The Journal of Digital Assets
- Volume 1
- August 2023
- Issue 2
AbstractAmidst the trends and advances in Artificial Intelligence (AI) techniques, the dataset’s key role in creating AI harms is a constant. Yet the literature’s attention is typically focussed on the machine-learning’s role, from which the dataset’s role is examined usually in terms of its potential for bias. Datasets also contribute to AI hallucinations. Again, the literature focuses on the composite roles of datasets and AI models in hallucinations, in isolation from a dataset’s contributory role in creating bias. Therefore, the overlapping and distinctive roles of datasets, as a dual contributor, are understudied in the literature. Through the legal concept of a duty of care, in the context of AI applications in financial services, this paper surveys the overlapping and distinctive roles of datasets. This paper’s focus matters for three reasons. First, it is important to identify and determine which parties, now part of a growing ecosystem of stakeholders, are legally liable for selecting and creating the dataset. Second, the tort of negligence is medium-agnostic and can adapt to governing AI harms, which will influence ongoing legislative efforts that are medium-based. And third, this paper provides a frame of reference to start a broader conversation, beyond Singapore, about how datasets will become more prominent in an AI zeitgeist, and that the common law can play a stabilising effect on the policy makers’ impetus to regulate through medium-based legislations.
Key words: AI hallucinations, Dataset bias, Dataset, Negligence, Duty of Care.