Reasoning from Experience in the Presence of Adversaries

The problem of reasoning from experience in the presence of adversaries found a potent symbol in Descartes’ famous “demon deceiver”: the philosopher wondered how a reasoner might acquire reliable knowledge about the world if his sensory input could be corrupted by a hostile third party. This challenge to empirical philosophy is also a grand challenge for systems that operate within and learn from open on-line environments, like social networks. Such systems are vulnerable to so-called denial of insight attacks that aim to bias their output by distorting or fabricating their input. Recent research in collaborative recommender systems points toward several classes of solution that may be applicable to this grand challenge. This paper explores these options and considers how they may be applicable to other forms of reasoning from experience.