• [2020-2023] First hire → Project manager → Digital collaboration consultant at ROER. But mainly, got to be mentored by Frank van Summeren on how to be a professional (❤️).
• [2014-2023] Tutor in various forms. E.g., 1-1 homework guidance for high-school students, group tutoring of classes, 1-year student-life mentoring
• [2012-2023] Tennis coach in various forms. From individual high-performance to large group recreational.

Artificial Morality: Incorporating Moral Concepts into Artificial Intelligence

Exploring the potential of Iris Murdoch’s ethical theory to increase the moral
decision making of a large language model

Supervised by Gunter Bombaerts

The societal motivation of the thesis

Recent developments in large language models (LLMs) have catapulted the capabilities of artificial intelligence (AI) applications, yet research into ensuring ethical behaviour of these models has been lacking. As autonomous AI systems are becoming progressively embedded in areas of society where ethical considerations are paramount, such as autonomous disease detection or vehicle control, there is a pressing need to ensure moral behaviour by these systems. There is little to none research on the impact of architectural decisions (based on ethical philosophy) that could lead to more moral behaviour. Since 1) ethical philosophers have thought for millennia about ethical human behaviour, and 2) modern AI systems are heavily inspired by the human brain; it is logical to explore the effectiveness of using these ethical theories for so-called ‘top-down’ (ethical behaviour by design) ethics into the Transformer – the mathematical architecture of the LLM.

Summary of the core idea of the thesis

Murdoch’s ethical insights on human attention can be applied to the design of the transformer’s attention mechanism to encourage ethical behavior according to Murdoch’s standards. This can be achieved by manipulating the attention mechanism to promote a selfless and morally-focused attention pattern.

Murdoch’s view on attention is that it is a central aspect of moral behavior, and that by training one’s attention on external objects and individuals, one can develop a more accurate and selfless understanding of the world, leading to more moral behavior.

The transformer’s attention mechanism is a mathematical function that allows the model to focus on specific parts of an input sequence when generating output. In other words, it is a mathematical representation of the model’s attention.

By modifying the attention mechanism to focus on information that promotes selfless and morally-focused attention, one could induce more ethical behavior in the transformer’s auto-regression. This could be achieved by, for example, weighting the attention mechanism towards information that is relevant to the needs and wishes of others, rather than information that is self-serving or focused on personal gain.

• Data Science (master) study association | Pattern/JADS
  • Future student outreach
  • Intro committee
• Physics (bachelor) study association | vdWaals
  • Future student outreach
  • Gala committee | Chair. Dealt with large group stakeholder management during the emotional period of the first Covid lockdowns.
  • Activity committee | Chair
  • Intro committee | External affairs and co-chair

This was fun! 🙂