2021: UNESCO Recommendation and the Global Ethics Dimension of AI Governance
By 2019, international AI governance had begun to develop a common vocabulary.
The OECD AI Principles had established an intergovernmental framework around trustworthy and human-centred AI.
Governments were developing national AI strategies.
Companies were publishing responsible-AI principles.
Researchers were increasingly studying algorithmic bias, transparency, safety and accountability.
But an important question remained.
What should global AI governance ultimately protect?
Was the objective simply to make AI systems more reliable?
Was it to encourage innovation?
Was it to reduce economic and technological risks?
Or was AI governance fundamentally about protecting people and societies?
In November 2021, UNESCO provided one of the most comprehensive answers to that question.
The UNESCO General Conference adopted the Recommendation on the Ethics of Artificial Intelligence on 23 November 2021.
The Recommendation represented a significant expansion of the global AI governance conversation.
It connected AI not only to innovation and technological responsibility, but also to human rights, human dignity, democracy, environmental sustainability, cultural diversity, education, development and social justice.
This was an important change in the evolution of AI governance.
THE MOVE FROM TRUSTWORTHY AI TO GLOBAL ETHICS
The OECD Principles and the UNESCO Recommendation share many common concerns.
Both recognise the importance of human-centred AI.
Both emphasise fairness, transparency, accountability and safety.
But UNESCO approached the issue from a broader global perspective.
The Recommendation treated AI ethics not simply as a question of how organisations should build trustworthy systems.
It also examined how AI could affect societies, institutions, cultures and human development.
This distinction matters.
A technical system can be safe in a narrow engineering sense and still create broader social problems.
An AI system can be accurate while contributing to discrimination.
A highly efficient system can still undermine human autonomy.
A powerful technology can generate economic value while increasing inequality.
A country can adopt AI rapidly while lacking the institutional capacity to govern it responsibly.
UNESCO’s approach therefore widened the governance lens.
AI was becoming a societal governance issue.
HUMAN DIGNITY AT THE CENTRE
One of the most important features of the UNESCO Recommendation is its strong connection to human dignity and human rights.
This placed AI governance within an established international normative framework.
Instead of treating AI as an entirely new policy domain, UNESCO connected it to principles that already exist in international law and public policy.
This included concerns related to equality, privacy, freedom of expression, non-discrimination, human oversight and human agency.
The importance of this approach is that AI systems do not operate outside society.
They operate within legal, economic, political and cultural institutions.
If AI changes how decisions are made, information is distributed, work is organised or public services are delivered, then the effects can ultimately reach fundamental rights.
AI governance therefore becomes partly a question of protecting the conditions under which people can exercise those rights.
HUMAN OVERSIGHT AND HUMAN AGENCY
Another important dimension of the UNESCO framework is the emphasis on human agency.
AI systems can make recommendations, predictions and increasingly complex decisions.
But the existence of an automated system does not eliminate human responsibility.
A central governance question is therefore how humans remain meaningfully involved.
Human oversight is not simply about putting a person somewhere in a technical workflow.
The person must have sufficient authority, knowledge and ability to intervene when necessary.
This distinction becomes especially important in high-impact environments.
Consider a public service decision.
If an AI system recommends that an individual should receive or be denied a service, human oversight only has meaning if an official can understand the relevant circumstances, question the output and override the system when appropriate.
This is one of the reasons why AI governance increasingly focuses on institutional design rather than only model design.
AI AND DEMOCRACY
UNESCO also placed significant attention on democracy.
This was particularly important because AI systems can influence information environments.
Recommendation systems can affect what people see.
Automated content generation can affect how information is produced.
Synthetic media can complicate the distinction between authentic and manipulated content.
Automated systems can influence political communication and public discourse.
AI therefore has implications for the information environment in which democratic societies operate.
The governance question is not whether AI should be allowed to influence public information at all.
Digital technologies have always influenced information systems.
The more important question is whether these systems operate in ways compatible with democratic values, transparency, human rights and public accountability.
UNESCO’s framework helped place these concerns within the broader AI ethics agenda.
CULTURAL DIVERSITY
Another particularly important contribution of the UNESCO Recommendation is its attention to cultural diversity.
This issue is sometimes overlooked in AI governance discussions.
Many AI systems are developed using datasets and linguistic resources that disproportionately represent certain countries, languages and cultures.
This can create uneven performance.
A model may work extremely well in one linguistic or cultural environment while performing poorly in another.
The problem is not simply technical.
It can affect who benefits from AI.
It can influence whose knowledge is represented.
It can shape how local languages are processed.
It can affect educational resources, public communication and digital participation.
For countries with smaller language communities, these questions can be particularly significant.
AI governance therefore includes questions about representation.
Who is represented in AI systems?
Whose data is available?
Which languages are supported?
Which cultural contexts are understood?
Which communities are visible in AI development?
These questions are increasingly important for the Global South.
THE GLOBAL SOUTH DIMENSION
One of the most strategically important aspects of UNESCO’s approach is that it does not treat AI governance entirely as a problem of advanced technology development.
It also considers development.
This is important because countries have very different levels of access to computing infrastructure, technical expertise, research funding and AI investment.
A governance framework that focuses only on controlling advanced AI systems may overlook another major challenge:
Many countries are still building the capacity to use AI effectively.
For developing countries, AI governance can therefore involve two simultaneous objectives.
The first is managing risks.
The second is building capacity.
Without the second, the first can become difficult to achieve.
A country that lacks AI expertise may struggle to evaluate imported AI systems.
It may struggle to understand the risks associated with automated decision-making.
It may depend heavily on foreign technology providers.
It may have limited ability to participate in international standard-setting.
It may also lack the research infrastructure required to develop locally relevant AI solutions.
This makes capacity building a governance issue in its own right.
AI GOVERNANCE AND INEQUALITY
UNESCO’s framework also draws attention to the possibility that AI could reinforce existing inequalities.
AI can create new economic opportunities.
But the benefits may not be distributed equally.
Countries with advanced infrastructure, capital, computing resources and research ecosystems may capture a disproportionate share of the economic value created by AI.
Meanwhile, countries with limited infrastructure may become primarily consumers of foreign AI systems.
This creates a potential structural imbalance.
The issue is therefore not only whether AI is “ethical.”
It is also whether the global AI economy is inclusive.
Who develops the technology?
Who owns the infrastructure?
Who controls the data?
Who benefits economically?
Who bears the risks?
Who participates in governance?
These questions would become increasingly important as AI developed into a strategic global technology.
THE ENVIRONMENTAL DIMENSION
Another major contribution of the UNESCO Recommendation is its attention to environmental sustainability.
AI systems require infrastructure.
Large-scale computing requires electricity.
Data centres require physical resources and cooling.
The development and operation of advanced AI systems therefore have environmental implications.
This adds another layer to AI governance.
Responsible AI cannot be considered only in terms of model behaviour.
The infrastructure required to build and operate AI also matters.
Environmental considerations therefore connect AI governance with energy policy, infrastructure planning and sustainability.
As AI systems become larger and more computationally demanding, this dimension is likely to become increasingly important.
AI AND EDUCATION
UNESCO’s institutional role also made education a central part of the discussion.
AI is changing not only how people work but also how they learn.
Educational institutions need to prepare students for an environment in which AI systems are increasingly present.
This creates several governance questions.
How should students learn to use AI responsibly?
How should institutions address academic integrity?
How should educators understand AI-generated content?
What skills become more important when AI can automate certain cognitive tasks?
How should education systems prepare students for changing labour markets?
AI governance therefore intersects with education policy.
This is particularly relevant for developing countries.
If education systems do not adapt, the gap between AI-enabled economies and less-prepared economies could widen.
UNESCO’S APPROACH TO GOVERNANCE
The Recommendation also emphasised that ethical AI requires institutional mechanisms.
Ethics cannot depend entirely on individual goodwill.
Governments need appropriate policies.
Organisations need governance structures.
Technical communities need standards and evaluation methods.
Educational institutions need capacity.
Public institutions need expertise.
Citizens need awareness and avenues for participation.
This is an important conceptual development.
AI ethics is not only about individual behaviour.
It is about institutions.
The question becomes:
What kind of governance system makes responsible AI more likely?
This is a much deeper question than simply asking whether developers have followed a list of ethical principles.
FROM PRINCIPLES TO IMPLEMENTATION
Like the OECD framework, the UNESCO Recommendation is not a global AI law.
It does not function as a single international regulator.
Its importance comes from establishing a global normative framework and encouraging member states to translate those principles into national policies and institutional practices.
This distinction is critical.
International recommendations can influence national policy.
They can shape legislation.
They can influence standards.
They can guide public institutions.
They can establish expectations for companies and researchers.
But implementation ultimately depends heavily on national and institutional capacity.
This creates a recurring problem in AI governance.
Global principles may be relatively easy to adopt.
Operational governance is much harder.
THE IMPORTANCE OF POLICY ASSESSMENT
One of the practical implications of the UNESCO approach is that countries need to assess their own readiness.
A government cannot meaningfully implement responsible AI policy without knowing:
What AI systems are already being used.
Which institutions are responsible.
What regulatory frameworks already exist.
What technical expertise is available.
What data governance systems exist.
How prepared universities and research institutions are.
What infrastructure is available.
What risks are emerging.
Where institutional gaps exist.
This turns AI governance into an assessment problem.
Before creating new rules, countries need to understand their existing governance capacity.
This is especially important for countries where AI policy institutions are still developing.
THE BANGLADESH RELEVANCE
For Bangladesh, this perspective is particularly important.
The question should not be limited to whether Bangladesh eventually adopts an AI policy.
A more fundamental question is whether the country possesses the institutional capacity required to implement one.
That includes government expertise, research capability, university readiness, technical standards, data governance, public-sector capacity and mechanisms for evaluating AI systems.
The UNESCO framework provides a useful lens for thinking about these questions because it treats AI governance as broader than regulation.
It connects governance with education, development, human rights, cultural diversity and institutional capacity.
For a developing country, that broader perspective may be more useful than simply copying the regulatory architecture of a technologically advanced economy.
A national AI governance framework needs to reflect national capability and development conditions.
THE LIMITATION OF GLOBAL PRINCIPLES
UNESCO’s Recommendation is ambitious.
But ambitious principles also face implementation challenges.
Countries differ substantially in institutional capacity.
The ability to enforce ethical requirements varies.
Technical expertise is unevenly distributed.
Regulatory institutions have different mandates.
AI systems are developed by companies operating across borders.
Some of the most important AI infrastructure is concentrated in a relatively small number of firms and countries.
This means that global ethical agreement does not automatically produce global governance convergence.
There may be broad agreement on principles while significant differences remain in implementation.
This distinction will become increasingly important.
THE GROWING GAP BETWEEN PRINCIPLES AND POWER
By 2021, AI governance therefore had developed an interesting tension.
International organisations were increasingly establishing principles around human rights, fairness, safety and accountability.
At the same time, AI capabilities and infrastructure were becoming increasingly concentrated among major technology companies and technologically advanced economies.
This created a gap.
Normative power and technological power were not necessarily distributed in the same way.
International organisations could articulate principles.
Governments could develop policy.
But a relatively small number of organisations possessed the computing infrastructure, data resources and technical capacity necessary to build the most advanced AI systems.
This would become one of the defining governance challenges of the 2020s.
AI GOVERNANCE IS ALSO A DEVELOPMENT ISSUE
The UNESCO framework therefore reveals an important principle:
AI governance is not only about preventing harm from advanced technology.
It is also about ensuring that societies have the ability to participate in the benefits of that technology.
This makes AI governance closely connected to development policy.
For the Global South, the governance agenda may therefore include questions about:
Access to infrastructure.
Technical education.
Research capacity.
Local-language AI.
Public-sector readiness.
Data governance.
International representation.
Technology transfer.
Investment.
Institutional capability.
These concerns do not replace AI safety or regulation.
They expand the governance agenda.
FROM 2021 TO THE GENERATIVE AI ERA
The UNESCO Recommendation was adopted before the widespread public explosion of generative AI.
ChatGPT would not launch until November 2022.
Large multimodal and foundation models would become dramatically more capable over the following years.
This timing is significant.
UNESCO established a broad ethical framework just before AI entered a new technological phase.
Generative AI would test almost every major principle that had been developed during the previous decade.
Transparency would become harder.
Accountability would become more complex.
Misinformation risks would increase.
Copyright questions would intensify.
Education systems would face new challenges.
Labour markets would be affected.
Synthetic media would become easier to create.
And questions about advanced AI safety would become more urgent.
The governance conversation would therefore move into a new phase.
THE SHIFT FROM ETHICS TO SYSTEM GOVERNANCE
The OECD Principles and UNESCO Recommendation helped establish a strong normative foundation.
But the next challenge was increasingly systemic.
Governments would have to think about the governance of general-purpose AI models.
They would have to consider model evaluation.
Risk management.
Testing.
Transparency requirements.
Incident reporting.
Auditing.
Content governance.
Compute infrastructure.
International cooperation.
And eventually, frontier AI safety.
This represented another transformation.
AI governance was moving from:
“What principles should responsible AI follow?”
toward:
“What institutions, rules and technical mechanisms can actually govern increasingly powerful AI systems?”
That question would become central after 2022.
CONCLUSION
The UNESCO Recommendation on the Ethics of Artificial Intelligence marked a major expansion of the global AI governance agenda.
The OECD Principles had helped establish an international language around trustworthy AI.
UNESCO broadened that language.
AI governance became increasingly connected to human dignity, human rights, democracy, cultural diversity, environmental sustainability, education and development.
The significance of the 2021 Recommendation therefore lies not only in the principles it articulated.
It lies in the governance philosophy behind them.
AI should not be evaluated only according to whether it works.
It should also be evaluated according to what it does to people, institutions and societies.
And AI governance should not be designed only around the interests and capacities of countries developing the most advanced technologies.
It must also consider the countries and communities that are adopting, adapting to and being affected by those technologies.
This creates a particularly important lesson for the Global South.
AI governance requires both risk management and capability development.
Countries need the ability not only to regulate AI, but also to understand it, evaluate it, research it and participate in shaping its future.
By 2021, the international AI governance conversation had therefore moved considerably beyond the question of ethics.
The foundations for a broader global governance system were being established.
But a new technological shock was about to arrive.
Generative AI would transform the scale, speed and complexity of the governance problem.
And that would force governments, companies and international organisations to reconsider whether existing principles were enough.
THE NEXT PHASE
The next article will examine the emergence of generative AI between 2022 and 2023.
The arrival of large language models and widely accessible generative AI changed the governance debate dramatically.
For the first time, highly capable general-purpose AI systems became accessible to hundreds of millions of people.
The central question shifted again:
How do we govern AI systems that are general-purpose, rapidly evolving, widely deployed and capable of creating new uses faster than policymakers can anticipate them?
That is where the modern AI governance era truly accelerates.
ATLAS AI GOVERNANCE SERIES:
“From Principles to Power: The Evolution of AI Governance”
Part 1 — What Is AI Governance—and Why Has It Become So Important?
Part 2 — Before AI Governance: What Did We Learn from Data Protection, Cybersecurity and Internet Governance?
Part 3 — Why Did AI Governance Become a Global Policy Issue in the 2010s?
Part 4 — 2019: OECD AI Principles and the Start of a New Era in AI Governance
Part 5 — 2021: UNESCO Recommendation and the Global Ethics Dimension of AI Governance
Part 6 — 2022–23: Generative AI and the New Governance Problem