A global taxonomy of interpretable AI: unifying the terminology for the technical and social sciences

Since its emergence in the 1960s, Artificial Intelligence (AI) has grown to conquer many technology products and their fields of application. Machine learning, as a major part of the current AI solutions, can learn from the data and through experience to reach high performance on various tasks. This growing success of AI algorithms has led to a need for interpretability to understand opaque models such as deep neural networks. Various requirements have been raised from different domains, together with numerous tools to debug, justify outcomes, and establish the safety, fairness and reliability of the models. This variety of tasks has led to inconsistencies in the terminology with, for instance, terms such as interpretable, explainable and transparent being often used interchangeably in methodology papers. These words, however, convey different meanings and are "weighted" differently across domains, for example in the technical and social sciences. In this paper, we propose an overarching terminology of interpretability of AI systems that can be referred to by the technical developers as much as by the social sciences community to pursue clarity and efficiency in the definition of regulations for ethical and reliable AI development. We show how our taxonomy and definition of interpretable AI differ from the ones in previous research and how they apply with high versatility to several domains and use cases, proposing a-highly needed-standard for the communication among interdisciplinary areas of AI.

On machine vision and photographic imagination

In this article we introduce the concept of implied optical perspective in deep learning computer vision systems. Taking the BBC's experimental television programme “Made by Machine: When AI met the Archive” (2018) as a case study, we trace a conceptual and material link between the system used to automatically “watch” the television archive and a specific type of photographic practice. From a computational aesthetics perspective, we show how deep learning machine vision relies on photography, its technical regimes and epistemic advantages, and we propose a novel way to identify the latent camera through which the BBC archive was seen by machine.

Record linking in the EHRI portal

Purpose

This paper aims to describe the European Holocaust Research Infrastructure (EHRI) project's ongoing efforts to virtually integrate trans-national archival sources via the reconstruction of collection provenance as it relates to copy collections (material copied from one archive to another) and the co-referencing of subject and authority terms across material held by distinct institutions.

Design/methodology/approach

This paper is a case study of approximately 6,000 words length. The authors describe the scope of the problem of archival fragmentation from both cultural and technical perspectives, with particular focus on Holocaust-related material, and describe, with graph-based visualisations, two ways in which EHRI seeks to better integrate information about fragmented material.

Findings

As a case study, the principal contributions of this paper include reports on our experience with extracting provenance-based connections between archival descriptions from encoded finding aids and the challenges of co-referencing access points in the absence of domain-specific controlled vocabularies.

Originality/value

Record linking in general is an important technique in computational approaches to humanities research and one that has rightly received significant attention from scholars. In the context of historical archives, however, the material itself is in most cases not digitised, meaning that computational attempts at linking must rely on finding aids which constitute much fewer rich data sources. The EHRI project’s work in this area is therefore quite pioneering and has implications for archival integration on a larger scale, where the disruptive potential of Linked Open Data is most obvious.

Developing the collection graph

Purpose

– In 2010 the European Holocaust Research Infrastructure (EHRI) was funded to support research into the Holocaust. The project follows on from significant efforts in the past to develop and record the collections of the Holocaust in several national initiatives. The purpose of this paper is to introduce the efforts by EHRI to create a flexible research environment using graph databases. The authors concentrate on the added features and design decisions to enable efficient processing of collection information as a graph.

Design/methodology/approach

– The paper concentrates on the specific customisations EHRI had to develop, as the graph database approach is new, and the authors could not rely on existing solutions. The authors describe the serialisations of collections in the graph to provide for efficient processing. Because the EHRI infrastructure is highly distributed, the authors also had to invest a lot of effort into reliable distributed access control mechanisms. Finally, the authors analyse the user-facing work on a portal and a virtual research environment (VRE) in order to discover, share and analyse Holocaust material.

Findings

– Using the novel graph database approach, the authors first present how we can model collection information as graphs and why this is effective. Second, we show how we make collection information persistent and describe the complex access management system we have developed. Third, we outline how we integrate user interaction with the data through a VRE.

Originality/value

– Scholars require specialised access to information. The authors present the results of the work to develop integrated research with collections on the Holocaust researchers and the proposals for a socio-technical ecosystem based on graph database technologies. The use of graph databases is new and the authors needed to work on several innovative customisations to make them work in the domain.

Ocropodium: open source OCR for small-scale historical archives

Large-scale digitization projects dealing with text-based historical material face challenges that are not well catered for by commercial software. This article discusses the results of a project to build a scalable OCR workflow for historical collections based on open source tools that is particularly tailored towards use in small-scale historical archives. It argues that open source tools allow for better customization to match these requirements, particularly with regard to character model training and per-project language modelling. We offer insights into our accuracy evaluation results of various open source OCR tools, as well as a case study about the challenges and opportunities of open source OCR in historical archives.

Methodological commons: arts and humanities e-Science fundamentals

The application of e-Science technologies to disciplines in the arts and humanities raises major questions as to how those technologies can be most usefully exploited, what tools and infrastructures are needed for that exploitation, and what new research approaches can be generated. This paper reviews a number of activities in the UK and Europe in the last 5 years which have sought to address these questions through processes of experimentation and targeted infrastructure development. In the UK, the AHeSSC (Arts and Humanities e-Science Support Centre) has played a coordinating role for seven projects funded by the Arts and Humanities e-Science Initiative. In Europe, DARIAH (Digital Research Infrastructure for the Arts and Humanities) has sought to develop a deeper understanding of research information and communication in the arts and humanities, and to inform the development of e-infrastructures accordingly. Both sets of activity have indicated a common requirement: to construct a framework which consistently describes the methods and functions of scholarly activity which underlie digital arts and humanities research, and the relationships between them. Such a 'methodological commons' has been formulated in the field of the digital humanities. This paper describes the application of this approach to arts and humanities e-Science, with reference to the early work of DARIAH and AHeSSC.

Deploying general-purpose virtual research environments for humanities research

Several virtual research environments (VREs) have been developed to address specific tasks or application domains. Building on the experiences and use cases coming out of these projects, this paper addresses the creation of more general-purpose VREs for the humanities, which move beyond specific, focused tasks, and instead provide services and environments that support more general-purpose humanities research activities. Specifically, we are investigating use cases related to the organization and integration of the dispersed and heterogeneous information on which such research is based. These use cases are highly interactive, interpretative and researcher centric, addressing topics such as annotation environments and support for 'active-reading' processes and scholarly dialogues. We present the background to our work and the technical approach taken, and report the results obtained so far.

Difference Frequency Generation in AgGaS(2): Sellmeier and Temperature-Dispersion Equations

Two single-mode diode lasers and AgGaS(2) as a nonlinear medium are applied for difference frequency generation (DFG) in the mid-infrared spectral range between 4.9 and 6.5 mum. Phase matching is achieved by either temperature tuning or angle tuning of the crystal. Experimentally measured sets of input wavelengths lambda(s) and lambda(p), the resulting DFG wavelength lambda(i), and corresponding phase-matching temperatures or angles are compared with the calculated values derived by use of different Sellmeier equations and coefficients and temperature-dispersion equations dn/dT. Our results show that only specific combinations of previously published Sellmeier equations, coefficients, and temperature-dispersion equations are suitable for exact calculations of phase-matching parameters. These combinations reproduce our experimentally obtained phase-matching temperatures and angles with an accuracy of better than 5% and are therefore of fundamental interest for the design of a mid-infrared DFG spectrometer with AgGaS(2) as a nonlinear medium.

Gonadal sex hormones and dystonia: experimental studies in genetically dystonic hamsters

In some kinds of idiopathic dystonia, including paroxysmal dystonia, a role of sex hormones has been suggested because of female predominance and onset, recurrence, or exacerbation of dystonic symptoms with pregnancy. Similar effects of pregnancy have recently been reported in a model of paroxysmal dystonia, the genetically dystonic hamster. Dystonia in mutant hamsters of both genders is transient, i.e., exhibits spontaneous remission at around puberty, strongly suggesting involvement of gonadal sex hormones. For exploration of the role of sex hormones in dystonia, we undertook a series of ontogenetic experiments in male and female dystonic hamsters. Mutant animals of both genders exhibited the same postnatal development of dystonia with maximum severity of dystonic attacks between weaning and approximately 40 days of age and spontaneous remission thereafter. As shown by plasma sex hormone determinations and, in females, vaginal cytology, spontaneous improvement of the movement disorder coincided with puberty in both genders. Male and female hamsters had about the same plasma levels of progesterone. Compared with nondystonic hamsters, onset of puberty was significantly retarded in both male and female dystonic hamsters. Furthermore, body weight gain was lower in dystonic animals, indicating retarded postnatal development. Gonadectomy at time of weaning did not alter the age-dependent development and remission of dystonia, suggesting that gonadal sex hormones are not critically involved in the disease in hamsters. We propose that transient paroxysmal dystonia in mutant hamsters is caused by postnatal retardation of brain development resulting in a temporary impairment of brain functions with spontaneous remission independent of gonadal sex hormones. In view of the fact that the brain can synthesize steroids such as progesterone independent of peripheral glands in both genders, such neurosteroids might be involved in the postnatal brain maturation that leads to remission of dystonia at around puberty in mutant hamsters.