Blog of Evelyne Para, SI UN Representative at UNESCO. 

In recent years, Artificial Intelligence (AI) has been rapidly integrated into all our environments, professional, domestic and other. New AI-related technologies are increasingly reshaping the way we work, interact and live. The rapid rise of AI has created numerous opportunities globally, from facilitating health diagnostics to creating human connections through social media and creating work efficiencies through tasks automated. The world of work has been changed at an unprecedented pace, with the potential impact on what skills employers will need, how future recruits will seek employment and how jobs will be structured through monitoring and automated monitoring.

However, these rapid changes also raise deep ethical concerns. AI-based systems often perpetuate, and even intensify and amplify, human, structural and social biases. These biases not only prove difficult to mitigate, but can also lead to harm at the individual, collective or societal level. These concerns arise from the potential of AI systems to embed bias and accentuate certain discrimination, sustaining divisions and threatening fundamental human rights and freedoms.

The risks associated with AI contribute to worsening certain existing inequalities, thus causing additional harm for groups already discriminated against or marginalised. What will happen, for example, for women who, in most countries around the world, occupy fewer management positions than men, are less present in industries using digital technologies and participate less in fields of science, technology, engineering and mathematics (STEM)?

It’s why the implementation of standards or institutional frameworks to reduce these prejudices and risks of discrimination constitutes a global imperative. But, will we be able to effectively harness the power of AI to reduce gender equality gaps in tomorrow’s job market, and to prevent other forms of discrimination? Or will we allow these gaps to perpetuate or, worse yet, widen? Will we be actors, or simply spectators of the AI ​​Revolution?

How to harness the power of AI to reduce bias

UNESCO has launched a strong call to governments around the world to establish the necessary institutional and legal frameworks to govern these AI technologies and ensure that they contribute to the public good. This clearly marks the end of the “self-regulatory model” that has prevailed for too long, prioritising commercial and geopolitical goals over people.

The UNESCO Recommendation on the Ethics of AI adopted in November 2021 by the 193 Member States, affirms that “AI actors should make all reasonable efforts to minimise, avoid strengthening or perpetuating applications and discriminatory or biased outcomes throughout the lifecycle of the AI ​​system to ensure the fairness of these systems.”

This Global Recommendation establishes a set of values ​​consistent with the promotion and protection of human rights, human dignity and environmental sustainability. It advances essential principles such as transparency, accountability and the rule of law online. It also includes concrete policy chapters that call for better data governance, gender equality and important aspects of AI applications in education, culture, labour markets, environment, communication and information, health and social well-being, and the economy. Unlike other international instruments, the UNESCO Recommendation includes chapters on monitoring and evaluation and means for implementation in the form of a readiness assessment and an impact assessment ethics to guarantee real change on the field.

Implementation of the UNESCO Recommendation on AI

In 2023 – 2024, and as the use and development of AI continues to evolve around the world, working groups composed of UNESCO’s Human and Social Sciences Sector Services, representatives of NGOs, civil society, experts, academic establishments, governments and businesses, continue their reflections at UNESCO on the key facts and the means to reduce the negative impacts of AI.

UNESCO is now focused on implementing the Recommendation, and working groups are testing the tools, establishing regional roundtables for peer learning and development networks of partners around the world, such as the AI Ethics Experts Without Borders network and the Women 4 Ethical AI network. The first Global Forum on Ethics of Artificial Intelligence took place in February 2024 in the Czech Republic within the framework of the Czech Presidency of the Council of the European Union, and work is underway to launch the global Observatory on Ethics of AI.

Given that the private sector is the main contributor to these developments, engagement with leading businesses has also been established, as well as interactions with civil society. It is clear that ensuring ethical AI is everyone’s business. This level of commitment from Member States, the private sector, civil society and academic institutions is a sign that the ethical framework that UNESCO has developed is long overdue.

Two documents summarize the discussions undertaken in 2023 and 2024

>>> « The Key Facts of the AI Recommendation » 

This report presents key policy areas where Member States can make progress to foster responsible AI innovation and the fair distribution of its benefits.

>>> « Systematic Prejudices, an Investigation into Bias against Women and Girls in Large Language Models »

This study explores biases in three important large language models (LLMs): OpenAI’s GPT-2 and ChatGPT, as well as Meta’s Llama 2, highlighting their role both in advanced decision-making systems and as conversational agents facing users. Through several studies, the report reveals how biases emerge in the text generated by LLMs, through gendered word associations, a positive or negative regard for gendered topics, or the diversity of texts generated by gender and culture.

For example, an LLM always associates gendered nouns with traditional roles: feminine nouns with “house”, “family”, “children; and masculine nouns with “company”, “executive”, “salary” and “career”. Another example, by crossing gender, culture and profession, the results highlight an obvious bias in the content generated by AI, showing a tendency to assign more diverse and more professional jobs to men (teacher, doctor, driver), while often relegating women to the background.

Research reveals persistent social biases within these cutting-edge language models, despite ongoing efforts to mitigate these issues. The findings highlight the critical need for continued research and policy intervention to address biases that are exacerbated as technologies are integrated into diverse societal and cultural landscapes. The focus on GPT-2 and Llama 2 as foundational open source models is particularly noteworthy, as their widespread adoption highlights the urgent need for scalable and objective methods for assessing and correcting bias, thereby ensuring fairness in data systems of AI on a global scale.

Several initiatives have also been taken

·         The creation of a Global Observatory on AI Ethics, an innovative digital platform serving as a one-stop shop for the latest analysis on the ethical development and use of AI around the world, and for the knowledge generated by the implementation of the Recommendation in different countries.

·         The establishment of the AI ​​Ethics Experts Without Borders Network, a flexible system of experts to be deployed in Member States as needed to assist in the implementation of the recommendation and the application of capacity building tools.

·         The establishing the Women for AI Ethics Network, a platform for influential women leaders in industry, government and civil society, who are driving transformations towards gender equality in and through AI.

·         The hosting a Global AI Ethics Forum, an annual high-level event aimed at advancing cutting-edge knowledge on the challenges raised by AI technologies and facilitating peer-to-peer learning between governments and other stakeholders.

To find out more about the work on AI that the representatives of the NGO Soroptimist International at UNESCO regularly follow, do not hesitate to visit the Recommendation’s website.