their un-debuggability, and their inability to "explain" their decisions in a human understandable and reconstructable way. So while AlphaGo or DeepStack can crush the best humans at Go or Poker, neither program has any internal model of its task; its representations defy interpretation by humans, there is no mechanism to explain their actions and behaviour, and furthermore, there is no obvious instructional value. .. the high performance systems can not help humans improve. Even when we understand the underlying mathematical scaffolding of current machine learning architectures, it is often impossible to get insight into the internal working of the models; we need explicit modeling and reasoning tools to explain how and why a result was achieved. We also know that a significant challenge for future AI is contextual adaptation, i.e., systems that incrementally help to construct explanatory models for solving real-world problems. Here it would be beneficial not to exclude human expertise, but to augment human intelligence with artificial intelligence.

No abstract

It is well recognized that JavaScript can be exploited to launch browser-based security attacks. We propose to battle such attacks using program instrumentation. Untrusted JavaScript code goes through a rewriting process which identifies relevant operations, modifies questionable behaviors, and prompts the user (a web page viewer) for decisions on how to proceed when appropriate. Our solution is parametric with respect to the security policy-the policy is implemented separately from the rewriting, and the same rewriting process is carried out regardless of which policy is in use. Be-sides providing a rigorous account of the correctness of our solution, we also discuss practical issues including policy management and prototype experiments. A useful by-product of our work is an operational semantics of a core subset of JavaScript, where code embedded in (HTML) documents may generate further document pieces (with new code embedded) at runtime, yielding a form of self-modifying code.

We use a state-transition approach to analyze and compare the core access control mechanisms that are characteristic of a variety of trust management, access control list, and capability-based systems. The framework, which characterizes the set of rights a subject has over an object after any sequence of actions, is based on abstract system states, state transitions, and logical deduction of access control judgments. We present abstract models representcontrol lists, and two versions of correspondence and models. The main resul access control lists (ACLs) and capabilities viewedm rows of the Lanipson access matrix and the (proper) subswnption of a form of ACLs by an "unforgeable reference" form of capabilities. The access control mechanism at the heart of distributed trust management systems is formally shown to provide a tractable compromise between unrestricted capability passing from the capability models and easy revocation provided by access control lists. The underlying simulations show how trust management compares with more established access control mechanisms, independent of features such as local name spaces and certijicate authorization hierarchies.

No abstract

No abstract

No abstract

With growing adoption of AI across fields such as healthcare, finance, and the justice system, explaining an AI decision has become more important than ever before. Development of human-centric explainable AI (XAI) systems necessitates an understanding of the requirements of the human-in-the-loop seeking the explanation. This includes the cognitive behavioral purpose that the explanation serves for its recipients, and the structure that the explanation uses to reach those ends. An understanding of the psychological foundations of explanations is thus vital for the development of effective human-centric XAI systems. Towards this end, we survey papers from the cognitive science literature that address the following broad questions: (1) what is an explanation, (2) what are explanations for, and 3) what are the characteristics of good and bad explanations. We organize the insights gained therein by means of highlighting the advantages and shortcomings of various explanation structures and theories, discuss their applicability across different domains, and analyze their utility to various types of humans-in-the-loop. We summarize the key takeaways for human-centric design of XAI systems, and recommend strategies to bridge the existing gap between XAI research and practical needs. We hope this work will spark the development of novel human-centric XAI systems.