This year I have completed my study at the Australian National University. I completed a double degree, and have graduated with two degrees. They are:
Bachelor of Advanced Computing (Hons)
⮡ Specialisation: Artificial Intelligence
⮡ Major: Cyber Security
Bachelor of Science
⮡ Specialisation: Advanced Mathematics
⮡ Major: Mathematics
⮡ Minor: Science Communication
The honours I have completed is based in Aritificial Intelligence theory, particularly constraint based planning problems. The goal of it was to translate ethical norms or frameworks into formal constraints for planners. The thesis is available on the website here. For an overview of the content, here is the abstract:
Abstract
The rise of autonomous agents’ relevance in everyday life work has increasingly focused on the question of ethical restraint. A common approach is to formalise these ethical norms into formal constraints which we impose upon the solving agent. Using these constraints the agent can solve for an optimal solution which adheres to these ethics, or can check after finding a plan whether it satisfies these constraints. Unfortunately ethical frameworks are numerous and varied, and most papers focus on one particular set of ethics (utilitarianism, act-based deontology, virtue ethics) and then produce a formalisation which is very specific to that philosophical model. In this project, we create a philosophically ambivalent operator that allows us to define a structure which we believe is intrinsic to many frameworks, a hierarchy of norms. The key difference between many philosophical frameworks is how they rank actions within the framework, how they judge bad, good, better, and best actions. The invariant property in this is the existence of the hierarchy. Thus instead of focusing on formalising how to rank these actions, we focus on communicating this ranking to the solver. Using our hierarchy operator, a structure can be defined over a sequence of norms, with which we can check for satisfaction from any given plan. We provide an initial (verbose) definition in first order logic, then a more concise definition using set comprehension and sequences of norms. Then using this sequence definition we can prove some interesting properties of hierarchies such as concatenation, removing duplicates, maintenance of prefix structure, and a partial order on the space of norms. We also provide an example showing how it correctly constrains a sample plan.