Dissertation defence
Dissertation defence – Lars Holmberg

Artificial intelligence and in particular Machine Learning (ML) increasingly impact human life by creating value from collected data. This assetization affects all aspects of human life, from selecting a significant other to recommending the next product to consume. This type of ML-based system thrives since they predict human behaviour based on average case performance metrics (like accuracy), but their usefulness is more limited when it comes to being transparent about their internal knowledge representations for singular decisions, e.g. explaining why they suggest a particular decision in a specific context.

The goal of this work is to let end users be in command of the usage of ML systems and thereby combine the strengths of humans and machines, machines that can propose transparent decisions. Artificial neural networks are an interesting candidate for a setting of this type since this technology has been successful in building knowledge representation from raw data. A neural network can be trained by exposing it to data from the target domain, it can then internalise knowledge representations from the domain and perform contextual tasks. In these situations, the fragment of the actual world internalised in an ML system has to be contextualised by a human to be useful and trustworthy in non-static settings.

This setting is explored through an overarching research question: Which challenges and opportunities may emerge when an end user uses neural networks in context to support singular decision-making? To address this question research through design is used as the central
methodology since this research approach can match the openness of the research question. Through six design experiments, I explore and expand on challenges and opportunities in settings where singular contextual decisions matter. The initial design experiments focus on opportunities in settings that augment human cognitive abilities. Later experiments explore challenges related to settings where neural networks can enhance human cognitive abilities, this part of the work concerns approaches that intend to explain promoted decisions.

This work contributes in three ways: (1) exploring learning related to neural networks in context to put forward a core terminology for contextual decision-making using ML systems, a terminology that includes the generative notions: true-to-the-domain, concept, out-of-distribution and generalisation, (2) a number of design guidelines, (3) the need to align internal knowledge representations with concepts if neural networks are to produce explainable decisions. I also argue that training neural networks to generalise basic concepts like shapes, and colours, concepts easily understandable by humans, is a path forward. This research direction leads toward neural network-based systems that can produce more complex explanations that build on basic generalisable concepts.