IGF 2022 WS #404 Missing Links in AI Governance

Organizer 1: Benjamin Prud'homme, Mila - Institut québécois d'intelligence artificielle
Organizer 2: Prateek Sibal, UNESCO
Organizer 3: Charline d'Oultremont, UNESCO
Organizer 4: Anna Jahn, Mila
Organizer 5: Amanda Leal , Mila

Speaker 1: Catherine Regis , Technical Community, Western European and Others Group (WEOG)
Speaker 2: Golnoosh Farnadi, Technical Community, Western European and Others Group (WEOG)
Speaker 3: Prateek Sibal, Intergovernmental Organization, Intergovernmental Organization

Moderator

Prateek Sibal, Intergovernmental Organization, Intergovernmental Organization

Online Moderator

Amanda Leal , Technical Community, Western European and Others Group (WEOG)

Rapporteur

Anna Jahn, Technical Community, Western European and Others Group (WEOG)

Format

Round Table - Circle - 60 Min

Policy Question(s)

What are the opportunities and limitations of the current AI governance landscape when it comes to including minorities?
How should the different stages of AI development - including the identification of the technical and social risks related thereto - affect policies, and vice-versa?
What are some ways to bridge the gap between technical and social approaches to help further women's and minority rights in the field of AI?

Connection with previous Messages:

SDGs

16. Peace, Justice and Strong Institutions


Targets: GOAL 5: The proposal focuses on gender diversity and Global inclusion for responsible AI development. There are different levels of marginalization in the global AI landscape. Women suffer even more from biases embedded in technology and from a lack of inclusivity in AI governance. For example, facial recognition fails women of color at a disproportionate rate because the AI system was trained on a pool of white men’s faces, and CV screening tools perpetuate gender discrimination in the marketplace because they were trained under criteria of success that privileges a typical male CV and doesn’t account for the historical gender gap in education and in the market. The gender perspective is a central pillar for responsible AI development - the principles of responsible AI present in the Montreal declaration point to the need for a response to the gender gap in AI. This is a crucial point in the responsible development of AI. This gap – beyond a gender issue – raises a much broader social issue, as the lack of diversity in the field of AI prevents the optimal development of responsible AI.

GOAL 10: SDG 10 on reducing inequality includes a target to "empower and promote the social, economic and political inclusion of all, irrespective of age, sex, disability, race, ethnicity, origin, religion or economic or another status." One way to achieve this goal is to use AI to identify and address the many contextual factors that can influence the development and application of AI. In particular, it will be important to consider how AI can help reduce inequality and empower marginalized groups.

GOAL 11: At the core of the diversity and inclusion requirement in AI development lies the inherent limitation of this technology: AI is context-based. Its sustainable application depends on a successful integration with their environments (mostly cities) and the communities on the ground. For AI to foster sustainable development, it needs to abide by fairness principles. Fairness in AI is a mathematical concept applied to the design of algorithms. However technical it is, it cannot be defined void of the local context. To explore the intricacies of fairness in AI for sustainable development, Golnoosh Farnadi, CIFAR chair on AI, will provide an overview of this field of research and its intersections with the SDG.

GOAL 17: Aligning AI with responsible and ethical development depends on forging partnerships at different levels. One of the objectives of SDG17 focuses on technology - “Enhance North-South, South-South, and triangular regional and international cooperation on and access to science, technology and innovation and enhance knowledge sharing on mutually agreed terms, including through improved coordination among existing mechanisms, in particular at the United Nations level, and through a global technology facilitation mechanism”. This SDG is well-covered by this proposal since the objective of the panel is to discuss how to bridge the gap between existing responsible AI governance frameworks and on-the-ground realities, especially in the Global South, so that we can achieve diversity and inclusion in this field.

Description:

Over the next decade, the development of Artificial Intelligence (AI) will continue to significantly impact society. Although AI is becoming a recurrent topic discussed in international forums, the large-scale application of AI and its governance and regulation are still incipient. Meanwhile, AI’s impacts are increasingly scraping borders, extending over multiple industries, and ultimately affecting everyone’s lives. AI is already being used in a wide variety of areas ranging from retail to education, law enforcement, the judiciary, healthcare, climate change, finance, etc.

However, further work needs to be conducted to better understand and harness the contextual impacts of AI. Inasmuch as regulations and frameworks are being developed, the dialogue needs to be expanded and strengthened around more cohesive avenues towards a large scale responsible AI governance. It is essential that all people - from developers, to users, to impacted populations and decision makers at different levels - be aware of the opportunities and limitations of AI systems. Risks related to AI such as potential biases in datasets, algorithm design and when implementing an AI can be predicted, mitigated and addressed. For instance, a common source of risks arises with the deployment of AI systems that are not fine-tuned for the on-the-ground realities. Those missing links between AI development and the myriad of contextual issues we face when applying AI - from sector-specific to cultural and geographical challenges - are presented in the UNESCO-Mila Publication on Missing Links in AI Governance.

As explored in the publication, governance frameworks for AI coupled with advances in ethical AI are helping to address concerns that arise when AI leaves lab settings to be applied to real-world contexts. Nevertheless, AI governance still lacks a broader perspective and inclusion of Global South stakeholders to reach a level of maturity that will allow for a global uptake of standards and practices aligned with responsible and ethical use of AI.

Expected Outcomes

1. The IGF is a forum for discussion of Internet governance issues, and a workshop on inclusive AI policymaking can help ensure that AI policy issues are duly considered in these discussions and enable transparency, safety, and accountability in the digital age;
2. Engage global participants in a participative dialogue on bridging the gap between technical and social approaches to advance women’s/minorities’ rights in AI;
3. Raise awareness about the MILA-UNESCO publication on Missing links in AI governance.

Hybrid Format: To ensure dynamic participation for those physically present at the IGF, our on-site moderator will allow hand raising, questions and a microphone directly connected to the online session for meaningful participation and fluent communication. Both moderators will ensure to include all participants online and offline during the session. In preparation for the IGF, we will ensure that our speakers participating online have sufficient connectivity to lead a successful hybrid session.

We will hold a synchronous hybrid conference session where both online and onsite participants are present at the same time and can interact with each other in real-time.

We will provide clear instructions to both onsite and online participants before the session. This means sending out an agenda, the draft outcome of the policy papers and principles, and providing clear instructions on how to join the conference and access any materials that will be used during the session.
We will make sure the audio and video quality of the speakers online, and of the participant’s room onsite is good for all participants. This means using a high-quality microphone and camera and testing the audio and video setup in advance. By ensuring a specific camera angle (onsite), we would like to have all participants in camera for the online viewers, to ensure fluent communication between both.

Online Participation



Usage of IGF Official Tool.