`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`______________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`______________
`
`LIBERTY MUTUAL INSURANCE CO.
`Petitioner
`
`v.
`
`PROGRESSIVE CASUALTY INSURANCE CO.
`Patent Owner
`______________
`
`Case CBM2012-00002
`Patent 6,064,970
`______________
`
`Before the Honorable JAMESON LEE, JONI Y. CHANG, and MICHAEL R.
`ZECHER, Administrative Patent Judges.
`
`REBUTTAL DECLARATION OF MARY L. O’NEIL ON BEHALF OF
`PETITIONER LIBERTY MUTUAL INSURANCE CO. REGARDING U.S.
`PATENT NO. 6,064,970
`
`I, Mary L. O’Neil, hereby declare under penalty of perjury:
`
`I have previously been asked by Liberty Mutual Insurance Co. (“Liberty
`
`Mutual”) to testify as an expert witness in this action. For purposes of this rebuttal
`
`declaration, I have been asked by Liberty Mutual to respond to certain assertions and
`
`opinions offered by Michael Miller and Progressive Casualty Insurance Co.
`
`(“Progressive”) concerning U.S. Patent No. 6,064,970 (“the ‘970 patent”) in this
`
`matter.
`
`
`
`
`
`
`
`Liberty Mutual Exhibit 1022
`Liberty Mutual v. Progressive
`CBM2012-00002
`Page 00001
`
`
`
`
`
`
`
`
`
`
`1.
`
`I am the same Mary L. O’Neil who provided a Declaration in this matter
`
`executed on September 14, 2012 as Exhibit 1009.
`
`2. My information regarding experience, qualifications, and compensation
`
`are provided along with my prior Declaration and Curriculum Vitae and case list
`
`(Exhibit 1010).
`
`I.
`
`Scope of Rebuttal Declaration
`
`3.
`
`I have been asked to respond to certain assertions and opinions of Mr.
`
`Michael Miller expressed in his declaration of May 1, 2013 as Exhibit 2010, his
`
`supplemental declaration of May 22, 2013 as Exhibit 2020, and certain assertions of
`
`Progressive in its Patent Owner’s Response of May 1, 2013.
`
`4.
`
`In developing my opinions below, and in addition to the materials
`
`identified in my prior declaration at paragraph 14, I have considered the following
`
`materials:
`
` Herrod Reference, GB2286369 (Ex. 1007);
`
` Declaration of Michael Miller (Ex. 2010);
`
` Supplemental Declaration of Michael Miller, including a document
`entitled “Actuarial Standard of Practice No. 12” (Ex. 2020);
`
` Document entitled “Risk Classification Statement of Principles” (Ex.
`2012);
`
` Patent Owner’s Response (Paper 27) (“Opposition” or “Opp.”);
`
`
`
`
`
`2
`
`Page 00002
`
`
`
`
`
`
`
`
`
`
` Board’s Decision on Institution of Covered Business Method Review
`(Paper 10);
`
` All other materials referenced as exhibits herein.
`II. Analysis and Opinions
`A. Mr. Miller’s Opinions and Progressive’s Assertions Regarding
`“Actuarial Classes” and Determining Auto Insurance Premiums
`
`5.
`
`In providing a definition of “actuarial class,” Mr. Miller states:
`
`In the field of motor vehicle insurance as of 1996, a person or ordinary
`skill in the art would have understood that “actuarial class” had the same
`meaning as risk class. . . . This definition is consistent with the definition
`in the Risk Classification Statement of Principles of the American
`Academy of Actuaries. A person of ordinary skill in the art in 1996
`would have adhered to this Statement of Principles.”
`
`Ex. 2010 ¶ 16 (Emphasis added).
`
`6. Mr. Miller has presented the Risk Classification Statement of Principles
`
`(Ex. 2012) as if it were a binding verbatim requirement to be followed. That is
`
`incorrect. The guidance provided by the Statement of Principles and its usage is
`
`explained by Interpretative Opinion 4: Actuarial Principles and Practices (1982) of
`
`the American Academy of Actuaries (AAA, the umbrella organization for all
`
`actuaries), which would have been in effect through 1996. In fact, Interpretative
`
`Opinion 4: Actuarial Principles and Practices (Ex. 1023)1 states in essence that the
`
`
`1 Exhibit 1023 is a true and correct copy of “Interpretative Opinion 3: Professional
`Communications of Actuaries and Interpretive Opinion 4: Actuarial Principles and
`
`
`3
`
`
`
`
`
`Page 00003
`
`
`
`
`
`
`
`Statement of Principles cited by Mr. Miller is only a guideline—one possible reference
`
`
`
`out of a large body of material that form the bases of Generally Accepted Actuarial
`
`Principles and Practices (which are a broad overview of how actuarial practice should
`
`be done):
`
`(a) Generally Accepted Actuarial Principles and Practices.
`
`Guide 4(b) requires the actuary to “exercise due diligence to ensure . . .
`that the methods employed are consistent with the sound actuarial
`principles and practices established by precedents or common usage
`within the profession. . .” Such “sound actuarial principles and practices”
`constitute Generally Accepted Actuarial Principles and Practices.
`
`(b) Sources of Generally Accepted Actuarial Principles and Practices.
`
`Sources of Generally Accepted Actuarial Principles and Practices emerge
`from the utilization and adoption of concepts described in actuarial
`literature. Such literature includes, but is not limited to, the Actuarial standards
`and Actuarial Compliance Guidelines adopted by the Actuarial Standards Board,
`the Recommendations and Interpretations published under the auspices of the
`American Academy of Actuaries; the professional journals of recognized professional
`actuarial organizations (including the Statements of Principles promulgated by the
`Society of Actuaries and the Casualty Actuarial Society); recognized actuarial
`textbooks and study materials; and applicable provisions of law and regulations; and
`may include standard textbooks or other professional publications in related fields
`such as mathematics, statistics, accounting, economics and law.
`
`Ex. 1023 at 6 (emphasis added, footnote omitted).
`
`7. Mr. Miller further incorrectly argues that a POSITA would have strictly
`
`adhered to the Statement of Principles in making “statistical considerations such as
`
`
`Practices,” adopted 1970-1982 by the American Academy of Actuaries and
`republished in 1992 by the Actuarial Standards Board, which I obtained online at
`
`
`4
`
`
`
`
`
`Page 00004
`
`
`
`
`
`
`
`the homogeneity, credibility, and predictive reliability of the claims data that will be
`
`
`
`gathered for each actuarial class.” Ex. 2010 ¶ 34. Rather, Opinion 4 again rebuts this
`
`assertion, stating ultimately in subsection (d) that “In all cases the professional
`
`judgment of the actuary should prevail” and provides Standards of Practice and
`
`Compliance Guidelines:
`
`(c) Standards of Practice and Compliance Guidelines.
`
`An actuary working in a specialized field should take into consideration any
`published Standard of Practice or Compliance Guideline of the Actuarial
`Standards Board. An actuary who uses principles or practices which
`differ materially from any published Standard of Practice or Compliance
`Guideline must be prepared to support the particular use of such
`principles or practices and should include in an actuarial communication
`appropriate and explicit information with respect to such principles and
`practices. . . . When dealing with a specific situation not covered by a published
`Standard of Practice or Compliance Guideline, the actuary should be aware of
`relevant precedent and generally available literature in deciding what constitutes
`Generally Accepted Actuarial Principles and Practices.
`
`Ex. 1023 at 6-7 (emphasis added); cf. also Ex. 2012 at 12.2
`
`8.
`
`Indeed, one such Standard of Practice—No. 12, attached to Mr. Miller’s
`
`supplemental declaration Ex. 2020—further belies Mr. Miller’s strict adherence to the
`
`
`http://www.actuarialstandardsboard.org/pdf/superseded/intopinion.PDF.
`2 “These statistical considerations—homogeneity, credibility and predictive stability—
`are somewhat conflicting” and “there is no one statistically correct risk classification
`system. . . . The decision as to relative weights to be applied will, in turn, be influenced
`by the nature of the risks, the management philosophy of the organization assuming
`the risk and the judgment of the designer of the system.”
`
`
`
`
`
`5
`
`Page 00005
`
`
`
`
`
`
`
`Statement of Principles and repeated insistence on the use of actual claims data to
`
`
`
`generate actuarial classes (see Ex. 2010 ¶¶ 16-18, 34-35, 41-42, 44, 51):
`
`“5.1 Methods to Demonstrate Cost Differences—A risk classification
`system is equitable if material differences in costs for risk characteristics
`are appropriately reflected in the rate. Classification subsidies result
`when the price paid by an individual or class of individuals fails to reflect
`differences in costs among the risk classes.
`
`A relationship between a risk characteristic and cost is demonstrated if it
`can be shown that experience is different when the characteristic is
`present. In demonstrating the relationship, the actuary can rely on actual or
`reasonably anticipated experience; the actuary is not constrained to using only the
`experience of the actuary’s client or company. Relevant information from any reliable
`source, including statistical or other mathematical analysis of available data, may be
`used. Information gained from clinical experience, or from expert
`opinion regarding the effects of change on future experience (e.g.,
`medical or engineering) may be used.
`
`In the absence of actual experience, an actuary may rely on clear actuarial evidence
`that differences in costs are related to a particular risk characteristic. In demonstrating
`this, the actuary may rely on clinical experience or expert opinion. For example, an
`environmental that which has been demonstrated by clinical experience
`to be related to additional deaths may be used until further actual
`experience becomes available.
`
`Sometimes it is appropriate for the actuary to make inferences without specific
`demonstration. For example, it would not be necessary to demonstrate that
`persons with seriously impaired, uncorrected vision would represent a
`high risk as operators of motor vehicles.”
`
`Ex. 2020 at Attachment pp.3-4 (emphasis added).
`
`9. Mr. Miller also opines that it is not required that actuarial classes are
`
`used to meet the statutory standard of charging premiums that are “not unfairly
`
`discriminatory.” Ex. 2010 ¶¶ 27-29. While it is true that the statute does not
`
`
`
`
`
`6
`
`Page 00006
`
`
`
`
`
`
`
`prescribe the means by which the law must be complied with, the utilization of
`
`
`
`actuarial classes is and has been the accepted means for such compliance. And,
`
`notably, Mr. Miller gives no alternative examples (or any examples of an approved
`
`policy that did not employ actuarial classes).
`
`10. At paragraph 18, Mr. Miller states:
`
`The future insurance loss (i.e., risk of loss) being estimated is the product
`of the probability of an occurrence of an insured claim times the likely
`cost of the claim. Because the probability of an insurance claim
`occurring is a different value than the probability of an auto accident
`occurring, auto insurance rates are typically calculated based on the
`likelihood of claim occurrence, not the likelihood of accident
`occurrence.
`
`11. Mr. Miller’s opinion here is muddled. Accurately stated, estimated future
`
`insurance loss costs are based on the expected pure premium, which is equal to the
`
`product of the expected claim frequency and the expected claim severity. The further
`
`opinion by Mr. Miller that an accident does not necessarily generate a one-to-one
`
`relationship to claims is not relevant to the concept of expected loss costs or expected
`
`pure premium as long as a consistent measure is utilized.
`
`12.
`
`For example, accident statistics, like any other risk characteristic that is
`
`or can be used to create an actuarial class (such as age, mileage, sudden braking
`
`events, or excessive speeding), does not necessarily have a one-to-one relationship to
`
`claims. Yet accident statistics can indeed in themselves be the basis for an actuarial
`
`class—see, e.g., ‘970 patent (Ex. 1001) at 1:28-43 (number of at-fault accidents)—just
`
`
`
`
`
`7
`
`Page 00007
`
`
`
`
`
`
`
`as, for example, the ‘970 points out that “number of sudden braking situations” can
`
`
`
`be an actuarial class at 4:30-45. It is the job of an actuary to determine how risk
`
`characteristics, such as number of accidents or sudden braking events, correlate to
`
`predicted future insurance losses so that an insurer can charge an individual the
`
`proper premium. This can be done—as explained, for example, in Standard of
`
`Practice No. 12—using “actual experience” (actual frequency and severity claims data)
`
`or “any reliable source, including statistical or other mathematical analysis of available
`
`data” (Ex. 2020 at Attachment pp.3-4 (emphasis added)).
`
`13.
`
`Importantly, I further note that Mr. Miller’s entire discussion of the
`
`necessity of collecting and applying actual claims data (see Ex. 2010 ¶¶ 16-18, 34-35,
`
`41-42, 44, 51) is wholly lacking from the ‘970 patent. Nowhere does the ‘970 patent
`
`explain how or where to obtain claims data, much less how to use such data to
`
`calculate overall claims losses, pure premiums, expected loss costs, etc. Rather, those
`
`techniques are left up to the knowledge of a POSITA, such as myself. While Mr.
`
`Miller repeatedly states that the prior art references involved are lacking such
`
`discussions, so is the ‘970 patent, but nevertheless, it does not mean that a POSITA
`
`would simply not possess such knowledge.3
`
`
`3 Although I have not addressed certain of Mr. Miller’s arguments in his declaration
`here, that does not mean I agree with them. Rather, I understand some of Mr.
`Miller’s assertions about, for example, certain terminology (e.g., “risk factor” or “rate
`
`
`8
`
`
`
`
`
`Page 00008
`
`
`
`
`
`
`
`
`
`
`14. Mr. Miller states that the unit of exposure for automobile insurance is
`
`the insured vehicle and further states that “[p]remiums for auto insurance are typically
`
`not calculated or quoted on a per person or per driver basis.” Ex. 2010 ¶ 21. Mr.
`
`Miller’s statement is again muddled and incomplete. Premiums are generally quoted
`
`on a policy basis. It is generally the risk characteristics of the insured driver that are
`
`applied (using actuarial classes) to the base premiums by coverage or car to determine
`
`the premium.4 For liability coverages, the specific car driven is not part of the rating
`
`process because the premium is determined using actuarial classes based on only the
`
`driver’s risk characteristics and the base premium for the selected coverage. See Ex. ‘970
`
`Patent at 1:28-2:12. For physical damage coverages, the premium is determined using
`
`actuarial classes based on the driver’s risk characteristics and the base premium by
`
`coverage for the specific insured vehicle. See id.
`
`15. Again, I also note that Mr. Miller’s opinions about any alleged issues
`
`arising regarding the “per driver” versus “per vehicle” distinction are completely
`
`lacking from the ‘970 patent. Thus, even if Mr. Miller’s opinions were accurate, he
`
`
`factor”) do not appear pertinent in this proceeding involving the ‘970 patent.
`4 Even Mr. Miller appears to contradict himself on this argument when defining “pure
`premium”: “The numerical value is a ratio of the expected loss of one actuarial class
`to another. The expected loss for an insured is sometimes called the ‘pure premium.’”
`Ex. 2010 ¶ 19. Here he has defined it in terms of the insured (driver) rather than the
`vehicle. Progressive also misstates my views on the definition of “pure premium”:
`“Ms. O’Neil referred to ‘pure premium,’ which has the same meaning as expected
`losses.” Opp. at 11 n.2. Rather, “pure premium” is the loss dollars per unit of
`
`
`9
`
`
`
`
`
`Page 00009
`
`
`
`
`
`
`
`relies exclusively on the knowledge of a POSITA—the same knowledge that the
`
`
`
`POSITA would have when reading any of the prior art references Mr. Miller criticizes.
`
`16. Mr. Miller also attempts to utilize the muddled distinction he has created
`
`between drivers and vehicles to conclude that insureds are not “assigned” to actuarial
`
`classes, stating as follows:
`
`In the case of auto insurance, it is not an insured person that is being
`“assigned” to an actuarial class of similar risks. It is the premium
`charged, and any future claim losses, associated with the insured car that
`are coded to an actuarial class for each of the risk characteristics used in
`determining the premium for the insured car.
`
`Ex. 2010 ¶ 30.
`
`17. This is incorrect and cannot be concluded from his former muddled
`
`point about driver versus vehicle. In fact, insurance policies typically identify the
`
`assigned classification of the driver for rating purposes. Similarly, distinct aspects of the
`
`vehicle’s classification are identified. The final premium is determined by a rating
`
`algorithm combining the driver, vehicle, and selected coverages. The resulting insurer
`
`premium and claim experience for a particular classification consists of the premium
`
`and claim amounts for all drivers combined who have been assigned to that
`
`classification.
`
`
`exposure.
`
`
`
`
`
`10
`
`Page 00010
`
`
`
`
`
`
`
`
`
`
`18.
`
`In an attempt to further his argument about driver versus vehicle, Mr.
`
`Miller provides an incorrect example to illustrate his contention that insured cars, not
`
`drivers, are the basis for classification analysis:
`
`For example, assume the premium for an insured car is determined
`based on only three risk characteristics: the rated-driver of the insured
`car is an adult driver, the coverage is subject to a $500 deductible, and
`the insured is eligible for a claims-free discount. The insurer’s
`policyholder records for this insured car will reflect a separate code for
`each of the three risk characteristics (i.e., adult driver, $500 deductible,
`and claims-free). When analyzing the difference in risk between adult
`drivers and youthful drivers, the premium and claims data of our
`hypothetical insured car will be included with the adult risk class. When
`analyzing the difference in risk between a $500 deductible and a $250
`deductible, the premium and claims data of our hypothetical insured car
`will be included with the $500 deductible risk class. When analyzing the
`difference in risk between insureds that are claims-free and those that are
`not, the premium and claims data of our hypothetical insured car will be
`included with the claims-free risk class.
`
`Ex. 2010 ¶ 31.
`
`19. Mr. Miller’s example is incomplete and incorrect. The driver in Mr.
`
`Miller’s example would have been assigned a classification of adult driver. The base
`
`rate for the physical damage coverage for the vehicle would be modified for a $500
`
`deductible (if some other deductible amount was reflected in the base rate). The final
`
`rate would be modified by a claims free discount. The comparison of premium and
`
`claim experience of adult and youthful drivers would be a comparison of data for the
`
`adult and youthful classifications based on their risk characteristics. For the deductible
`
`option, the comparison of premium and claim data would be a comparison of the
`
`
`
`
`
`11
`
`Page 00011
`
`
`
`
`
`
`
`data for the $500 and $250 classification options. Finally, the claim-free discount
`
`
`
`would compare the premium and claim data for classifications with the discount and
`
`without the discount. In each of these rate making analyses, the claim data would be
`
`collected for each resulting exposure, or the pure premium. This analysis remains
`
`consistent with the initial assignment of drivers to rating classifications in order to
`
`determine the policy premium.
`
`20. Mr. Miller argues that obtaining “household” data is necessary to rate an
`
`insurance policy:
`
`The “rated-driver” is the specific driver among multiple drivers in the
`household, whose risk characteristics impact on the premium calculation
`for a specific insured auto. . . . The premium is calculated to reflect the
`total insured risk associated with the ownership and operation of a
`vehicle that potentially has multiple operators. The insureds under a
`personal auto insurance policy are the named insured(s) listed on the
`policy’s declaration page, as well as all relatives resident in the household.
`
`* * * *
`
`[A] POSITA would have understood that to accurately determine an
`auto insurance premium, the risk characteristics of all drivers resident in
`the household are needed. These data are necessary so that the insurer
`can determine which of the operators in the household should be the
`“rated-driver” on each insured car in the household.
`
`Ex. 2010 ¶¶ 22-24, 49.
`
`21. Mr. Miller is incorrect. There is no requirement that household data be
`
`collected because, for example, the insurer’s rating algorithm can be applied even if
`
`only one driver participates in the monitoring. Also, an individual driver can get a
`
`
`
`
`
`12
`
`Page 00012
`
`
`
`
`
`
`
`policy on a single car for himself only, so there is no requirement that an insurer get
`
`
`
`information on multiple drivers or “household” data.
`
`22. This household “requirement” argued by Mr. Miller is yet another point
`
`that is not discussed in the ‘970 patent. So, again, even if Mr. Miller’s opinions were
`
`correct, his exclusive reliance on the knowledge of a POSITA (rather than any ‘970
`
`disclosures) is the same knowledge the POSITA would have in analyzing the prior art
`
`references.
`
`B. Mr. Miller’s Opinions and Progressive’s Assertions Regarding
`Kosaka and Fuzzy Logic
`
`23. Regarding a POSITA’s awareness of fuzzy logic, Mr. Miller states that
`
`“[t]he person of ordinary skill would not have had experience using or applying fuzzy
`
`logic to the determination of insurance premiums, but would have been relatively
`
`sophisticated in the use of multi-variant statistical analysis of risk classification data.”
`
`Ex. 2010 ¶ 14.
`
`24. Mr. Miller’s opinion improperly narrows the knowledge of a POSITA.
`
`Although fuzzy logic was not the predominant mathematical process applied in rate
`
`filings for determination of the overall premium in 1996, the literature confirms that
`
`the methodology was well known and applications to classification rating and
`
`
`
`
`
`13
`
`Page 00013
`
`
`
`
`
`
`
`underwriting were well-documented. See, e.g., Ex. 1024 (Shapiro Article) 5 (providing a
`
`
`
`history of the application of fuzzy logic in insurance since 1982 and an extensive list
`
`of references at pages 57 through 61, many predating 1996; this reference also
`
`demonstrates the application of fuzzy logic to rating territories and classifications
`
`based on age groupings); Ex. 1025 (Carreno Article)6 (describing “knowledge based
`
`system . . . that combines fuzzy processing with [a] rule-based expert system” for
`
`“improved decision aid for evaluating risk for life insurance” (p.536) ; the “output of
`
`the system consists of a crisp value for Risk in the range [0, 1]” (p.538)); see also Ex.
`
`1026 (Lemaire Article)7 ((summarizing the history of fuzzy logic), 54-55 (extensive
`
`bibliography showing many papers applying fuzzy logic); Ex. 1027 (Derrig Article)8
`
`(presenting an application of fuzzy techniques to derive Massachusetts automobile
`
`rating territories); Ex. 1028 (Young Article)9 (extensive bibliography at 761-62
`
`
`5 Arnold F. Shapiro, An Overview of Insurance Uses of Fuzzy Logic, in Paul P. Wang, et al.,
`eds., Computational Intelligence in Economics and Finance Volume II pp. 25-61 (Chapter
`One) (Springer Berlin Heidelberg, 2007).
`6 Luis A. Carreno, et al., A Fuzzy Expert System Approach to Insurance Risk Assessment
`Using FuzzyCLIPS, in WESCON Conference Record pp.536-541 (1993) (reference no. 13
`from Shapiro Article at 58).
`7 Jean Lemaire, Fuzzy Insurance, ASTIN Bulletin International Actuarial Association
`Vol. 20, No. 1, pp.33-56 (1990) (date of article shown on download page at end of
`exhibit).
`8 Richard A. Derrig, et al., Fuzzy Techniques of Pattern Recognition in Risk and Claim
`Classification, Journal of Risk and Insurance Vol. 62, No.3, Sept. 1995.
`9 Virginia R. Young, Adjusting Indicated Insurance Rates: Fuzzy Rules that Consider Both
`Experience and Auxiliary Data, in Proceedings of the Casualty Actuarial Society Casualty
`
`
`14
`
`
`
`
`
`Page 00014
`
`
`
`
`
`
`
`consisting primarily of papers written about fuzzy logic in insurance well before 1996).
`
`
`
`The abundance of literature available in 1996 related to the application of fuzzy logic
`
`to insurance further demonstrates that a POSITA would have had knowledge of the
`
`topic and its applications to classification ratemaking, and would have known how to
`
`address it (such as where to look for any further needed information). The rapid
`
`growth in the topic of fuzzy logic took place well before 1996, including in the
`
`insurance industry. Hence, Mr. Miller has inaccurately narrowed the knowledge of a
`
`POSITA.
`
`25. Mr. Miller states his understanding of Kosaka as follows:
`
`Kosaka discloses a risk evaluation device and an insurance premium
`determination device. The risk evaluation device detects the speed
`relative to a preceding vehicle. I understand that the speed and wave
`length data are among the input values used to determine a range of
`fuzzy risk values via fuzzy logic.
`
`Ex. 2010 ¶ 36.
`
`26. This is a muddled and incomplete understanding of Kosaka. The risk
`
`evaluation device measures certain operating characteristics of the vehicle, with
`
`Kosaka providing examples such as speed and relative distance. These readings are
`
`utilized via the process of fuzzy logic to determine risk evaluation values with a
`
`defined range. The overall risk evaluation value is a single numerical value (based on
`
`
`Actuarial Society - Arlington, Virginia, 1997, pp.734-765 (date of article shown on
`download pages at end of exhibit).
`
`
`
`
`
`15
`
`Page 00015
`
`
`
`
`
`
`
`fuzzy inputs) that is indicative of risk and that Kosaka uses to adjust the premium. See
`
`
`
`Ex. 1004 (Kosaka) at 1, 8, Fig. 11. The process of using fuzzy logic has been well
`
`described in several of the references which I have listed. One example of the process
`
`is for Massachusetts rating territories (Ex. 1027).
`
`27. Mr. Miller also argues that Kosaka teaches nothing regarding premium
`
`determination using actuarial classes and could not be combined with the “crisp” logic
`
`system of Herrod. See Ex. 2010 ¶¶ 39-44, 48-51. First, Mr. Miller states that the
`
`POSITA would have had no experience with fuzzy logic, would not know how to use
`
`fuzzy logic, or know how to apply fuzzy logic to establish actuarial classes; nor, he
`
`says, would the POSITA know how to process the data in Kosaka using fuzzy logic.
`
`Ex. 2010 ¶¶ 36-39, 43-44.
`
`28. Mr. Miller is incorrect in his opinions about a POSITA’s understanding
`
`of Kosaka’s fuzzy logic disclosures. Importantly, I note that Kosaka explicitly states
`
`that fuzzy logic need not be used: “fuzzy logic was used as the means for determining
`
`risk evaluation values in this example of embodiment, but determination may be
`
`carried out without using fuzzy logic. Calculation may also be carried out using a
`
`common insurance table.” Ex. 1004 at 6. His criticism about fuzzy logic and the
`
`supposed incompatibility with Herrod is thus misplaced. Yet, in any event, as
`
`commented above, Mr. Miller has unnecessarily limited the definition of POSITA
`
`
`
`
`
`16
`
`Page 00016
`
`
`
`
`
`
`
`given that there was significant literature on the subject of fuzzy logic prior to 1996
`
`
`
`regardless of whether or not its application was widely utilized.
`
`29.
`
` Moreover, fuzzy logic is not the key point in the Kosaka reference. It is
`
`merely a different way to process the vehicle operation data collected using the data
`
`collection device. Fuzzy logic is applied at various steps in an example of the process
`
`of deriving overall risk evaluation values in Kosaka, and Kosaka indicates (as noted
`
`above) that these values are produced whether or not fuzzy logic is employed. The
`
`POSITA would have recognized that the actual data derived from the operation of an
`
`automobile in Kosaka could be translated into risk evaluation values (with or without
`
`the application of fuzzy logic). And a POSITA would have understood that the
`
`Kosaka overall output risk evaluation values are single numerical values, which
`
`presents no hurdle to using those values with the system in Herrod
`
`30. As supposed support for Mr. Miller’s opinions about fuzzy logic, Mr.
`
`Miller states:
`
`I understand that fuzzy logic relies on a fuzzy-set mathematical theory
`that results in data sets that are not mutually exclusive. The person of
`ordinary skill could not be certain there was any true difference in risk
`between two Kosaka risk values produced via fuzzy logic, especially if
`those risk values were near the intersection of two sets. Two operators
`with different Kosaka risk values would not necessarily have objectively
`different driving patterns or objectively different insurance risks. The
`ordinary skilled artisan would not use such data for purposes of
`establishing actuarial classes.
`
`Ex. 2010 ¶ 39.
`
`
`
`
`
`17
`
`Page 00017
`
`
`
`
`
`
`
`
`
`
`31. Mr. Miller has confused the process with the result. Kosaka utilizes (as
`
`one example) fuzzy logic on the data collected by the data collection device and
`
`generates distinct risk evaluation values to assess insurance risk. Indeed, as I
`
`mentioned above, fuzzy logic applications for insurance in general, and, in particular,
`
`classification rate making, have been well demonstrated in the literature. See, e.g., Ex.
`
`1027 (Derrig Article) (describing an application of fuzzy techniques to derive
`
`Massachusetts automobile rating territories; the resulting rating territories conform to
`
`the required criteria for rating classifications such as non-overlap between classes); Ex.
`
`1025 (Carreno Article) (describing an insurance system that combines “fuzzy
`
`processing with [a] rule-based expert system” and outputs a “crisp value for Risk in
`
`the range [0, 1]”).
`
`32. Mr. Miller further argues that the Kosaka risk evaluation values were
`
`incomplete because they must be relative to a safety or risk standard, and he proposes
`
`a hypothetical example where two cars may have the same risk value though driving in
`
`different types of traffic on a “city street versus a four-lane expressway.” Ex. 2010
`
`¶ 40.
`
`33. Mr. Miller’s reasoning is incomplete. Actuarial classes in general (which
`
`may, e.g., be based on such characteristics as age, gender, etc.) are not necessarily
`
`based on external safety standards such as speed limits or traffic congestion as
`
`suggested by Mr. Miller. Actuarial classes are subsets of the total group of insureds.
`18
`
`
`
`
`
`Page 00018
`
`
`
`
`
`
`
`These subsets have been identified by distinctions in expected pure premium, and
`
`
`
`contrary to Mr. Miller’s suggestion not every actuarial class depends on location of the vehicle’s
`
`operation. As discussed above, even the ‘970 patent recognizes this—suggesting
`
`actuarial classes based on, for example, driving time, sudden accelerations, sudden
`
`decelerations, seatbelt usage, etc. (compare with parking or garage location, which
`
`clearly are location-dependent). See Ex. 1001 at 4:30-52; see id. at claims 6, 18 (actuarial
`
`classes used without specifying recording of location data). A particular actuarial class
`
`does not reflect every risk characteristic. Indeed, a combination of separate actuarial
`
`classes is used ultimately to determine a premium. See, e.g., Ex. 1001 at 1:28-2:21.
`
`34. This hypothetical example presented by Mr. Miller also ignores that a
`
`POSITA would recognize that the risk evaluation values are based on numerous
`
`observations of driving behavior and if distinctions in expected loss costs are
`
`observed between subsets of drivers, then valid actuarial classes are generated and
`
`applied. The city versus highway driving scenario noted by Mr. Miller is inapposite.
`
`The risk of the type of driving has been measured within the risk evaluation values.
`
`Such an evaluation need not be location-dependent. It may arise that additional
`
`location-dependent risk characteristics may be tested and applied. No classification
`
`rating system measures and applies every possible risk characteristic.
`
`35. Mr. Miller argues that Kosaka does not teach “homogeneity in actuarial
`
`class data”—citing again the gui