Research Papers

Towards Artificial Virtuous Agents: Games, Dilemmas and Machine Learning

Ajay Vishwanath, Einar Duenger Bøhn, Ole-Christoffer Granmo +2

Machine ethics has received increasing attention over the past few years because of the need to ensure safe and reliable artificial intelligence (AI). The two dominantly used theories in machine ethics are deontological and utilitarian ethics. Virtue ethics, on the other hand, has often been mentioned as an alternative ethical theory. While this interesting approach has certain advantages over popular ethical theories, little effort has been put into engineering artificial virtuous agents due to challenges in their formalization, codifiability, and the resolution of ethical dilemmas to train virtuous agents. We propose to bridge this gap by using role-playing games riddled with moral dilemmas. There are several such games in existence, such as Papers, Please and Life is Strange, where the main character encounters situations where they must choose the right course of action by giving up something else dear to them. We draw inspiration from such games to show how a systemic role-playing game can be designed to develop virtues within an artificial agent. Using modern day AI techniques, such as affinity-based reinforcement learning and explainable AI, we motivate the implementation of virtuous agents that play such role-playing games, and the examination of their decisions through a virtue ethical lens. The development of such agents and environments is a first step towards practically formalizing and demonstrating the value of virtue ethics in the development of ethical agents.

The success of DNNs is driven by the counter-intuitive ability of over-parameterized networks to generalize, even when they perfectly fit the training data. In practice, test error often continues to decrease with increasing over-parameterization, referred to as double descent. This allows practitioners to instantiate large models without having to worry about over-fitting. Despite its benefits, however, prior work has shown that over-parameterization can exacerbate bias against minority subgroups. Several fairness-constrained DNN training methods have been proposed to address this concern. Here, we critically examine MinDiff, a fairness-constrained training procedure implemented within TensorFlow's Responsible AI Toolkit, that aims to achieve Equality of Opportunity. We show that although MinDiff improves fairness for under-parameterized models, it is likely to be ineffective in the over-parameterized regime. This is because an overfit model with zero training loss is trivially group-wise fair on training data, creating an "illusion of fairness," thus turning off the MinDiff optimization (this will apply to any disparity-based measures which care about errors or accuracy. It won't apply to demographic parity). Within specified fairness constraints, under-parameterized MinDiff models can even have lower error compared to their over-parameterized counterparts (despite baseline over-parameterized models having lower error). We further show that MinDiff optimization is very sensitive to choice of batch size in the under-parameterized regime. Thus, fair model training using MinDiff requires time-consuming hyper-parameter searches. Finally, we suggest using previously proposed regularization techniques, viz. L2, early stopping and flooding in conjunction with MinDiff to train fair over-parameterized models.

With the widespread use of sophisticated machine learning models in sensitive applications, understanding their decision-making has become an essential task. Models trained on tabular data have witnessed significant progress in explanations of their underlying decision making processes by virtue of having a small number of discrete features. However, applying these methods to high-dimensional inputs such as images is not a trivial task. Images are composed of pixels at an atomic level and do not carry any interpretability by themselves. In this work, we seek to use annotated high-level interpretable features of images to provide explanations. We leverage the Shapley value framework from Game Theory, which has garnered wide acceptance in general XAI problems. By developing a pipeline to generate counterfactuals and subsequently using it to estimate Shapley values, we obtain contrastive and interpretable explanations with strong axiomatic guarantees.

There is a clear desire to model and comprehend human behavior. Trends in research covering this topic show a clear assumption that many view human reasoning as the presupposed standard in artificial reasoning. As such, topics such as game theory, theory of mind, machine learning, etc. all integrate concepts which are assumed components of human reasoning. These serve as techniques to attempt to both replicate and understand the behaviors of humans. In addition, next generation autonomous and adaptive systems will largely include AI agents and humans working together as teams. To make this possible, autonomous agents will require the ability to embed practical models of human behavior, which allow them not only to replicate human models as a technique to "learn", but to to understand the actions of users and anticipate their behavior, so as to truly operate in symbiosis with them. The main objective of this paper it to provide a succinct yet systematic review of the most important approaches in two areas dealing with quantitative models of human behaviors. Specifically, we focus on (i) techniques which learn a model or policy of behavior through exploration and feedback, such as Reinforcement Learning, and (ii) directly model mechanisms of human reasoning, such as beliefs and bias, without going necessarily learning via trial-and-error.

It is known that recommendations of AI-based systems can be incorrect or unfair. Hence, it is often proposed that a human be the final decision-maker. Prior work has argued that explanations are an essential pathway to help human decision-makers enhance decision quality and mitigate bias, i.e., facilitate human-AI complementarity. For these benefits to materialize, explanations should enable humans to appropriately rely on AI recommendations and override the algorithmic recommendation when necessary to increase distributive fairness of decisions. The literature, however, does not provide conclusive empirical evidence as to whether explanations enable such complementarity in practice. In this work, we (a) provide a conceptual framework to articulate the relationships between explanations, fairness perceptions, reliance, and distributive fairness, (b) apply it to understand (seemingly) contradictory research findings at the intersection of explanations and fairness, and (c) derive cohesive implications for the formulation of research questions and the design of experiments.

Integrating Prior Knowledge in Post-hoc Explanations

Adulam Jeyasothy, Thibault Laugel, Marie-Jeanne Lesot +2

In the field of eXplainable Artificial Intelligence (XAI), post-hoc interpretability methods aim at explaining to a user the predictions of a trained decision model. Integrating prior knowledge into such interpretability methods aims at improving the explanation understandability and allowing for personalised explanations adapted to each user. In this paper, we propose to define a cost function that explicitly integrates prior knowledge into the interpretability objectives: we present a general framework for the optimization problem of post-hoc interpretability methods, and show that user knowledge can thus be integrated to any method by adding a compatibility term in the cost function. We instantiate the proposed formalization in the case of counterfactual explanations and propose a new interpretability method called Knowledge Integration in Counterfactual Explanation (KICE) to optimize it. The paper performs an experimental study on several benchmark data sets to characterize the counterfactual instances generated by KICE, as compared to reference methods.

EVOTER: Evolution of Transparent Explainable Rule-sets

Hormoz Shahrzad, Babak Hodjat, Risto Miikkulainen

Most AI systems are black boxes generating reasonable outputs for given inputs. Some domains, however, have explainability and trustworthiness requirements that cannot be directly met by these approaches. Various methods have therefore been developed to interpret black-box models after training. This paper advocates an alternative approach where the models are transparent and explainable to begin with. This approach, EVOTER, evolves rule-sets based on simple logical expressions. The approach is evaluated in several prediction/classification and prescription/policy search domains with and without a surrogate. It is shown to discover meaningful rule sets that perform similarly to black-box models. The rules can provide insight into the domain, and make biases hidden in the data explicit. It may also be possible to edit them directly to remove biases and add constraints. EVOTER thus forms a promising foundation for building trustworthy AI systems for real-world applications in the future.

Lung cancer has the highest mortality rate of deadly cancers in the world. Early detection is essential to treatment of lung cancer. However, detection and accurate diagnosis of pulmonary nodules depend heavily on the experiences of radiologists and can be a heavy workload for them. Computer-aided diagnosis (CAD) systems have been developed to assist radiologists in nodule detection and diagnosis, greatly easing the workload while increasing diagnosis accuracy. Recent development of deep learning, greatly improved the performance of CAD systems. However, lack of model reliability and interpretability remains a major obstacle for its large-scale clinical application. In this work, we proposed a multi-task explainable deep-learning model for pulmonary nodule diagnosis. Our neural model can not only predict lesion malignancy but also identify relevant manifestations. Further, the location of each manifestation can also be visualized for visual interpretability. Our proposed neural model achieved a test AUC of 0.992 on LIDC public dataset and a test AUC of 0.923 on our in-house dataset. Moreover, our experimental results proved that by incorporating manifestation identification tasks into the multi-task model, the accuracy of the malignancy classification can also be improved. This multi-task explainable model may provide a scheme for better interaction with the radiologists in a clinical environment.

Multi-modal multi-objective model-based genetic programming to find multiple diverse high-quality models

E. M. C. Sijben, T. Alderliesten, P. A. N. Bosman

Explainable artificial intelligence (XAI) is an important and rapidly expanding research topic. The goal of XAI is to gain trust in a machine learning (ML) model through clear insights into how the model arrives at its predictions. Genetic programming (GP) is often cited as being uniquely well-suited to contribute to XAI because of its capacity to learn (small) symbolic models that have the potential to be interpreted. Nevertheless, like many ML algorithms, GP typically results in a single best model. However, in practice, the best model in terms of training error may well not be the most suitable one as judged by a domain expert for various reasons, including overfitting, multiple different models existing that have similar accuracy, and unwanted errors on particular data points due to typical accuracy measures like mean squared error. Hence, to increase chances that domain experts deem a resulting model plausible, it becomes important to be able to explicitly search for multiple, diverse, high-quality models that trade-off different meanings of accuracy. In this paper, we achieve exactly this with a novel multi-modal multi-tree multi-objective GP approach that extends a modern model-based GP algorithm known as GP-GOMEA that is already effective at searching for small expressions.

Sparse Subspace Clustering for Concept Discovery (SSCCD)

Johanna Vielhaben, Stefan Blücher, Nils Strodthoff

Concepts are key building blocks of higher level human understanding. Explainable AI (XAI) methods have shown tremendous progress in recent years, however, local attribution methods do not allow to identify coherent model behavior across samples and therefore miss this essential component. In this work, we study concept-based explanations and put forward a new definition of concepts as low-dimensional subspaces of hidden feature layers. We novelly apply sparse subspace clustering to discover these concept subspaces. Moving forward, we derive insights from concept subspaces in terms of localized input (concept) maps, show how to quantify concept relevances and lastly, evaluate similarities and transferability between concepts. We empirically demonstrate the soundness of the proposed Sparse Subspace Clustering for Concept Discovery (SSCCD) method for a variety of different image classification tasks. This approach allows for deeper insights into the actual model behavior that would remain hidden from conventional input-level heatmaps.

Legislation and public sentiment throughout the world have promoted fairness metrics, explainability, and interpretability as prescriptions for the responsible development of ethical artificial intelligence systems. Despite the importance of these three pillars in the foundation of the field, they can be challenging to operationalize and attempts to solve the problems in production environments often feel Sisyphean. This difficulty stems from a number of factors: fairness metrics are computationally difficult to incorporate into training and rarely alleviate all of the harms perpetrated by these systems. Interpretability and explainability can be gamed to appear fair, may inadvertently reduce the privacy of personal information contained in training data, and increase user confidence in predictions -- even when the explanations are wrong. In this work, we propose a framework for responsibly developing artificial intelligence systems by incorporating lessons from the field of information security and the secure development lifecycle to overcome challenges associated with protecting users in adversarial settings. In particular, we propose leveraging the concepts of threat modeling, design review, penetration testing, and incident response in the context of developing AI systems as ways to resolve shortcomings in the aforementioned methods.

Deep reinforcement learning has led to numerous notable results in robotics. However, deep neural networks (DNNs) are unintuitive, which makes it difficult to understand their predictions and strongly limits their potential for real-world applications due to economic, safety, and assurance reasons. To remedy this problem, a number of explainable AI methods have been presented, such as SHAP and LIME, but these can be either be too costly to be used in real-time robotic applications or provide only local explanations. In this paper, the main contribution is the use of a linear model tree (LMT) to approximate a DNN policy, originally trained via proximal policy optimization(PPO), for an autonomous surface vehicle with five control inputs performing a docking operation. The two main benefits of the proposed approach are: a) LMTs are transparent which makes it possible to associate directly the outputs (control actions, in our case) with specific values of the input features, b) LMTs are computationally efficient and can provide information in real-time. In our simulations, the opaque DNN policy controls the vehicle and the LMT runs in parallel to provide explanations in the form of feature attributions. Our results indicate that LMTs can be a useful component within digital assurance frameworks for autonomous ships.