AI: Australia’s Ethical Framework
Governance Institute recently lodged a submission to the Department of Industry, Innovation and Science on its discussion paper — Artificial Intelligence: Australia’s Ethical Framework. In our submission we expressed our support for the ongoing discussion about the need for AI opportunities to be balanced with risks. Our members consider that a set of ethical AI principles would greatly aid public trust as these technologies continue to penetrate communities and would also ensure the right mix of innovation and the avoidance of potential harms.
In our submission we observed that there is a considerable volume of documents already published on principles, guidelines and frameworks for AI ethics, including many identified within the discussion paper, as well as several attempts to consolidate the growing amount of material being developed globally. There is significant variation however, in the principles identified. This variation is seen in the number of principles, the level of detail and the language used. We noted that one area of commonality is that almost all of this material provides for the ability to voluntarily commit to the relevant principles, guidelines or framework, rather than providing recommendations for operationalising, overseeing or enforcing them.
We suggested that before looking at the individual principles proposed in the Discussion Paper, there should be consideration about how the principles are structured. Our members consider that it would aid in understanding the principles, if they were grouped into categories or themes. This would not only allow a smaller number of categories which may make understanding the overall intention of the principles easier, but also provide an opportunity for including a larger number of more detailed principles within those groupings. We suggested that a tiered framework of Core Principle, Supporting Principle and Ethical AI issues could be a useful way of structuring any principles. We also suggested two different approaches for developing the core principles. The first is based on the medical ethical principles of respect for ‘autonomy’, ‘non-maleficence’, ‘beneficence’ and ‘justice’. Many of the existing AI ethics documents are based on these four principles, often with a fifth of ‘explicability’. An alternative approach to categorising the proposed principles into a collection of core principles could be to consider the key components of governance: ‘stewardship’, ‘transparency’, ‘accountability’, ‘integrity’ and ‘risk management’, or STAIR.