Cameron Buckner - phd

The last five years have seen a series of remarkable achievements in deep-neural-network-based Artificial Intelligence (AI) research, and some modellers have argued that their performance compares favourably to human cognition. Critics, however, have argued that processing in deep neural networks is unlike human cognition for four reasons: they are i) data-hungry, ii) brittle, and iii) inscrutable black boxes that merely iv) reward-hack rather than learn real solutions to problems. This paper rebuts these criticisms by exposing comparative bias within them, in the process extracting some more general lessons that may also be useful for future debates.

Deep neural networks are currently the most widespread and successful technology in artificial intelligence. However, these systems exhibit bewildering new vulnerabilities: most notably a susceptibility to adversarial examples. Here, I review recent empirical research on adversarial examples that suggests that deep neural networks may be detecting in them features that are predictively useful, though inscrutable to humans. To understand the implications of this research, we should contend with some older philosophical puzzles about scientific reasoning, helping us to determine whether these features are reliable targets of scientific investigation or just the distinctive processing artefacts of deep neural networks.

Deep learning is currently the most prominent and widely successful method in artificial intelligence. Despite having played an active role in earlier artificial intelligence and neural network research, philosophers have been largely silent on this technology so far. This is remarkable, given that deep learning neural networks have blown past predicted upper limits on artificial intelligence performance—recognizing complex objects in natural photographs and defeating world champions in strategy games as complex as Go and chess—yet there remains no universally accepted explanation as to why they work so well. This article provides an introduction to these networks as well as an opinionated guidebook on the philosophical significance of their structure and achievements. It argues that deep learning neural networks differ importantly in their structure and mathematical properties from the shallower neural networks that were the subject of so much philosophical reflection in the 1980s and 1990s. The article then explores several different explanations for their success and ends by proposing three areas of inquiry that would benefit from future engagement by philosophers of mind and science.

Ontology evaluation poses a number of diffcult challenges requiring different evaluation methodologies, particularly for a "dynamic ontology" generated by a combination of automatic and semi-automatic methods. We review evaluation methods that focus solely on syntactic (formal) correctness, on the preservation of semantic structure, or on pragmatic utility. We propose two novel methods for dynamic ontology evaluation and describe the use of these methods for evaluating the different taxonomic representations that are generated at different times or with different amounts of expert feedback. These methods are then applied to the Indiana Philosophy Ontology (InPhO), and used to guide the ontology enrichment process.
 

The application of digital humanities techniques to philosophy is changing the way scholars approach the discipline. This paper seeks to open a discussion about the difficulties, methods, opportunities, and dangers of creating and utilizing a formal representation of the discipline of philosophy. We review our current project, the Indiana Philosophy Ontology (InPhO) project, which uses a combination of automated methods and expert feedback to create a dynamic computational ontology for the discipline of philosophy. We argue that our distributed, expert-based approach to modeling the discipline carries substantial practical and philosophical benefits over alternatives. We also discuss challenges facing our project (and any other similar project) as well as the future directions for digital philosophy afforded by formal modeling. 
 

Recent experimental philosophy arguments have raised trouble for philosophers' reliance on armchair intuitions. One popular line of response has been the expertise defense: philosophers are highly-trained experts, whereas the subjects in the experimental philosophy studies have generally been ordinary undergraduates, and so there's no reason to think philosophers will make the same mistakes. But this deploys a substantive empirical claim, that philosophers' training indeed inculcates sufficient protection from such mistakes. We canvass the psychological literature on expertise, which indicates that people are not generally very good at reckoning who will develop expertise under what circumstances. We consider three promising hypotheses concerning what philosophical expertise might consist in: (i) better conceptual schemata; (ii) mastery of entrenched theories; and (iii) general practical know-how with the entertaining of hypotheticals. On inspection, none seem to provide us with good reason to endorse this key empirical premise of the expertise defense.
 

The "wisdom of crowds" is accomplishing tasks that are cumbersome for individuals yet cannot be fully automated by means of specialized computer algorithms. One such task is the construction of thesauri and other types of concept hierarchies. Human expert feedback on the relatedness and relative generality of terms, however, can be aggregated to dynamically construct evolving concept hierarchies. The InPhO (Indiana Philosophy Ontology) project bootstraps feedback from volunteer users unskilled in ontology design into a precise representation of a specific domain. The approach combines statistical text processing methods with expert feedback and logic programming to create a dynamic semantic representation of the discipline of philosophy. In this paper, we show that results of comparable quality can be achieved by leveraging the workforce of crowdsourcing services such as the Amazon Mechanical Turk (AMT). In an extensive empirical study, we compare the feedback obtained from AMT's workers with that from the InPhO volunteer users providing an insight into qualitative differences of the two groups. Furthermore, we present a set of strategies for assessing the quality of different users when gold standards are missing. We finally use these methods to construct a concept hierarchy based on the feedback acquired from AMT workers.

The Indiana Philosophy Ontology (InPhO) project is presented as one of the first social-semantic web endeavors which aims to bootstrap feedback from users unskilled in ontology design into a precise representation of a specific domain. Our approach combines statistical text processing methods with expert feedback and logic programming approaches to create a dynamic semantic representation of the discipline of philosophy. We describe the basic principles and initial experimental results of our system.
 

InPhO is a system that combines statistical text processing, information extraction, human expert feedback, and logic programming to populate and extend a dynamic ontology for the field of philosophy. Integrated in the editorial work flow of the Stanford Encyclopedia of Philosophy (SEP), it will provide important metadata features such as automated generation of cross-references, semantic search, and ontology driven conceptual navigation.
 

The Indiana Philosophy Ontology (InPhO) is a "dynamic ontology" for the domain of philosophy derived from human input and software analysis. The structured nature of the ontology supports machine reasoning about philosophers and their ideas. It is dynamic because it tracks changes in the content of the online Stanford Encyclopedia of Philosophy. This paper discusses ways of managing the varying expertise of people who supply input to the InPhO and provide feedback on the automated methods.
 

The next generation of online reference works will require structured representations of their contents in order to support scholarly functions such as semantic search, automated generation of cross-references, tables of contents, and ontology-driven conceptual navigation. Many of these works can be expected to contain massive amounts of data and be updated dynamically, which limits the feasibility of "manually" coded ontologies to keep up with changes in content. However, relationships relevant to inferring an ontology can be recovered from statistical text processing, and these estimates can be verified with carefully-solicited expert feedback. In this paper, we explain a method by which we have used answer set programming on such expert feedback to dynamically populate and partially infer an ontology for a well-established, open-access reference work, the Stanford Encyclopedia of Philosophy.
 

This paper describes the design of new algorithms and the adjustment of existing algorithms to support the automated and semi-automated management of domain-rich metadata for an established digital humanities project, the Stanford Encyclopedia of Philosophy. Our approach starts with a "hand-built" formal ontology that is modified and extended by a combination of automated and semi-automated methods, thus becoming a "dynamic ontology". We assess the suitability of current information retrieval and information extraction methods for the task of automatically maintaining the ontology. We describe a novel measure of term-relatedness that appears to be particularly helpful for predicting hierarchical relationships in the ontology. We believe that our project makes a further contribution to information science by being the first to harness the collaboration inherent in a expert-maintained dynamic reference work to the task of maintaining and extending a formal ontology. We place special emphasis on the task of bringing domain expertise to bear on all phases of the development and deployment of the system, from the initial design of the software and ontology to its dynamic use in a fully operational digital reference work.
 

The goals of the Indiana Philosophy Ontology (InPhO) project are to build and maintain a "dynamic ontology" for the discipline of philosophy, and to deploy this ontology in a variety of digital philosophy applications. Automated information-retrieval methods are combined with human feedback to build and manage a machine-readable representation (i.e., a "formal ontology") of the relations among philosophical ideas and thinkers. The applications we hope to develop that will employ the ontology include automatic generation of cross-references for Stanford Encyclopedia of Philosophy (SEP) articles, semantic search of the SEP and other philosophical resources (including guided searching with Noesis), conceptual navigation through the SEP using information visualization techniques, and web access to the biographical and citational information contained in the InPhO. Moreover, we will archive the dynamically generated versions of the ontology.