Chance Discovery

The fundamental philosophy of chance discovery is introduced. By comparison with the cyclic model of knowledge discovery, this chapter describes the essentials for realizing chance discovery. From these discussions, three keys for chance discovery are proposed, i.e. communication, context shifting, and data mining. As a result, the double helix and the subsumption architecture are presented as methods for realizing chance discovery.

Recognition of chance can be influenced by a person’s situations or surroundings at a specific time. It is useful to know the situations and contexts in which people use the word chance because, by considering the relationships between a sentence including the word chance and the context, we can construct the logic of our recognition of chance. In this paper we have collected texts which include the word from recent articles in newspapers and magazines, and, taking its etymology into consideration, have explored the use of the word and its implications. The sentences and the contexts show the ways we recognize some events as chances.

This chapter addresses the relation and difference between prediction, forecasting, and chance discovery. Prediction and forecasting have a long history. So far, many studies have been devoted to prediction and forecasting. However, in complex real-world systems, contrary to scientific laws, it is sometimes very difficult to predict the future. In such situations, model creation, model selection, and parameter fitting are all important in the complex changing real world. Chance discovery targets three aspects that prediction and forecasting methods have not shed light on, i.e. emphasis on model and variable creation and discovery, emphasis on rare events, and emphasis on human and computer interaction.

The present chapter deals with chance discovery from the perspective of complex systems. We will limit ourselves to a type of complex systems known as self-organized critical systems; to discuss these systems we need to make clear what we mean by notions such as complex, scale free, critical, and self-organization. To fix some ideas we will introduce two self-organizing complex models, i.e. a simple cellular automaton model and a model of evolutionary ecology. This highlights how seemingly innocent local perturbations may propagate through the entire system, totally altering the composition of the system. We consider the possibility of prediction in complex systems, and elaborate on the specifics of chance discovery in interconnected and highly sensitive complex systems.

Human-to-human communication can provide an opportunity for chance discovery. We tend to think ‘information to be communicated’ is one its sender wants to convey or one its receiver needs, and others are noise. However, information apt to be overlooked as noise may have potential business opportunities or be a latent pitfall to failure. Although people rarely notice such ‘noise information’, a person behaving as an agent of communication can bring this subtle sign to a business opportunity without passing it up, if he/she has a high level of awareness. This chapter describes chance discovery in the human-to-human communication in retail business and emails in R & D projects.

Chance discovery in online communities is the serendipitous meeting of two people with a background or interest in common. It is a solution to some problem that the community has, but that solution must come from without. In this chapter, we separate the area into three facets, chance discovery of online communities, between communities, and within a community. This separation is adequate to capture most contemporary research. We examine and illuminate the technological case where computer systems have been designed to actively assist humans in the discovery process.

This chapter shows how to derive postsemantic context based on vector representations of words (described in Chap.8). The core problem is to discover relevant word associations in relation to seed words in the utterance. This may involve uncovering implicit associations or reweighting explicit associations more highly. The set of such associations forms a part of ‘conversational implicature’. The chapter describes techniques for computing associations in a dimensional space that have shown promise in the literature. The goal is to provide some initial insights as to their usefulness for mining conversational implicature by applying them to a small set of email utterances.

This chapter considers chance discovery and management in a community of intelligent, autonomous, software agents, where agents may have differing beliefs and intentions. For such a community of agents, we derive a set of five requirements for the design of languages and protocols for communications between the agents when discussing chance discovery and management. We then use these requirements to assess two proposals in the multi-agent systems community for agent communications: generic languages, such as the FIPA ACL, and dialogue game protocols. The latter are found to have greater potential capability to support dialogues over chance discovery and management between autonomous agents.

If multiple autonomous entities — agents — are involved in chance discovery and management, then the agents involved may disagree as to what constitutes a chance event, and what action, if any, to take in response. One approach to agent communication in this situation is to insist that agents not only send messages, but also support them with reasons why those messages are appropriate. This is argumentation-based communication. In this chapter, we review some of our work on argumentation-based communication, discussing the issues we consider to be important in developing systems for argumentation-based communication between agents in chance discovery and management domains.

In this chapter, we consider the Web as the environment, on which how to help a human notice and evaluate topics is discussed. As the Web provides us with new topics constantly, it is difficult to grasp the trends or change of topics on the Web. Although the hugeness of the Web as well as its dynamic nature is a burden for the users, it will also bring them a chance for business and research if they can notice the trends or movement of the real world from the Web, which can not be found from a single document but from a sequence of document sets.

Scenic information is a source of chances, playing a significant role as the ‘key 2’ for chance discovery mentioned in Chap. 1. It stimulates the contextual shifts of a human, which carries him/her into deeper concerns with actions on chances to be discovered. In other words, scenic information guides a human to the imagination of new environments with new chances. This chapter shows the role of scenic information in up-to-date application domains.

Analogical reminding in humans and machines is a great source for chance discoveries because analogical reminding can produce representational change and thereby produce insights. Here, we present a new kind of representational change associated with analogical reminding called packing. We derived the algorithm in part from human data we have on packing. Here, we explain packing and its role in analogy making, and then present a computer model of packing in a micro-domain. We conclude that packing is likely to be used in human chance discoveries, and is needed if our machines are to make their own chance discoveries.

“Chance favors only the prepared mind.” — Louis Pasteur

In this chapter, we first introduce abduction and analogy as a discovery reasoning. Second, we show a hypothetical reasoning system, Theorist, as an example of computational abductive reasoning. This hypothetical reasoning system can be applied to explanatory reasoning such as design and diagnosis. However, it can not generate new hypotheses. In our explanation of hypothetical reasoning, we also show the possibilities and the limitations of conventional abduction when we use it in the context of chance discovery. Third, we show abductive analogical reasoning (AAR), which can generate new hypotheses. AAR is an extension of hypothetical reasoning that is achieved by combining abduction and analogical mapping. Finally, we show AAR as a tool for chance discovery and explain the roles of abduction and analogy in chance discovery.

In this chapter, the criteria for information selection and two-dimensional interfaces for application are presented. We require multi-sided analysis for shrinking enormous amounts of data, and we should understand visualized outputs by intuition for acquiring new ideas. Visualization methods will be heavily relied upon for aiding humans’ comprehension by intuition, and promoting the urge for finding biased outputs. Such a system will provide clues for discovering chances.

In this chapter, we see whether chance discovery in the form of KeyGraphs can be used to reveal deep building blocks for competent genetic algorithms(GAs), thereby speeding up innovation in particularly difficult problems. On an intellectual level, showing the connection between KeyGraphs and genetic algorithms as related pieces of the innovation puzzle is both scientifically and computationally interesting. GAs represent that aspect of human innovation that tries to innovate through the exchange or cross fertilization of notions contained in different ideas; the KeyGraph procedure represents that portion of human innovation that pays special attention to and interprets salient fortuitous events. The chapter goes beyond mere conjecture and performs pilot studies that show how KeyGraphs and competent GAs can work together to solve the problem of deep building blocks; the work is promising and steps toward a practical computational combination of the two procedures are suggested.

This chapter presents a notion of enhancing our daily conversations for increasing opportunities to encounter new ideas and future partners for collaboration. We show two systems. One is AIDE, a system which facilitates online discussion with visualization of the discussion structure and a virtual discussant. The users of AIDE can mutually notice the similarity and difference among their viewpoints against common topics. The other is AgentSalon, a system which facilitates casual face-to-face chatting in the real-space setting such as museums and conference sites. AgentSalon has a big screen showing conversations among animated agents belonging to users. By observing a chat of the agents, the users can effectively obtain appropriate topics: that is, it tempts them to follow the chat.

KeyGraph, the document-indexing (keyword-extraction) algorithm, is applied for another purpose: extracting active faults with risks of near-future big earthquakes from earthquake sequences. This presents an exemplification of KeyGraph as a tool for abstracting causalities from an event sequence. From the results here, we can validate KeyGraph as a tool for showing which active faults are risky, as well as for showing which words abstract a document finely. The risky faults empirically obtained by KeyGraph correspond finely with the real occurrence of earthquakes and seismologists’ estimation of risks. Even more important, the risks are accountable. That is, the relation between the obtained risky faults and faults quaking frequently are visualized by the output of KeyGraph.

In this chapter we apply the double helix introduced in Chap. 1 to model the process of chance discovery, applied to the survey of peoples’ behaviors on the information from the Internet and in the real world being motivated by unknown factors. This, at the end of this chapter, leads to the process model of chance discovery. Simply abstracting the double helix (in Chap. 1), we review what the process of chance discovery means and how it helps social scientists.

This chapter focuses on quite a new direction in marketing, the power shift from suppliers to consumers, or the empowerment of consumers. The techniques for chance discovery can be applied to support not only suppliers’ but also consumers’ chance discovery. These two directions seem completely different but they are based more or less on the same underlying methodology, namely, analyzing consumer preferences and seeking potentially desirable states. We will discuss the background behind our research: recent trends in marketing, the increasing demand for recommender systems, and the applicability of chance discovery there. Then we will propose our own approach to supporting consumers’ chance discovery, and show the results of a user test of a prototype.

This chapter introduces a method of chance discovery, i.e. awareness of and explanation of a significant situation for decision making, to a group interview of housewives who daily make decisions to serve meals to their families. The experimental stimulus was a simple visual representation of family food consumption relations. The visual representation of rarely consumed foods worked as chances for the housewives’ awareness of unnoticed fundamental issues to be considered in meal service, and for their proposals for new services. Meaningful proposals then grew into acceptable service decisions, in the discussion which followed presentation of the visual stimulus.