end users' software environment. Most custom-designed expert systems also require programming.

¶ 65 NLESB also imposes a limit on the number of propositions that can be included in any one rule. I would have liked to combine the consequences of several rules, but then the rules would have had two antecedent conditions and ten consequences. Twelve propositions in one rule would have overtaxed this system's structuring capabilities.

¶ 66 NLESB was developed to incorporate rules with up to eight propositions, because empirical studies showed that readers found statutes confusing when they contained more than seven or eight distinct propositions. Overly long statutes with more than seven propositions should generally be shortened to help the reader understand them.

¶ 67 Table 1 lists twenty-nine features that end users have found helpful in an expert system. From this list, NLESB had about 60% of these features. The main advantages of NLESB is ease with which rules can be loaded, the system can be installed, and an application built in a short amount of time. The main disadvantage of NLESB is the lack of developer control over the substance and the order in which information is displayed to the end user. Again, the rules in the knowledge are printed in the Appendix.

¶ 68 Expert systems can assist with legal analysis of power curtailment decisions. Theese systems offer various advantages and disadvantages depending on the types of contracts being analyzed. The more an utility or power marketing firm uses the same wording in its power sales transactions, the more an expert system can save time and money in eliminating routine calls to in-house counsel. Changing terms in power sales contracts requires updating an expert system knowledge base.

¶ 69 At present, traded products in the electricity markets are evolving rapidly. Products types that used to be thinly traded (illiquid), such as options on electricity forward contracts, are becoming more actively traded (liquid) as more firms learn about using derivatives to hedge operational and price risks. The knowledge base of an expert system would have to be updated to reflect the terms associated with a new product type. If the expert system knowledge base contains rules that apply to nonfirm energy, firm energy, and capacity, then the expert system should be able to analyze the effects of curtailing one of these products.

¶ 70 However, a problem with ambiguity remains for expert systems, in general, and those applied to the law, in particular. This article has discussed at least two separate forms of ambiguity. The first is logical ambiguity and deals with the causal relationship between premises and consequences. In Section 3, we explained how the process of placing ordinary legal text into the normalized form often highlights latent ambiguities or implied logical relations in the original text. "Logical ambiguity is present when it is unclear how the concepts are to be combined to form complex propositions and arguments."⁽²⁶⁾

¶ 71 The second major source of ambiguity is the choice of words or legal terminology. As mentioned in Section 3, some contracts may contain the phrase "the buyer causes a breach," while others use the phrase "Breach Period" or "a breach has occurred." Even if the logical ambiguity could be eliminated, some residual ambiguity in diction would remain. "An expert system employing such terms may still operate, but judgment will be required on the part of the user in answering the questions put by the computer."⁽²⁷⁾ Experts systems can be used as a tool to educate business staff and other non-lawyers about the legal consequences of curtailing power, but they are no substitute for lawyers and human judgment.

¶ 72 For that reason, end users would need to confirm the conclusions they draw from an expert system with the utility's in-house counsel. Otherwise, they risk the possibility that they failed to notice a key term in the contract or perhaps answered one of the expert system questions incorrectly. Having confirmed the legal consequences with a lawyer, a business staff member could then proceed under the same type of contract to analyze the legal consequences of curtailing power and recommend strategies without consulting the lawyer each time.

¶ 73 Expert systems will never replace lawyers or allow companies to operate without a full-time legal staff. An expert system can make a mistake if the knowledge engineer has encoded erroneous legal deductions or left in too much logical ambiguity. If a manager relies on the expert system conclusion and it turns out to be wrong, then his or her firm will suffer the same consequences that one might anticipate when a lawyer gives wrong advice.

¶ 74 But the simple fact remains that a person does not need a law degree or a license to practice law in order to know that nonfirm power may generally be curtailed without legal consequences or that a supplier may have to pay liquidated damages if it curtails firm energy. An expert system can help educate the nonlawyer into understanding how legal conclusions are drawn and what key terms in a contract affect liability for curtailment. In the same manner that a lawyer needs to study recent developments to keep abreast of changes in the law, so does the knowledge base of the expert system need periodic updating.

¶ 75 Many lawyers have no idea that expert systems have been applied to the law, yet these applications have existed for many years. A professional society has been formed to study the impact of artificial intelligence methods on the law, and their journal, Artificial Intelligence and Law, presents state-of-the-art modeling designs and descriptions of working expert systems. I am happy to report that the law of power sales contracts appears to be a suitable domain for expert system development, and hopefully this news will encourage others, both in and out of the power industry, to build similar systems for their own transactions.

Rule 11

IF

(1) the aggregate quantity of energy that the seller reduces or interrupts exceeds any amount reduced or interrupted in accordance with the provisions of paragraph 10(b) of this Confirmation in two of the four months of the Supply Period in the same calendar year, AND

(2) (A) the aggregate quantity of energy that the seller reduces or interrupts is NOT otherwise excused under this Agreement, OR

(B) the aggregate quantity of energy that the seller reduces or interrupts is NOT an agreed upon interruption under paragraph 10(d) of this Confirmation,

THEN

(3) the supply shall be deemed "unreliable."

Rule 11

IF

(1) the seller's supply of energy to buyer is deemed unreliable in any calendar year,

THEN

(2) buyer may, at buyer's election, terminate the remaining portion of the transaction provided for by this Confirmation by giving the seller written notice of such termination by December 30 of that year.

Rule 12

IF

(1) the seller is unable for any reason and at any time to make energy available at the Points of Delivery at which it has initially notified buyer it will make energy available then,

THEN

(2) the seller may meet its obligations to buyer by making energy available to buyer at any one or more alternate Points of Delivery which at the time are designated by seller as alternative Points of Delivery at which seller is able to schedule energy for delivery to buyer.

Rule 13

IF

(1) the seller is NOT excused from supplying energy under this Agreement, AND

(2) the seller has NOT exceeded the specified maximum aggregate amount during the Supply Period,

THEN

(3) the seller shall make available to buyer up to the specified hourly amount of peaking firm energy during all hours of all days of the Supply Period, subject to a specified maximum aggregate amount during the Supply Period, AND

(4) the buyer shall schedule, take, and pay for, up to a specified hourly amount of peaking firm energy during all hours of all days of the Supply Period, subject to a specified maximum aggregate amount during the Supply Period.

Rule 14

IF

(1) the buyer has scheduled energy, AND

(2) (A) seller suffers a force majeure event, OR

(B) buyer suffers a force majeure event, AND

(3) the buyer fails to receive energy,

THEN

(4) then buyer may continue to schedule or reschedule up to a specified hourly amount of peaking firm energy during all hours of all days of the Supply Period, subject to the specified maximum aggregate amount during the Supply Period.

Rule 15

IF

(1) (A) the buyer causes a Breach Period to occur, OR

(B) the seller causes a Breach Period to occur,

THEN

(2) the buyer and the seller will consult with each other as to the extent of the Breach Period to facilitate the efforts of the nondefaulting party to mitigate its direct damages.

Rule 16

IF

(1) the seller arranges to resell energy during a Breach Period caused by the buyer, AND

(2) the buyer has NOT indicated that the buyer will NOT resume taking energy after the 48 hour period has expired,

THEN

(3) the seller's arrangement to resell energy may NOT exceed 48 hours in duration.

Rule 17

IF

(1) the buyer arranges to purchase energy during a Breach Period caused by the seller, AND

(2) the seller has NOT indicated that the seller will NOT resume supplying energy after the 48 hour period has expired,

THEN

(3) the buyer's arrangement to purchase energy may NOT exceed the agreed upon damages for a seller breach.

Rule 18

IF

(1) the seller arranges to resell energy during a Breach Period caused by the buyer, AND

(2) the buyer has indicated that the buyer will NOT resume taking energy after the 48 hour period has expired,

THEN

(3) the seller's arrangement to resell energy may exceed 48 hours in duration.

Rule 19

IF

(1) the buyer arranges to purchase energy during a Breach Period caused by the seller, AND

(2) the seller has indicated that the seller will NOT resume supplying energy after the 48 hour period has expired,

THEN

(3) the buyer's arrangement to purchase energy may exceed the agreed upon damages for a seller breach.

Rule 20 - third party supplier, excused situation

IF

(1) buyer and seller enter into an Agreement for the sale of energy, AND

(2) seller plans to meet the conditions of this sale by purchasing energy from a third party, AND

(3) the seller's purchase is reduced, interrupted, or curtailed,

THEN

(4) seller will have the right to reduce, interrupt, or curtail the seller's sale to buyer, AND

(5) seller will make all reasonable efforts to provide buyer with notice of such reduction, interruption, or curtailments as soon as possible after seller receives notice.

Rule 21

IF

(1) buyer and seller enter into an Agreement for the sale of energy, AND

(2) buyer fails to schedule and receive the electric energy, AND

(3) (A) the failure was NOT excused by Uncontrollable Forces, OR

(B) the failure was NOT caused by the seller,

THEN

(4) buyer will pay seller an amount for each MWh equal to the positive difference between [i] the contract Energy Price and [ii] the price at which seller was able to sell the energy.

Rule 22

IF

(1) buyer and seller enter into an Agreement for the sale of nonfirm energy,

THEN

(2) seller shall have the right to reduce nonfirm energy to the extent seller deems appropriate, AND

(3) seller shall have the right to interrupt or curtail delivery to buyer to the extent seller deems appropriate, AND

(4) seller shall provide buyer advance notice of such interruption as soon as possible.

Rule 23

IF

(1) buyer and seller enter into an Agreement for the sale of nonfirm energy, AND

(2) buyer fails to take delivery of energy from seller, AND

(3) uncontrollable forces, as provided in Section III of the Interconnection Agreement, did NOT excuse the buyer's failure to take delivery of energy,

THEN

(4) buyer shall pay to seller the positive difference between the amount buyer would have paid seller under this Confirmation for such interrupted MWh and the price, if any, at which seller is able to resell such energy.

Rule 24

IF

(1) the seller fails to supply short-term capacity and energy as scheduled in this Confirmation, AND

(2) the buyer and the seller have NOT expressly agreed in writing to damages that the buyer would incur as a result of this default of the seller's obligation to supply short term capacity and energy,

THEN

(3) the seller's liability shall be limited to direct damages only, AND

(4) the direct damages of buyer shall be, as applicable, the increased expenses of replacement capacity and energy purchases, the increased expenses of generation to replace the purchase from the seller, or, if neither alternative is possible, the loss of margin associated with curtailment of load (revenue less system-wide power costs) not to exceed the specified dollar amount per MWh in such case, AND

(5) the seller shall pay the buyer the actual amount of such direct damages upon receipt of an invoice therefor.

Rule 25

IF

(1) the seller fails to supply short-term capacity and energy as scheduled in this Confirmation, AND

(2) the buyer and the seller have NOT expressly agreed in writing to damages that the buyer would incur as a result of this default of the seller's obligation to supply short term capacity and energy,

THEN

(3) the seller shall have the right to verify any claim submitted for direct damages by examining the buyer's records pertaining thereto through a third-party auditing entity acceptable to both the buyer and the seller, AND

(4) such records, including cost information, that are indicated as proprietary and confidential shall not be disclosed to the seller, AND

(5) the buyer shall only make available to the auditing agency such records and information on the condition that the auditing agency will not disclose such records and information, except as required by law, to the seller or others without the express written approval of the buyer, AND

(6) the auditing agency shall use the records and cost information only to verify the buyer's direct damage claims, AND

(7) the buyer and the seller do not waive any rights, remedies, defenses, or causes of action to obtain recovery of a direct damage claim or to defend against any such claim.

Rule 26

IF

(1) the buyer fails to take short-term capacity and energy as scheduled in this Confirmation, AND

(2) the buyer and the seller have NOT expressly agreed in writing to damages that the seller would incur as a result of this default of the buyer's obligation to take short-term capacity and energy,

THEN

(3) the buyer's liability shall be limited to direct damages only, AND

(4) the buyer shall pay the seller the actual amount of such direct damages upon receipt of an invoice therefor, AND

(5) the buyer shall have the right to verify any claim submitted for direct damages by examining the seller's records pertaining thereto through a third-party auditing entity acceptable to both the buyer and the seller, AND

(6) such records, including cost information, that are indicated as proprietary and confidential shall not be disclosed to the buyer.

Rule 27

IF

(1) the buyer fails to take short-term capacity and energy as scheduled in this Confirmation, AND

(2) the buyer and the seller have NOT expressly agreed in writing to damages that the seller would incur as a result of this default of the buyer's obligation to take short-term capacity and energy,

THEN

(3) the seller shall only make available to the auditing agency such records and information on the condition that the auditing agency will not disclose such records and information, except as required by law, to the buyer or others without the express written approval of the seller, AND

(4) the auditing agency shall use the records and cost information only to verify the seller's direct damage claims, AND

(5) the buyer and the seller do not waive any rights, remedies, defenses, or causes of action to obtain recovery of a direct damage claim or to defend against any such claim.

[END OF DOCUMENT

Endnotes

1. Michael Guth serves as a Risk Management Consultant to electric utilities in the United States. He has a B.A. (Economics) from Rice University, an M.S. (Mathematical Economics) from Caltech, and a Ph.D. (Economics), and J.D., both from University of Tennessee. When he is not advising utilities, he practices law in Oak Ridge, Tennessee. E-mail: mike@@michaelguth.com.

2. I was working as a consultant to a large utility in the Southeast during this period, and the prices mentioned in this paragraph reflect my direct observations including analysis of monthly price averages in the Southeast for the preceding four years.

3. "The controversy was triggered when Midwestern utilities reported electricity shortages and sharp price hikes on the open spot electric power market during a heat wave in June. Illinois Power Co. reported that a combination of sharply rising demand combined with limited supply - caused in part by the failure of several traders to meet deliver commitments - caused the spot market's asking price for wholesale generation to reach as high as $10,000/MWh, skyrocketing from a normal average of around $30/MWh. One buyer paid $7,500/MWh, according to the utility." FERC to Investigate Electricity Price Hikes, 4 Foster Natural Gas Report No. 2192 (July 16, 1998).

4. "Utilities themselves aren't immune from power shortages. Employees at American Electric Power, Columbus, Ohio, one of the country's largest utilities serving seven states from Ohio to Virginia, took stairs instead of some elevators in their 31-story building and worked with the lights off to preserve energy. Meanwhile, AEP sought voluntary reductions from its largest consumers and prepared plans for more service interruptions." Kathryn Kranhold, "Power Cuts Hit Midwest and May Spread," Wall Street Journal (June 26, 1998).

5. See, e.g., Kathryn Kranhold, "Power Cuts Hit Midwest and May Spread," Wall Street Journal (June 26, 1998) ("After years of utility-industry warnings about impending electricity shortages, power interruptions have become a reality for some big customers in the Midwest."); "Get on Energy Conservation Track," Milwaukee Journal Sentinel (July 13, 1998)("The reality now is that Wisconsin customers are experiencing the harsh payback in the form of power shortages, while at the same time encountering a dangerous cutback in energy conservation investment.").

6. The defaults led some power marketing firms to sustain enormous losses and ultimately drove firms such as Pacificorp and LG&E to eliminate their power marketing subsidiaries. "Huge Power Price Spikes Force LG&E from Wholesale Marketing Businesses," Inside FERC Gas Mkt. Rep. 10 (Aug. 7, 1998) ("LG&E Energy Corp., on of the 20-largest gas marketers in North America, posted a $225 million second-quarter loss and subsequently shut down its wholesale trading operations. . . . Some other marketing companies, including Engage Energy US L.P., have reported losing considerable sums of money on the power price spike, but none to this extent.").

7."El Paso Energy Marketing has sued the City of Springfield, Ill. in a Texas district court, seeking $7.4 million in damages due to Springfield's canceling a power supply agreement on June 24, one day after Federal Energy Sales Inc. defaulted on its supply obligations to Springfield and numerous other retail power providers, including First Energy Corp. of Illinois. A month earlier, First Energy - parent of Ohio Edison, Pennsylvania Power, Cleveland Electric Illuminating and Toledo Edison - had pointed its finger at Federal Energy in an Ohio state court for failing to fulfill its contractual obligations to deliver power. In a letter to shareholders, First Energy revealed that it lost $80 million because of excessive power prices it was forced to pay during the recent heat wave. First Energy was not the only trader that defaulted during the June market fiasco, however. The Power Company of America and the Barr Devlin Electricity Trading Group also defaulted." FERC to Investigate Electricity Price Hikes, 4 Foster Natural Gas Report No. 2192 (July 16, 1998).

8. We see this emphasis on keeping a transaction viable in the following quotation attributed to the PECO Energy Power Team. "Our first priority, however, is not to win a lawsuit. Our first priority is to establish whether we have a viable business partner." FERC, Responding to Pressure, Says It Will Investigate Midwest Turmoil, Electric Utility Week 3 (July 20, 1998).

9. Aluminum Company Of America V. Central Lincoln Peoples' Utility District et al., 467 U.S. 380, 383, 104 S.Ct. 2472, 2476 (1984) (citations and footnotes omitted). See also Donald A. Haagensen and Jay T. Waldron, "Aluminum Co. of America v. Central Lincoln Peoples' Utility District: Supreme Court Deference to Appellate Advocacy," 15 Envtl L. 343, 345 n.9 (1985) ("Firm power is the amount of power which BPA, relying primarily on a hydroelectric generating system, can deliver even in the driest or 'critical' years. In most years, BPA's reservoirs have more water from rain and snow melt than in a critical year, and BPA can generate additional power. BPA labels this additional power nonfirm.").

10. "Expert systems are programs that mimic the decision-making process of a human expert. Conceptually, these systems are similar to a decision tree. By branching in response to user input, the expert system seeks an answer to the user's question." James N. Calvin, Tax Expert Systems Begin to be Feasible in Accounting Offices, 47 Tax'n for Acct. 381, 381 (1991).

11. Donald Waterman and Mark Peterson at the RAND Corporation's Center for Civil Justice pioneered the development of expert systems applied to the law. Donald Waterman and Mark Peterson, Models of Legal Decision Making, 37 Rand Corporation Memo R-2717-ICJ (1981).

12. Cleveland Electric Illuminating Co. v. City of Cleveland, 50 Ohio App.2d 275, 363 N.E.2d 759 (1976)(Electric utility successfully sued city when the city fell behind in payments to the utility due under a power sales agreement).

13. See, e.g., Energy Tactics, Inc. v. Niagara Mohawk Power Corp., 219 A.D.2d 577, 631 N.Y.S.2d 697 (NY App. 1995) (Power sales contract obligated utility to purchase at contractual rate all electricity produced by alternative energy producer that was normal and comparable with past production rates during the first five years of the contract.). For a case concerning antitrust implications of curtailing power, although not a dispute involving a purchaser and seller, see Schuylkill Energy Resources, Inc. v. Pennsylvania Power & Light Co., 113 F.3d 405 (3^rd Circuit, 1997), cert. den., 118 S.Ct. 435 (1997) (Allegation that an utility's curtailment policy limited the availability to consumers of alternative energy sources did not constitute an antitrust injury, because the cogenerator was not the utility's competitor).

14. Two examples of cases that did not turn up in this search involved decisions requiring utilities to sell nonfirm power, when it was produced, to be delivered to nonpreference customers. See Volunteer Elec. Corp. v. Tennessee Valley Auth., 139 F.Supp. 22 (E.D. Tenn. 1954), aff'd 231 F.2d 446 (6^th Cir. 1956); Arkansas Power & Light Co. v. Schlesinger, No. 79-1263 (D.D.C. Oct. 20, 1980).

15. For examples of relevant federal regulations promulgated during this same period, see F.E.R.C. Reg. No. 151, 26-27, 66-80, 21 FERC (CCH) ¶ 61,021 (1982) (reasonableness of provisions for withdrawal of preference power from nonpreference customers under Niagara Redevelopment Act must be determinated as of the time the contracts were executed with nonpreference customers.); Bonneville Power Administration, U.S. Dept. of Energy, Revised Proposed Nonfirm Energy Policy for Consumer Alternative Fuel Loads, 49 Fed. Reg. 35,853 (1984).

16. NLESB is written in Turbo-Prolog 2.0 and runs on an IBM-compatible personal computer with at least 640K RAM, DOS 3.0 or higher, and about 1 megabyte of hard disk space. NLESB also runs on a Windows platform, version 3.1 and higher.

17. Normalization is a process of "rigorously consistent use of a limited set of syntax terms to express unambiguously the relationship among conditions and legal consequences stated in an outline format." Grayfred Gray, Computerization in Law Practice in 1986: Needs and Uses, 14 (unpublished manuscript, University of Tennessee College of Law, Knoxville, TN). See also, L. Allen, Symbolic Logic: A Razor-Edged Tool for Drafting and Interpreting Legal Documents, 66 Yale L. J. 833 (1957); G. Gray, Reducing Unintended Ambiguity in Statutes: An Introduction to Normalization for Original Drafting, 54 Tenn. L. R. 433 (1987).

18. The conventions of normalization include using parenthetical numbers to represent each condition or proposition that must occur for the consequence to occur (i.e., each set of conditions joined by an AND operator), and using parenthetical letters for conditions that are alternative ways for a given condition to be met (i.e., one or more conditions joined by an OR operator). Thus if the antecedent portion of a rule contained three conditions that had to be met before the consequence would follow, these conditions would be listed as (1) AND (2) AND (3). If, however, the consequence followed from two conditions, and the second condition could be met in one of two ways, this logic would be listed as (1) AND [(2)(A) OR (2)(B)], where (A) and (B) represent two alternatives, either one of which will satisfy the second condition.

Another important convention of normalization is that the OR-operator could be used in the antecedent portion of a rule, but not in the consequence. Thus, it is possible to have a rule of the form P OR Q --> R, but it is not possible to have a rule of the form P --> Q OR R. It is beyond the scope of this paper to prove why including an OR-operator in the consequence of a rule would violate the fundamental postulates of logic, but it involves an indeterminate logic in which a conclusion cannot be traced back uniquely to the premises that caused it. Layman F. Allen, "Language, Law, and Logic: Plain Legal Drafting for the Electronics Age," in COMPUTER SCIENCE AND LAW 75-100 (Bryan Niblett, Ed., 1980).

19. See, e.g., Trevor Bench-Capon and F. P. Coenen, Isomorphism and Legal Knowledge Based Systems, 1 Artif. Intel. & L. 65 (1992).

20. Layman Allen emphasizes this point in his writings. See, generally, Layman Allen, Problem of Communication and the Legal System, __ N. Y. L. J. __ (Oct. 27, 1967); Layman E. Allen & C. Rudy Engholm, Normalized Legal Drafting and the Query Method, 29 J. Legal Educ. 380, 402-03 (1978) (describing normalized statutes as if-then statements); Layman Allen and Rudy Engholm, The Need for Clear Structure in "Plain Language" Legal Drafting, 13 U. Mich. J. L. Ref. 455 (1980) (illustrating the usefulness of modern logic for achieving clarity in legal drafting). C.f. Grayfred B. Gray, Reducing Unintended Ambiguity in Statutes: An Introduction to Normalization of Statutory Drafting, 54 Tenn. L. Rev. 433 (1987) (providing examples of statutes expressed as if-then statements).

21. Breach Period is capitalized, because it is a defined term within the contracts from which the excerpts were taken.

22. Firm energy represents an enforceable contract for the supply of specific quantities of energy a date, time, place, and price. In contrast, nonfirm energy is sold without any assurance of availability of the energy to the purchaser. See, e.g., Bonneville Power Administration Definitions 45 (1979) (available in the Bonneville Power Administration Library, Portland, Oregon).

23. Most expert systems contain backward-chaining rules. See Edwina L. Rissland, Artificial Intelligence and Law: Stepping Stones to a Model of Legal Reasoning, 99 Yale L.J. 1957, 1981 n. 36 (1990) ("The systems can work either "forward"' by reasoning from facts to a desired conclusion supported by them, or "backward"' from a desired conclusion to find facts supporting it").

24. Id. at 1966.

25. The rules must be normalized outline form, but otherwise NLESB uses English or any written language as input.

26. Kevin W. Saunders, "What Logic Can and Cannot Tell Us About the Law," 73 Notre Dame L. Rev. 667, 680 n.50 (1998).

27. Id.