Thesis Chapters by Will Bridewell
Anomalous data lead to scientific discoveries. Although machine learning systems can be forced to... more Anomalous data lead to scientific discoveries. Although machine learning systems can be forced to resolve anomalous data, these systems use general learning algorithms to do so. To determine whether anomaly-driven approaches to discovery produce more accurate models than the standard approaches, we built a program called Kalpana. We also used Kalpana to explore means for identifying those anomaly resolutions that are acceptable to domain experts. Our experiments indicated that anomaly-driven approaches can lead to a richer set of model revisions than standard methods. Additionally we identified semantic and syntactic measures that are significantly correlated with the acceptability of model revisions. These results suggest that by interpreting data within the context of a model and by interpreting model revisions within the context of domain knowledge, discovery systems can more readily suggest accurate and acceptable anomaly resolutions.
Papers by Will Bridewell
If arti¯cial agents are to be created such that they occupy space in our social and cultural mili... more If arti¯cial agents are to be created such that they occupy space in our social and cultural milieu, then we should expect them to be targets of folk psychological explanation. That is to say, their behavior ought to be explicable in terms of beliefs, desires, obligations, and especially intentions. Herein, we focus on the concept of intentional action, and especially its relationship to consciousness. After outlining some lessons learned from philosophy and psychology that give insight into the structure of intentional action, we¯nd that attention plays a critical role in agency, and indeed, in the production of intentional action. We argue that the insights o®ered by the literature on agency and intentional action motivate a particular kind of computational cognitive architecture, and one that hasn't been well-explicated or computationally°eshed out among the community of AI researchers and computational cognitive scientists who work on cognitive systems. To give a sense of what such a system might look like, we present the ARCADIA attention-driven cognitive system as¯rst steps toward an architecture to support the type of agency that rich humanÀmachine interaction will undoubtedly demand.
AI Magazine, 2017
Over the decades, the view of agency in artificial intelligence (AI) has narrowed to one that emp... more Over the decades, the view of agency in artificial intelligence (AI) has narrowed to one that emphasizes acting in a way that maximizes reward. This perspective fails to make contact with the broader academic and legal communities where agency is bound up with personal accountability. To explore this gap in meaning, we introduce a spectrum of control that characterizes standard approaches to constructing agents and points the way toward agents that can be held responsible. The linchpin that enables agents to control their actions in the “right way” is attention. Broadly construed, attention lets an agent that is responsive to its environment consider the relationships among its actions, goals, and norms while also avoiding distraction. This ability enables strategic norm violations and opens the door to artificial, human-level agency.
Robot Ethics 2.0: From Autonomous Cars to Artificial Intelligence, 2017
It is easy to see that social robots will need the ability to detect and evaluate deceptive speec... more It is easy to see that social robots will need the ability to detect and evaluate deceptive speech; otherwise they will be vulnerable to manipulation by malevolent humans. More surprisingly, we argue that effective social robots must also be able to produce deceptive speech. Many forms of technically deceptive speech perform a positive pro-social function, and the social integration of artificial agents will be possible only if they participate in this market of constructive deceit. We demonstrate that a crucial condition for detecting and producing deceptive speech is possession of a theory of mind. Furthermore, strategic reasoning about deception requires identifying a type of goal distinguished by its priority over the norms of conversation, which we call an ulterior motive. We argue that this goal is the appropriate target for ethical evaluation, not the veridicality of speech per se. Consequently, deception-capable robots are compatible with the most prominent programs to ensure that robots behave ethically.
We define the inductive process modeling task as the automated construction of quantitative proce... more We define the inductive process modeling task as the automated construction of quantitative process models from time series and background knowledge. In this task, the background knowledge comprises generic processes that along with a given set of entities define the space of candidate model structures. Typically this space grows exponentially with the size of the library, so past research introduced a hierarchical organization on the processes to constrain that space to a limited set of plausible configurations. However, organizing the processes into a hierarchy takes considerable effort, leads to implicit constraints, and creates a complex relationship between the knowledge of what processes exist and the knowledge of how one can combine them. To address these problems, we developed SC-IPM 1 , an inductive process modeler that uses declarative constraints to reduce the size of the model structure space. In this paper, we describe the constraint formalism and how it guides SC-IPM's search.
Building models of a complex system such as an ecosystem or a chemical plant is an arduous task t... more Building models of a complex system such as an ecosystem or a chemical plant is an arduous task that can take several person months to complete. One rarely knows the scope of the model, its assumptions and claims, at the outset of the task, let alone how to state those in a formal language. To make this task manageable, modelers start at the whiteboard -by making free-form drawings that capture their current understanding of the studied system. These drawings need not conform to any particular ontology and may lack internal coherency or consistency. Nevertheless, such drawings can help organize one's thoughts and can capture key participants and relationships in the dynamic system. We argue that these free-form drawings facilitate the modeling process, based on evidence from modeling in practice. We analyze the relationship between free-form drawings and formally encoded models. We then suggest how to exploit these relationships to develop a modeling environment that supports a tighter integration between conceptual and detailed modeling.
Science comprises some of the most challenging cognitive tasks in which humans engage, which make... more Science comprises some of the most challenging cognitive tasks in which humans engage, which makes it a natural target for AI research. first proposed the idea that we might explain scientific discovery in computational terms and automate the processes involved on a computer. DENDRAL (Feigenbaum et al., 1971) demonstrated this by inferring the structures of organic molecules from mass spectra, a problem previously solved only by experienced chemists. Somewhat later, AM and Langley's (1981) BACON rediscovered a number of conjectures and laws from the history of mathematics and science. Research continued during the 1980s, leading to multiple books on the topic (e.g., Shrager & Langley, 1990). Research in this period also focused on historical examples, but the 1990s saw repeated application of these ideas to discover new scientific knowledge, as Langley has recounted.
In previous publications, we have reported a computational approach to constructing quantitative ... more In previous publications, we have reported a computational approach to constructing quantitative process models of dynamic systems from time-series data and background knowledge. However, our experience with these systems suggests that process knowledge is insufficient to avoid the consideration of implausible models. To this end, we have identified and introduced constraints that specify which processes can or must occur together, which in turn limit search to candidates that scientists will consider acceptable. We have also developed methods for inducing such constraints from the results of search through the model space. We maintain that the ability to specify, utilize, and induce constraints on quantitative process models will support deeper understanding of complex systems and constitutes an important addition to eScience.
In this paper, we review the paradigm of inductive process modeling and examine its application t... more In this paper, we review the paradigm of inductive process modeling and examine its application to human physiology. This framework represents models as a set of interacting processes, each with associated differential or algebraic equations that express causal relations among variables. Simulating such a quantitative process model produces trajectories for variables over time that one can compare to observations. Background knowledge about candidate processes enables search through the space of model structures and their associated parameters, and thus identify quantitative models that explain time-series data. We present an initial process model for aspects of human physiology, consider its uses for health monitoring, and discuss the induction of such models. In closing, we consider related efforts on physiological modeling and our plans for collecting data to evaluate our framework in this domain.
Network visualization of temporal data offers insights into the practical application of treatmen... more Network visualization of temporal data offers insights into the practical application of treatment guidelines. Using publicly available data on sequential HIV treatments, we apply network models to visualize switches in regimens and to understand when and if guideline-recommended regimens were used.
Supporting Innovative Construction of Explanatory Scientific Models
Science comprises some of the most challenging cognitive tasks in which humans engage, which make... more Science comprises some of the most challenging cognitive tasks in which humans engage, which makes it a natural target for AI research. first proposed the idea that we might explain scientific discovery in computational terms and automate the processes involved on a computer. DENDRAL (Feigenbaum et al., 1971) demonstrated this by inferring the structures of organic molecules from mass spectra, a problem previously solved only by experienced chemists. Somewhat later, AM and Langley's (1981) BACON rediscovered a number of conjectures and laws from the history of mathematics and science. Research continued during the 1980s, leading to multiple books on the topic (e.g., Shrager & Langley, 1990). Research in this period also focused on historical examples, but the 1990s saw repeated application of these ideas to discover new scientific knowledge, as Langley has recounted.
Process Modeling Framework: An Introduction
We define the inductive process modeling task as the automated construction of quantitative proce... more We define the inductive process modeling task as the automated construction of quantitative process models from time series and background knowledge. In this task, the background knowledge comprises generic processes that along with a given set of entities define the space of candidate model structures. Typically this space grows exponentially with the size of the library, so past research introduced a hierarchical organization on the processes to constrain that space to a limited set of plausible configurations. However, organizing the processes into a hierarchy takes considerable effort, leads to implicit constraints, and creates a complex relationship between the knowledge of what processes exist and the knowledge of how one can combine them. To address these problems, we developed SC-IPM 1 , an inductive process modeler that uses declarative constraints to reduce the size of the model structure space. In this paper, we describe the constraint formalism and how it guides SC-IPM's search.
In previous publications, we have reported a computational approach to constructing quantitative ... more In previous publications, we have reported a computational approach to constructing quantitative process models of dynamic systems from time-series data and background knowledge. However, our experience with these systems suggests that process knowledge is insufficient to avoid the consideration of implausible models. To this end, we have identified and introduced constraints that specify which processes can or must occur together, which in turn limit search to candidates that scientists will consider acceptable. We have also developed methods for inducing such constraints from the results of search through the model space. We maintain that the ability to specify, utilize, and induce constraints on quantitative process models will support deeper understanding of complex systems and constitutes an important addition to eScience.
Whither History of Science in Discovery Informatics?
In this paper, we review the paradigm of inductive process modeling and examine its application t... more In this paper, we review the paradigm of inductive process modeling and examine its application to human physiology. This framework represents models as a set of interacting processes, each with associated differential or algebraic equations that express causal relations among variables. Simulating such a quantitative process model produces trajectories for variables over time that one can compare to observations. Background knowledge about candidate processes enables search through the space of model structures and their associated parameters, and thus identify quantitative models that explain time-series data. We present an initial process model for aspects of human physiology, consider its uses for health monitoring, and discuss the induction of such models. In closing, we consider related efforts on physiological modeling and our plans for collecting data to evaluate our framework in this domain.
Social cognition is a key feature of human-level intelligence. However, social reasoning facultie... more Social cognition is a key feature of human-level intelligence. However, social reasoning faculties are rarely included in cognitive systems. To encourage research in this direction, we introduce a practical, computational framework that enables socially aware inference. We demonstrate the framework's ability to model a common, complex, and under-investigated aspect of human social behavior: deception. Moreover, we show how a system implementing this framework could dynamically respond once it has detected a lie. We then discuss some of the challenges associated with deception, ending with an outline of future research directions.
Building models of a complex system such as an ecosystem or a chemical plant is an arduous task t... more Building models of a complex system such as an ecosystem or a chemical plant is an arduous task that can take several person months to complete. One rarely knows the scope of the model, its assumptions and claims, at the outset of the task, let alone how to state those in a formal language. To make this task manageable, modelers start at the whiteboard -by making free-form drawings that capture their current understanding of the studied system. These drawings need not conform to any particular ontology and may lack internal coherency or consistency. Nevertheless, such drawings can help organize one's thoughts and can capture key participants and relationships in the dynamic system. We argue that these free-form drawings facilitate the modeling process, based on evidence from modeling in practice. We analyze the relationship between free-form drawings and formally encoded models. We then suggest how to exploit these relationships to develop a modeling environment that supports a tighter integration between conceptual and detailed modeling.
In this paper, we explore the representational and inferential requirements for supporting a rich... more In this paper, we explore the representational and inferential requirements for supporting a rich
notion of belief revision. Our analysis extends beyond the typical case of a single agent revising its
beliefs in light of new information into the realm of social engagement. More to the point, we argue
that, although belief revision mechanisms surely operate at the level of single agents, we must also
consider the need to lift an agent’s understanding of the belief revision process to the knowledge
level in order to intentionally guide other agents’ revision processes with whom it socially interacts.
In exploring belief revision at the knowledge level, we identify reasons for rejecting classical
formulations of the problem and identify constraints by which alternative accounts must abide.
Uploads
Thesis Chapters by Will Bridewell
Papers by Will Bridewell
notion of belief revision. Our analysis extends beyond the typical case of a single agent revising its
beliefs in light of new information into the realm of social engagement. More to the point, we argue
that, although belief revision mechanisms surely operate at the level of single agents, we must also
consider the need to lift an agent’s understanding of the belief revision process to the knowledge
level in order to intentionally guide other agents’ revision processes with whom it socially interacts.
In exploring belief revision at the knowledge level, we identify reasons for rejecting classical
formulations of the problem and identify constraints by which alternative accounts must abide.