Probabilistic Logics in Foresight
Panula-Ontto, Juha (2019)
Panula-Ontto, Juha
Tampereen yliopisto
2019
Tietojenkäsittelyoppi - Computer Science
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Väitöspäivä
2019-05-14
Julkaisun pysyvä osoite on
https://urn.fi/URN:ISBN:978-952-03-1059-2
https://urn.fi/URN:ISBN:978-952-03-1059-2
Tiivistelmä
A prudent decision-maker facing a complicated strategic decision considers the factors relevant to the decision, gathers information about the identified factors, and attempts to formulate the best course of action based on the available information. Careful consideration of any alternative course of action might reveal that in addition to the desirable intended consequences, a number of less desirable outcomes are likely to follow as well. Facing a complicatedly entangled net of considerations, entwined positive and negative outcomes, and uncertainty, the decision-maker will attempt to organize the available information and make the decision by using some strategy of reasoning on the information.
A logic is away of reasoning adherent to rules, based on structured knowledge. A modeling language and inference rules comprise a logic. The language of a logic is formal, consisting of a defined set of building blocks having well defined meanings. The decision-maker can use a modeling language to describe the information pertinent to the decision-making problem, and organize the information by giving it a structure, which specifies the relationships between the individual considerations. While reasoning about the extensive amount of information in its disorganized form may be overwhelming, in a structured form the information becomes much more useful for the decision-maker, as nowit can be analyzed in a systematic fashion. Inference is systematic reasoning about structured information. As the information is described in a formal and structured way and the process of reasoning about it is systematic, the inference may be automated. Computational inference permits reasoning that would not be possible by intuition in cases where the amount of considerations and their interdependencies exceeds human cognitive capacity. The decision-maker may direct the efforts to describing the decision factors and knowledge with the formal language, with a narrower and more manageable frame of attention, and perform the inference with a computer.
Probabilistic language gives room for haziness in knowledge description, and is thus suitable for describing knowledge originating from humans, conveyed to the decision-maker in a non-formal format, such as viewpoints and opinions. Many domains of decision-making and planning use human sourced knowledge, especially if the informants are knowledgeable people or experts with relevant, developed understanding on the domain issues. The expert views can augment the knowledge bases in cases where other forms of information, such as empirical or statistical data, are lacking or completely absent, or do not capture or represent considerations important for the decision-making. This is a typical setting for strategic decision-making, long range planning, and foresight, which have to account for developments and phenomena that do not yet exist in the form they might in the future, or at all.
This work discusses approaches for decision support and foresight oriented modeling of expert knowledge bases and inference based on such knowledge bases. Two novel approaches developed by the author are presented and positioned against previous work on cross-impact analysis, structural and morphological analysis, and Bayesian networks. The proposed approaches are called EXIT and AXIOM. EXIT is a conceptually simple approach for structural analysis, based on a previously unutilized computational process for discovery of higher-order influences in a structural model. The analytical output is, in relation to comparable approaches, easier to interpret considering the causal information content of the structural model. AXIOM is a versatile probabilistic logic, combining ideas of structural analysis, morphological analysis, cross-impact analysis and Bayesian belief networks. It provides outputs comparable to Bayesian networks, but has higher fitness for full model parameterization through expert elicitation. A guiding idea of the methodological development work has been that the slightly aged toolset of cross-impact analysis can be updated, improved and extended, and brought to be more interoperable with the Bayesian approach.
A logic is away of reasoning adherent to rules, based on structured knowledge. A modeling language and inference rules comprise a logic. The language of a logic is formal, consisting of a defined set of building blocks having well defined meanings. The decision-maker can use a modeling language to describe the information pertinent to the decision-making problem, and organize the information by giving it a structure, which specifies the relationships between the individual considerations. While reasoning about the extensive amount of information in its disorganized form may be overwhelming, in a structured form the information becomes much more useful for the decision-maker, as nowit can be analyzed in a systematic fashion. Inference is systematic reasoning about structured information. As the information is described in a formal and structured way and the process of reasoning about it is systematic, the inference may be automated. Computational inference permits reasoning that would not be possible by intuition in cases where the amount of considerations and their interdependencies exceeds human cognitive capacity. The decision-maker may direct the efforts to describing the decision factors and knowledge with the formal language, with a narrower and more manageable frame of attention, and perform the inference with a computer.
Probabilistic language gives room for haziness in knowledge description, and is thus suitable for describing knowledge originating from humans, conveyed to the decision-maker in a non-formal format, such as viewpoints and opinions. Many domains of decision-making and planning use human sourced knowledge, especially if the informants are knowledgeable people or experts with relevant, developed understanding on the domain issues. The expert views can augment the knowledge bases in cases where other forms of information, such as empirical or statistical data, are lacking or completely absent, or do not capture or represent considerations important for the decision-making. This is a typical setting for strategic decision-making, long range planning, and foresight, which have to account for developments and phenomena that do not yet exist in the form they might in the future, or at all.
This work discusses approaches for decision support and foresight oriented modeling of expert knowledge bases and inference based on such knowledge bases. Two novel approaches developed by the author are presented and positioned against previous work on cross-impact analysis, structural and morphological analysis, and Bayesian networks. The proposed approaches are called EXIT and AXIOM. EXIT is a conceptually simple approach for structural analysis, based on a previously unutilized computational process for discovery of higher-order influences in a structural model. The analytical output is, in relation to comparable approaches, easier to interpret considering the causal information content of the structural model. AXIOM is a versatile probabilistic logic, combining ideas of structural analysis, morphological analysis, cross-impact analysis and Bayesian belief networks. It provides outputs comparable to Bayesian networks, but has higher fitness for full model parameterization through expert elicitation. A guiding idea of the methodological development work has been that the slightly aged toolset of cross-impact analysis can be updated, improved and extended, and brought to be more interoperable with the Bayesian approach.
Kokoelmat
- Väitöskirjat [4885]