1. 1 How to test a philosophical theory empirically Barry Smith Institute for Formal Ontology and Medical Information Science http://ifomis.de sponsored by the Wolfgang Paul Program of the Alexander von Humboldt Stiftung

2. 2 and by the des Bundes

3. 3 Institute for Formal Ontology and Medical Information Science

4. 4 Ontology as a branch of philosophy seeks to establish the science of the kinds and structures of objects, properties, events, processes and relations in every domain of reality

5. 5 Ontology a kind of generalized chemistry or zoology (Aristotle’s ontology grew out of biological classification)

6. 6 Aristotle world’s first ontologist

7. 7 World‘s first ontology ( from Porphyry’s Commentary on Aristotle’s Categories)

8. 8 Linnaean Ontology

9. 9 Medical Diagnostic Ontology

10. 10 Ontology is distinguished from the special sciences it seeks to study all of the various types of entities existing at all levels of granularity

11. 11 and to establish how they hang together to form a single whole (‘reality’ or ‘being’)

12. 12 Sources for ontological theorizing: the study of ancient texts thought experiments (we are philosophers, after all) the development of formal theories the results of natural science now also: working with computers

13. 13 The existence of computers and of large databases allows us to express old philosophical problems in a new light

14. 14 Example: The problem of the unity of science The logical positivist solution to this problem addressed a world in which sciences are identified with printed texts What if sciences are identified with Large Databases ?

15. 15

16. 16 The Database Tower of Babel Problem Each family of databases has its own idiosyncratic terms and concepts by means of which it represents the information it receives How to resolve the incompatibilities which result when databases need to be merged? Compare: how to unify biology and chemistry?

17. 17 The term ‘ontology’ now used by information scientists to describe the building of standardized taxonomies which are designed to make databases mutually compatible

18. 18 An ‘ontology’ is a dictionary of terms formulated in a canonical syntax and with commonly accepted definitions and axioms

19. 19 How has this idea been realized? How have information systems engineers built ontologies? From where did they take the term ‘ontology’?

20. 20 From Quine

21. 21 Quine: each natural science has its own catalogue of types of objects to the existence of which it is committed

22. 22 the study of ontology is identified with the study of such Ontological Commitments … for Quineans, the ontologist studies, not reality, but scientific theories

23. 23 Quineanism: ontology is the study of the ontological commitments or presuppositions embodied in the different natural sciences

24. 24 In the hands of information scientists this is transformed into the view that ontology should study the concepts people use Ontology becomes the study of concept systems or conceptual models (becomes almost a branch of psychology)

25. 25 Arguments for Ontology as Conceptual Modeling Ontology is hard. Life is short. Let’s do conceptual modeling instead

26. 26 Ontological engineers thus neglect the standard of truth in favor of other, putatively more practical standards: above all programmability

27. 27 For an information system ontology there is no reality other than the one created through the system itself, so that the system is, by definition, correct

28. 28 Ontological engineering concerns itself with concept systems It does not care whether these are true of some independently existing reality.

29. 29 ‘Ontology’ is tremendously popular in information systems research today … b u t

30. 30 ATTEMPTS TO SOLVE THE TOWER OF BABEL PROBLEM VIA ONTOLOGIES AS CONCEPTUAL MODELS HAVE FAILED

31. 31 To see why let us consider some examples of concept systems in the medical domain

32. 32 Attempts at such standardization include: 1.UMLS 2.SNOMED 3.GALEN

33. 33 Example 1: UMLS Universal Medical Language System Very large taxonomy maintained by National Library of Medicine in Washington DC

34. 34 Example 1: UMLS 134 semantic types 800,000 concepts 10 million interconcept relationships UMLS is the product of fusion of several source vocabularies (built out of concept trees)

35. 35 Example 2: SNOMED-RT Systematized Nomenclature of Medicine A Reference Terminology with Legal Force

36. 36 Example 2: SNOMED-RT 121,000 concepts, 340,000 relationships “common reference point for comparison and aggregation of data throughout the entire healthcare process”

37. 37 Problems with UMLS and SNOMED Each is a ‘fusion’ of several source vocabularies, some of dubious quality They were fused without an ontological system being established first They contain circularities, taxonomic gaps, and unnatural ad hoc determinations

38. 38 Example 3: GALEN G eneralised A rchitecture for L anguages, E ncyclopaedias and N omenclatures in Medicine Applied especially to surgical procedures

39. 39 Problems with GALEN Ontology is ramshackle and has been subject to repeated fixes Unnaturalness makes coding slow and expensive, hence narrow scope Not gained wide acceptance

40. 40 Blood

41. 41 Representation of Blood in GALEN Blood has two states, LiquidBlood and CoagulatedBlood Blood SoftTissue DomainCategory Phenomenon Substance Tissue GeneralisedSubstanceSubstanceorPhysicalStructure

42. 42 Representation of Blood in UMLS Blood Tissue Entity Physical Object Anatomical Structure Fully Formed Anatomical Structure Body SubstanceBody FluidSoft Tissue Blood as tissue

43. 43 Representation of Blood in SNOMED Blood Liquid Substance Substance categorized by physical state Body fluid Body Substance Substance Blood as Fluid

44. 44 Database standardization is desparately needed in medicine … to enable the huge amounts of data resulting from clinical trials by different groups working on the same drugs/therapies/diagnostic methods …to be fused together

45. 45 How make ONE SYSTEM out of this? To reap the benefits of standardization we need to resolve such incompatibilities? But how? Not just by looking at the concepts underlying the respective systems For how, just by looking at separate concepts, could we establish how these concept systems relate to each other?

46. 46 different conceptual systems

47. 47 need not interconnect at all

48. 48 the only way to make them interconnect is by looking not just at concepts

49. 49 but also at the reality beyond

50. 50 How to solve the Tower of Babel Problem

51. 51 Look not at concept systems alone but at how concept systems relate to the world beyond

52. 52 Concept systems which are transparent to reality have a reasonable chance of being integrated together into a single ontological system

53. 53 This means we need to return to the traditional view of ontology

54. 54 … as a maximally opportunistic theory of reality Ontology should be modeled not on psychology but (as with Aristotle) on biology or chemistry

55. 55 Maximally opportunistic means: don’t just look at concepts look at the objects themselves towards which such concepts are directed

56. 56 … look at the objects from every available direction both formal and informal scientific and non-scientific empirical and theoretical attempting always to establish how these objects hang together ontologically

57. 57 Maximally opportunistic means: look at concepts critically and always in such a way as to include independent ways to access the objects themselves

58. 58 How to test a philosophical theory empirically?

59. 59 IFOMIS Institute for Formal Ontology and Medical Information Science Faculty of Medicine University of Leipzig

60. 60 IFOMIS in collaboration with those groups of ontological engineers who have recognized that they can improve their methods by drawing on the results of the philosophical work in ontology carried out over the last 2000 years

61. 61 … above all: LADSEB, Padua/Trento ITBM-CNR, Rome ONTEK Corporation Language and Computing EV, Belgium

62. 62 It will develop medical ontologies at different levels of granularity: cell ontology drug ontology * protein ontology gene ontology * * = already exists (but in a variety of mutually incompatible forms)

63. 63 and also anatomical ontology * epidemiological ontology disease ontology therapy ontology pathology ontology *

64. 64 physician’s ontology patient’s ontology and even hospital management ontology * together with

65. 65 Ontology like cartography must work with maps at different scales How fit these maps (conceptual grids) together into a single system?

66. 66

67. 67 Consider them as grids transparent to reality allowing our directedness towards objects beyond

68. 68 Cartographic Projection

69. 69 Optical Projection

70. 70 conceptual grids treated always only as mediators towards objects in reality intentionality = the directedness towards objects via conceptual grids object

71. 71 intentionality = the directedness towards concepts concepts

72. 72 conceptual grids treated always only as mediators towards objects in reality intentionality = the directedness towards objects via conceptual grids object

73. 73 Intentional directedness … is effected via conceptual grids we are able to reach out to the objects themselves because our conceptual grids are transparent

74. 74 Kantianism = the inability to appreciate the fact that our conceptual grids can be transparent to reality beyond

75. 75 there are many compatible map-like partitions at different scales, which are all transparent to the reality beyond

76. 76 Universe/Periodic Table animal bird canary ostrich fish ontology of biological species ontology of DNA space

77. 77 Universe/Periodic Table animal bird canary ostrich fish both are transparent partitions of one and the same reality

78. 78 Ontological Zooming

79. 79 The job of the ontologist is to understand how different partitions of the same reality interrelate

80. 80 IFOMIS partners in Leipzig Coordination centre for clinical trials Competence Network Malignant Lymphomas Project group HaematoWork (Dynamic knowledge based Workflow Adaption) Research project WISMA (Ontology-based Clinical Trial Management)

81. 81 WISMA Motivation and Objectives Faster and effective registration of clinical trial data Improvement of data quality Improvement of existing communication infrastructures Support clinical trial management in regard to: Information Transactions Definitions and Interpretations Evaluation better feedback and collaboration higher quality

82. 82 Competence Network Malignant Lymphomas Established clinical trial groups Hodgkin-Lymphomas High-malignant Non-Hodgkin-Lymphomas Low-malignant Non-Hodgkin-Lymphomas with over 30 protocols with up to 300 clinics/practitioners about 10,000 new cases per year

83. data input, requests information, acknowledgement information and communication portal notification documentation depositions documents participants' specific query notification findings material clinical trial centres clinical trial groups: HD lgNHL hgNHL CLL security area DB security area DBdirectory communication and information centre diverse information software tools therapy handbook material database partners sp 1 sp3 sp4 sp5 sp9 security area DB participants patients and relatives, hospitals, clinics and oncology specialists pathologists and radiotherapists images judgement

84. 84 How to test a philosophical theory empirically

85. 85 The Tests Uniform top-level ontology for medicine applicable at distinct granularities Test-case development of partial medical domain ontologies applied to: Standardization of clinical trial protocols Clinical trial dictionary Processing of unstructured patient records (www.landc.be)

86. 86 The Goals Uniform top-level ontology for testing in medical domain ONE YEAR Applicable at distinct granularities (e.g. gene ontology) FOUR YEARS Standardization of clinical trial protocol TWO YEARS Clinical trial Merkmal-dictionary TWO YEARS Processing of unstructured patient records (www.landc.be) THREE YEARS

87. 87 Measures of Success Uniform top-level ontology for medicine NO COMPETITOR applicable at distinct granularities NO COMPETITOR Standardization of clinical trial protocol NO SERIOUS COMPETITOR Clinical trial Merkmal-dictionary NO COMPETITOR Processing of unstructured patient records MANY COMPETITORS, BUT GOOD MEASURES OF EFFECTIVENESS

88. 88 The End