Automatic generation of explanations: AGE

Humans seem to have a natural instinct for wanting to understand and make sense of their environment and things in it. In expert systems (ES), explanation can be used to clarify the reasoning process to users such that they can gain a better understanding of how the system functions. With the help of good explanation facilities, a user can know why an ES is asking a particular question, how the expert system will act if given a certain input, and how the ES reaches a particular conclusion. This is especially important when an ES application is used as a high level advisor to professionals who must retain responsibility for the decisions which are made. However, most ES explanation components require acquiring additional knowledge for explanation, thus increasing the effort of implementing an ES with explanation capabilities. The primary goal of this work is to present a methodology for automatically generating explanations during and at the end of the reasoning process. The developed explanation components can deal with different knowledge representation schemes that are used by problem solving methods namely, the "generate and confirm hypotheses" that is based on the CommonKADS methodology[1], and the routine design generic task . As a proof of concept the explanation components were developed and integrated into the agricultural expert system generic tool (AESGT) .The developed explanation components can be easily reused with expert systems developed by the tool to automatically generate explanation for the reasoning process.

ArXiv, 2021

Nowadays, it is growing interest to make Machine Learning (ML) systems more understandable and trusting to general users. Thus, generating explanations for ML system behaviours that are understandable to human beings is a central scientific and technological issue addressed by the rapidly growing research area of eXplainable Artificial Intelligence (XAI). Recently, it is becoming more and more evident that new directions to create better explanations should take into account what a good explanation is to a human user, and consequently, develop XAI solutions able to provide user-centred explanations. This paper suggests taking advantage of developing an XAI general approach that allows producing explanations for an ML system behaviour in terms of different and user-selected input features, i.e., explanations composed of input properties that the human user can select according to his background knowledge and goals. To this end, we propose an XAI general approach which is able: 1) to ...

2020

Neural networks lack the ability to reason about qualitative physics and so cannot generalize to scenarios and tasks unseen during training. We propose ESPRIT, a framework for commonsense reasoning about qualitative physics in natural language that generates interpretable descriptions of physical events. We use a two-step approach of first identifying the pivotal physical events in an environment and then generating natural language descriptions of those events using a data-to-text approach. Our framework learns to generate explanations of how the physical simulation will causally evolve so that an agent or a human can easily reason about a solution using those interpretable descriptions. Human evaluations indicate that ESPRIT produces crucial fine-grained details and has high coverage of physical concepts compared to even human annotations. Dataset, code and documentation are available at https://github.com/salesforce/esprit.

Artificial Intelligence, 1996

Effective problem solving requires building adequate models that embody the simplifications, abstractions, and approximations that parsimoniously describe the relevant system phenomena for tbe task at hand. Compositional modeling is a framework for constructing adequate device models by composiqg model fragments selected from a model fragment library. While model selection using compo,sitional modeling has been shown to be intractable, it is tractable when all model fragment approximations are causal upproximations. This paper addresses the reasoning and knowledge representation issues that arise in building practical systems for constructing adequate device models that provide parsimonious causal explanations of how a device functions. We make four important contributions. First, we present a representation of class level descriptions of model fragments and their relationships. The representation yields a practical model fragment library organization that facilitates knowledge base construction and supports focused generation of device models. Second, we show how the structural, behavioral, and functional contexts of the device define model adequacy and provide the task focus and additional constraints to guide the search for adequate models. Third, we describe a novel model selection algorithm that incorporates device behavior with order of magnitude reasoning and focuses model selection with component interaction heuristics. Fourth, we present the results of our implementation that produces adequate models and causal explanations of a variety of electromechanical devices drawn from a library of 20 components and 150 model fragments.

Computers in Human Behavior, 1996

Explanations were construed for an expert system in the domain of protein purification and based upon the multiple-explanation construction model (MEC model). Various explanations were construed covering different relevant aspects of the explanation space, expert-level explanations (quantitative representation), low-level explanations (qualitative representation), grounds explanations (background knowledge) and backing explanations (general abstract principles). These were tested on laboratory staff working at Pharmacia LKB Biotechnology AB in Uppsala, Sweden. The variables being tested were learning, understanding, usability, and novelty of the explanation types. The results indicate that the model is valuable in construing explanations with different "knowledge levels" with the purpose of fulfilling the needs of experts as well as "less-experts" covering different important aspects of the explanation space. In the context of learning, the results show that experts prefer expert-level explanations and low-level explanations whereas less-experts prefer a combination of all explanation types. A multiexplanation perspective has to be taken, where explanations covering different aspects of the explanation space on different levels have to be available to less-experts to facilitate learning from explanations in a specific complex domain. These results can have strong implications for learning, for example in the context of computer-supported education. Explanations Within Knowledge-Based Systems Explanations within knowledge-based systems have lately gained a lot of attention. One important aspect in this context is knowledge representation since the way knowledge is represented in a reasoning system, for example, strongly

The paper discusses the issue of generating explanations in intelligent tutoring systems. Specifically, it shows how explanations are generated according to the GET-BITS model of intelligent tutoring systems. The major concern is what software components are needed in order to generate meaningful explanations to different classes of the end-users of such systems. The process of explanation is considered in the context of the types of knowledge present in the knowledge base of an intelligent tutoring system. Throughout the paper, the process of explanation generation is treated from the software engineering point of view. Some design examples, describing several classes developed in support of explanation generation based on the GET-BITS model, are also presented.

2011

ABSTRACT Artificial intelligence research on creative design has led to Structure-Behavior-Function (SBF) models that emphasize functions as abstractions for organizing understanding of physical systems. Empirical studies on understanding complex systems suggest that novice understanding is shallow, typically focusing on their visible structures and showing minimal understanding of their functions and invisible causal behaviors.

Abstract To explain complex phenomena, an explanation system must be able to select information from a formal representation of domain knowledge, organize the selected information into multisentential discourse plans, and realize the discourse plans in text. Although recent years have witnessed significant progress in the development of sophisticated computational mechanisms for explanation, empirical results have been limited.

We present a domain-independent method for generation of natural language explanations of rules in expert systems. The method is based on explanatory rules written in a procedural formal language, which build the explanation from predefined natural language texts fragments. For better style, a specific text fragment is randomly selected from a group of synonymous expressions. We have implemented 16 groups of explanatory rules and 74 groups of explanatory texts containing about 200 text fragments.

We present a domain-independent method for generation of natural language explanations of rules in expert systems. The method is based on explanatory rules written in a procedural formal language, which build the explanation from predefined natural language texts fragments. For better style, a specific text fragment is randomly selected from a group of synonymous expressions. We have implemented 16 groups of explanatory rules and 74 groups of explanatory texts containing about 200 text fragments.