Abstract: We are increasingly developing sophisticated systems which are aware of the context that they are used in, and intelligently adapt their behavior to this context. This keynote delves into the essence of ‘context’, acknowledging its diverse conceptualizations in literature, ranging from ‘any information’ to a few categories. Despite acknowledging that the relevance of context is domain-specific, it often remains unclear what is relevant specifically. Technically, context representations aim to objectively capture measurable context elements. However, practical significance often lies at a different abstraction level, where a context element’s relevance and meaning may shift based on how other context elements manifest. For instance, spacial coordinates have no immediate connection with the real word; their relevance and meaning are defined by what is there else. An individual’s (past) experience further complicates relevance: Is it an arbitrary house or is the one where you grew up? Do you currently live there? It context could become more intricate if a stranger enters that house. In this keynote address, I will embrace the subtleties of context, emphasizing that the compound of context elements matters, and underscoring that objective context representations may often only serve as proxies for truly significant experienced context.

Abstract: There has been a huge amount of research on personalisation. It has also moved well beyond the research lab and is widely used in real world applications, websites, and much more. A model of the user and their context drives that personalisation. This, like many other forms of AI, have typically been inscrutable. This is because system builders have failed to prioritise the design of their systems so that the user can scrutinise the key elements of the personalisation process: the data used, the time and context of that data, the user modelling process and the personalisation. As the broad public increasingly calls for better understanding and control of smart systems, we need to establish ways for system builders to focus on scrutability from the design foundations, right through to system building and deployment. This talk will outline a vision for ways to do this, drawing on my team's previous work across diverse areas, ranging from learning, recommenders and smart buildings.