On the Need for Collaborative Intelligence in Cybersecurity

Abstract

The success of artificial intelligence (and particularly data-driven machine learning) in classifying and making predictions from large bodies of data has led to an expectation that autonomous AI systems can be deployed in cybersecurity applications. In this position paper we outline some of the problems facing machine learning in cybersecurity and argue for a collaborative approach where humans contribute insight and understanding, whilst machines are used to gather, filter and process data into a convenient and understandable form. In turn this requires a convenient representation for exchanging information between machine and human, and we argue that graded concepts are suitable, allowing summarisation at multiple levels of discernibility (granularity). We conclude with some suggestions for developing a hierarchical and graded representation.