Australian Institute for Machine Learning, Adelaide University, South Australia

In my opinion, placing focus on including stakeholders and experts in safe implementation, regulation, and policy of AI moreso than on pure technical design will ensure success for the Global Dialogue. A successful outcome would involve recognition of the complex socio-technical, cultural environments in which AI exists, requiring substantial evaluative techniques of AI from the social sciences and implementation science. For instance, "respecting, protecting and promoting human rights: transparency, accountability and human oversight" requires more than just values but instead, methodologies and socio-cultural frameworks which allow for proper exploration of these important concepts. Further, the outcomes should empower communities and individuals and de-centralise large tech companies, empowering others toward local governance and autonomy in use and evaluation of AI.

Social, economic, ethical, cultural, linguistic and technical implications of AI;Transparency, accountability, and human oversight;Interoperability of governance approaches;Protection and promotion of human rights;

Though safe and trustworthy AI, capacity-building, and open source models are all extremely important and valuable highly technical approaches to AI safety, there is significant need for understanding human behaviour and rights in the age of AI away from pure model building. The understanding of how AI systems integrate into socio-cultural systems, different governance processes, and the human oversight needed for such systems is poor. Not enough engagement with interdisciplinary, human and ethics focussed disciplines is continually placing the responsibility and utility of AI in the hands of developers. Instead, the General Assembly Resolution should emphasise areas which require more diverse stakeholder and expert engagement, such as in psychology, philosophy, ethics, and cultural areas of study.

One issue that I believe should be captured further is the domain of human factors. Though there is some attention increasing in this area of how humans make decisions and operate cognitively with AI, there can be greater engagement with how human factors is forming a significant barrier but also pathway to success with AI. A great example is in medicine, where clinical impact cannot be garnered from high performing models alone. Instead, how these models are used by clinicians without compromising existing human expertise is an underrated area of model evaluation. By integrating human factors and its many facets (engineering, ergonomics, cognitive science, ethics) into the Resolution, we may be better positioned to draw attention to this important area.

Governance, protection and promotion of human rights, and transparency and human oversight are of great importance in Australia. Though there are some safety frameworks, much of these are largely values driven and not easily understood in their practical applications. For instance, though we might understand that AI in medicine might require different levels of oversight for varying levels of risk, neither the TGA nor large health bodies enforce any particular governance approaches that can ensure that risk is managed in an accountable and feasible manner. Unlike the EU AI act, it is hard to ensure that AI use and deployment in Australia can be held to a national accord. The global dialogue on AI governance could greatly improve upon these gaps by ensuring there is an internationally applicable approach to safe AI governance to which Australia could be held accountable to. If this dialogue is empowering for local communities and the protection of rights in the age of AI, we have an opportunity to ensure that AI governance privileges greater humanity over profit.

In advancing international cooperation on AI governance, the AI Dialogue can play a critical role to be accountable to communities rather than for-profit tech companies and lobbyists. Already we can see the impact big tech has on lawmaking; OpenAI in the US, and Anthropic in Australia. These are concerning examples of how AI governance can be profiteered to benefit big tech companies over the needs of the population. The UN has a history of ensuring that human rights, autonomy, and cooperation is placed above the needs of 'bad actors' and large corporations. As AI becomes increasingly autonomised, anthropormorphised, and opaque, the AI dialogue can ensure that countries are accountable to cooperation on AI governance that is human-centered.

The AI dialogue should engage with DAIR - the Distributed AI Research Institute which conducts community-rooted research on AI technologies. Their initiative brings to light the darker side of AI technologies, such as the use of data labellers and their experiences, perspectives on race and human rights, and the de-bunking of myths on AI capabilities. Their grassroots driven approach and engagement of community members who are underrepresented in AI is essential for reaching the lived experience of the underserved population. If the AI Dialogue is to adequately address the extent of governance approaches needed to represent global perspectives, DAIR is a fantastic and important initiative with experience in this area.

Stakeholders can use real-world lived examples of AI in their communities or jurisdiction to show the areas of governance which the AI Dialogue should serve. A variety of diverse stakeholders should be represented across academia, lawmaking, industry, organisations, and everyday consumers of AI. Youth in particular have views and contributions about AI which can be universally applicable, and I would recommend the engagement of youth. AI is impacting their future, and the AI Dialogue should be prepared to represent their beliefs and issues with as much care as adult perspectives.

As aforementioned, the engagement of youth is a valuable perspective in global discussions on AI governance. The impact of AI is most acutely felt in the younger generations who feel their autonomy and input on AI may be dwindling. It is important to show that the UN can prioritise young people and their concerns in the future of AI governance. Further, the role of data labellers and data workers who contribute to the creation of models such as large language models will be extremely valuable. Their work is highly traumatic and exploited, and yet is the backbone on which big tech is reliant on. Bringing light to these underserved voices who are forgotten in the decisions made in AI is an excellent opportunity to show the value of the UN in such matters that concern human rights. Many researchers engage with data workers and could empower them by sharing their perspectives through their work in the Dialogue. Further, the use of case studies and applying conversations in the AI Dialogue to varying use cases of different communities is a good way to expand thinking during the Dialogue.

Small group discussion, panels, and activities where different stakeholders work together during the AI Dialogue can enable pragmatic problem solving. Members with different disciplinary perspectives, cultural backgrounds, and problem solving strategies working together is a meaningful way to engage diverse people in the AI dialogue. Talks and seminars that involve representatives of particular communities is a good way to place emphasise on community-driven needs of AI governance. Most importantly, the more that different members can work together to explore complex issues can make sure certain areas are not siloed, and that the output of the AI dialogue is highly diverse.

The CANAIRI project led by Associate Professor Melissa McCradden is an international consortium investigating the silent phase of AI evaluation in medicine which is also relevant to local governance of AI in health institutions. This initiative promotes effective AI governance but also emphasises the importance of accessible methodologies and evidence in the AI pipeline. The EU AI act is also a historic step for legal guidance and governance of AI technologies, establishing a gold standard which was otherwise lacking. The banning of harmful AI practices, such as surveillance, is an important push toward recognition of human rights in AI governance.