Research Papers

While we have witnessed a rapid growth of ethics documents meant to guide AI development, the promotion of AI ethics has nonetheless proceeded with little input from AI practitioners themselves. Given the proliferation of AI for Social Good initiatives, this is an emerging gap that needs to be addressed in order to develop more meaningful ethical approaches to AI use and development. This paper offers a methodology, a shared fairness approach, aimed at identifying the needs of AI practitioners when it comes to confronting and resolving ethical challenges and to find a third space where their operational language can be married with that of the more abstract principles that presently remain at the periphery of their work experiences. We offer a grassroots approach to operational ethics based on dialog and mutualised responsibility. This methodology is centred around conversations intended to elicit practitioners perceived ethical attribution and distribution over key value laden operational decisions, to identify when these decisions arise and what ethical challenges they confront, and to engage in a language of ethics and responsibility which enables practitioners to internalise ethical responsibility. The methodology bridges responsibility imbalances that rest in structural decision making power and elite technical knowledge, by commencing with personal, facilitated conversations, returning the ethical discourse to those meant to give it meaning at the sharp end of the ecosystem. Our primary contribution is to add to the recent literature seeking to bring AI practitioners' experiences to the fore by offering a methodology for understanding how ethics manifests as a relational and interdependent sociotechnical practice in their work.

Current discussions of "Explainable AI" (XAI) do not much consider the role of abduction in explanatory reasoning (see Mueller, et al., 2018). It might be worthwhile to pursue this, to develop intelligent systems that allow for the observation and analysis of abductive reasoning and the assessment of abductive reasoning as a learnable skill. Abductive inference has been defined in many ways. For example, it has been defined as the achievement of insight. Most often abduction is taken as a single, punctuated act of syllogistic reasoning, like making a deductive or inductive inference from given premises. In contrast, the originator of the concept of abduction---the American scientist/philosopher Charles Sanders Peirce---regarded abduction as an exploratory activity. In this regard, Peirce's insights about reasoning align with conclusions from modern psychological research. Since abduction is often defined as "inferring the best explanation," the challenge of implementing abductive reasoning and the challenge of automating the explanation process are closely linked. We explore these linkages in this report. This analysis provides a theoretical framework for understanding what the XAI researchers are already doing, it explains why some XAI projects are succeeding (or might succeed), and it leads to design advice.

In this paper, we proposed a transfer learning-based English language learning chatbot, whose output generated by GPT-2 can be explained by corresponding ontology graph rooted by fine-tuning dataset. We design three levels for systematically English learning, including phonetics level for speech recognition and pronunciation correction, semantic level for specific domain conversation, and the simulation of free-style conversation in English - the highest level of language chatbot communication as free-style conversation agent. For academic contribution, we implement the ontology graph to explain the performance of free-style conversation, following the concept of XAI (Explainable Artificial Intelligence) to visualize the connections of neural network in bionics, and explain the output sentence from language model. From implementation perspective, our Language Learning agent integrated the mini-program in WeChat as front-end, and fine-tuned GPT-2 model of transfer learning as back-end to interpret the responses by ontology graph.

With the advent of Industry 4.0, Data Science and Explainable Artificial Intelligence (XAI) has received considerable intrest in recent literature. However, the entry threshold into XAI, in terms of computer coding and the requisite mathematical apparatus, is really high. For fault diagnosis of steel plates, this work reports on a methodology of incorporating XAI based insights into the Data Science process of development of high precision classifier. Using Synthetic Minority Oversampling Technique (SMOTE) and notion of medoids, insights from XAI tools viz. Ceteris Peribus profiles, Partial Dependence and Breakdown profiles have been harvested. Additionally, insights in the form of IF-THEN rules have also been extracted from an optimized Random Forest and Association Rule Mining. Incorporating all the insights into a single ensemble classifier, a 10 fold cross validated performance of 94% has been achieved. In sum total, this work makes three main contributions viz.: methodology based upon utilization of medoids and SMOTE, of gleaning insights and incorporating into model development process. Secondly the insights themselves are contribution, as they benefit the human experts of steel manufacturing industry, and thirdly a high precision fault diagnosis classifier has been developed.

As technology continues to advance, there is increasing concern about individuals being left behind. Many businesses are striving to adopt responsible design practices and avoid any unintended consequences of their products and services, ranging from privacy vulnerabilities to algorithmic bias. We propose a novel approach to fairness and inclusiveness based on experimentation. We use experimentation because we want to assess not only the intrinsic properties of products and algorithms but also their impact on people. We do this by introducing an inequality approach to A/B testing, leveraging the Atkinson index from the economics literature. We show how to perform causal inference over this inequality measure. We also introduce the concept of site-wide inequality impact, which captures the inclusiveness impact of targeting specific subpopulations for experiments, and show how to conduct statistical inference on this impact. We provide real examples from LinkedIn, as well as an open-source, highly scalable implementation of the computation of the Atkinson index and its variance in Spark/Scala. We also provide over a year's worth of learnings -- gathered by deploying our method at scale and analyzing thousands of experiments -- on which areas and which kinds of product innovations seem to inherently foster fairness through inclusiveness.

Iterative and Adaptive Sampling with Spatial Attention for Black-Box Model Explanations

Bhavan Vasu, Chengjiang Long

Deep neural networks have achieved great success in many real-world applications, yet it remains unclear and difficult to explain their decision-making process to an end-user. In this paper, we address the explainable AI problem for deep neural networks with our proposed framework, named IASSA, which generates an importance map indicating how salient each pixel is for the model's prediction with an iterative and adaptive sampling module. We employ an affinity matrix calculated on multi-level deep learning features to explore long-range pixel-to-pixel correlation, which can shift the saliency values guided by our long-range and parameter-free spatial attention. Extensive experiments on the MS-COCO dataset show that our proposed approach matches or exceeds the performance of state-of-the-art black-box explanation methods.

Towards a Unified Evaluation of Explanation Methods without Ground Truth

Hao Zhang, Jiayi Chen, Haotian Xue +1

This paper proposes a set of criteria to evaluate the objectiveness of explanation methods of neural networks, which is crucial for the development of explainable AI, but it also presents significant challenges. The core challenge is that people usually cannot obtain ground-truth explanations of the neural network. To this end, we design four metrics to evaluate explanation results without ground-truth explanations. Our metrics can be broadly applied to nine benchmark methods of interpreting neural networks, which provides new insights of explanation methods.

The field of high-energy physics (HEP), along with many scientific disciplines, is currently experiencing a dramatic influx of new methodologies powered by modern machine learning techniques. Over the last few years, a growing body of HEP literature has focused on identifying promising applications of deep learning in particular, and more recently these techniques are starting to be realized in an increasing number of experimental measurements. The overall conclusion from this impressive and extensive set of studies is that rarer and more complex physics signatures can be identified with the new set of powerful tools from deep learning. However, there is an unstudied systematic risk associated with combining the traditional HEP workflow and deep learning with high-dimensional data. In particular, calibrating and validating the response of deep neural networks is in general not experimentally feasible, and therefore current methods may be biased in ways that are not covered by current uncertainty estimates. By borrowing ideas from AI safety, we illustrate these potential issues and propose a method to bound the size of unaccounted for uncertainty. In addition to providing a pragmatic diagnostic, this work will hopefully begin a dialogue within the community about the robust application of deep learning to experimental analyses.

A new challenge for knowledge graph reasoning started in 2018. Deep learning has promoted the application of artificial intelligence (AI) techniques to a wide variety of social problems. Accordingly, being able to explain the reason for an AI decision is becoming important to ensure the secure and safe use of AI techniques. Thus, we, the Special Interest Group on Semantic Web and Ontology of the Japanese Society for AI, organized a challenge calling for techniques that reason and/or estimate which characters are criminals while providing a reasonable explanation based on an open knowledge graph of a well-known Sherlock Holmes mystery story. This paper presents a summary report of the first challenge held in 2018, including the knowledge graph construction, the techniques proposed for reasoning and/or estimation, the evaluation metrics, and the results. The first prize went to an approach that formalized the problem as a constraint satisfaction problem and solved it using a lightweight formal method; the second prize went to an approach that used SPARQL and rules; the best resource prize went to a submission that constructed word embedding of characters from all sentences of Sherlock Holmes novels; and the best idea prize went to a discussion multi-agents model. We conclude this paper with the plans and issues for the next challenge in 2019.

As reinforcement learning (RL) scales to solve increasingly complex tasks, interest continues to grow in the fields of AI safety and machine ethics. As a contribution to these fields, this paper introduces an extension to Deep Q-Networks (DQNs), called Empathic DQN, that is loosely inspired both by empathy and the golden rule ("Do unto others as you would have them do unto you"). Empathic DQN aims to help mitigate negative side effects to other agents resulting from myopic goal-directed behavior. We assume a setting where a learning agent coexists with other independent agents (who receive unknown rewards), where some types of reward (e.g. negative rewards from physical harm) may generalize across agents. Empathic DQN combines the typical (self-centered) value with the estimated value of other agents, by imagining (by its own standards) the value of it being in the other's situation (by considering constructed states where both agents are swapped). Proof-of-concept results in two gridworld environments highlight the approach's potential to decrease collateral harms. While extending Empathic DQN to complex environments is non-trivial, we believe that this first step highlights the potential of bridge-work between machine ethics and RL to contribute useful priors for norm-abiding RL agents.

Learning via social awareness: Improving a deep generative sketching model with facial feedback

Natasha Jaques, Jennifer McCleary, Jesse Engel +4

In the quest towards general artificial intelligence (AI), researchers have explored developing loss functions that act as intrinsic motivators in the absence of external rewards. This paper argues that such research has overlooked an important and useful intrinsic motivator: social interaction. We posit that making an AI agent aware of implicit social feedback from humans can allow for faster learning of more generalizable and useful representations, and could potentially impact AI safety. We collect social feedback in the form of facial expression reactions to samples from Sketch RNN, an LSTM-based variational autoencoder (VAE) designed to produce sketch drawings. We use a Latent Constraints GAN (LC-GAN) to learn from the facial feedback of a small group of viewers, by optimizing the model to produce sketches that it predicts will lead to more positive facial expressions. We show in multiple independent evaluations that the model trained with facial feedback produced sketches that are more highly rated, and induce significantly more positive facial expressions. Thus, we establish that implicit social feedback can improve the output of a deep learning model.

Characterizing human values is a topic deeply interwoven with the sciences, humanities, art, and many other human endeavors. In recent years, a number of thinkers have argued that accelerating trends in computer science, cognitive science, and related disciplines foreshadow the creation of intelligent machines which meet and ultimately surpass the cognitive abilities of human beings, thereby entangling an understanding of human values with future technological development. Contemporary research accomplishments suggest sophisticated AI systems becoming widespread and responsible for managing many aspects of the modern world, from preemptively planning users' travel schedules and logistics, to fully autonomous vehicles, to domestic robots assisting in daily living. The extrapolation of these trends has been most forcefully described in the context of a hypothetical "intelligence explosion," in which the capabilities of an intelligent software agent would rapidly increase due to the presence of feedback loops unavailable to biological organisms. The possibility of superintelligent agents, or simply the widespread deployment of sophisticated, autonomous AI systems, highlights an important theoretical problem: the need to separate the cognitive and rational capacities of an agent from the fundamental goal structure, or value system, which constrains and guides the agent's actions. The "value alignment problem" is to specify a goal structure for autonomous agents compatible with human values. In this brief article, we suggest that recent ideas from affective neuroscience and related disciplines aimed at characterizing neurological and behavioral universals in the mammalian class provide important conceptual foundations relevant to describing human values. We argue that the notion of "mammalian value systems" points to a potential avenue for fundamental research in AI safety and AI ethics.