Module 13 Content: AI Governance and the Future of AI

AI systems introduce risks that individual users cannot manage alone. Governance addresses these issues through rules, institutional practices, and shared standards.

Data governance

Data governance is a foundation of AI governance because AI systems depend on data as input. If data is inaccurate, biased, or poorly managed, the system will reflect those problems.

Data governance typically includes:

data quality, such as accuracy and completeness
data access and sharing rules
data privacy and protection
data documentation and transparency
accountability for how data is collected and used

Further information

AI governance

AI governance refers to how AI systems are shaped, constrained, and managed in practice.

This includes institutional rules and platform-level controls that influence how AI is developed and used.

There are different levels of AI governance.

Government and international frameworks

Examples include:

These frameworks define general expectations such as transparency, fairness, privacy, and accountability.

They guide national policies and organizational decisions, rather than regulating specific tools directly.

Institutional policies

Organizations develop their own rules for using AI in practice.

Universities defining acceptable AI use accross campus (see UK Advance Team’s guidlines and recommendations).

Publishers setting authorship and disclosure policies. Examples include Sage, Springer, Elsevier, and Taylor & Francis. Also see COPE Position Statement on Authorship and AI Tools.

Professional organizations issuing guidelines that shape how AI is used in practice. For example:

These policies translate broad principles into specific contexts.

In-class activity: Comparing AI policies across contexts

Review 2–3 example AI policies from different types of organizations.

For example:

a university AI policy
a corporate AI policy
a professional or publisher guideline

As you read, focus on:

how the policies differ in what they allow or restrict
how they treat data use, privacy, or authorship
how responsibility is assigned

Then consider:

which policy feels more strict or more flexible
what assumptions each policy makes about its users
how the context (e.g., university vs company) shapes the rules

Platform and corporate governance

AI systems are also governed by platform-level rules. Examples include Claude’s Constitution developed by Anthropic and OpenAI’s usage policies.

These rules shape not only what users can do, but also how organizations deploy AI and how data is collected and processed. They may change over time.

Also see OpenAI’s suggestions: Industrial Policy for the Intelligence Age: Ideas to Keep People First

How to evaluate AI

AI systems should be evaluated based on how they perform in real tasks, not only on technical benchmarks. Evaluation typically includes:

Task performance, such as accuracy, relevance, or completeness
Consistency across repeated runs
Robustness, such as how the system handles noisy, misleading, or intentionally difficult inputs
Safety, including harmful or inappropriate outputs
Usefulness, such as whether the output actually supports user needs

Evaluation can take place at different stages:

during development, using test datasets or benchmark tasks
before deployment, using predefined tasks or situations used to test how an AI system performs
after deployment, through user feedback and real-world monitoring

Evaluation is an ongoing process rather than a one-time check. Systems may change over time, and their performance must be continuously assessed.

Further information

Common principles across frameworks

Across frameworks such as OECD and UNESCO, several ideas appear repeatedly:

transparency
fairness
privacy
accountability
human oversight

These principles are defined differently across contexts, but they reflect shared concerns about how AI systems should operate.

Challenges of AI governance

AI governance involves tradeoffs.

Regulation often lags behind technical change. Rules developed in one country may not apply in another. Restricting systems may reduce risk but also limit access or innovation. Open systems make AI easier for more people to use, but also make it harder to set limits on how they are used, because they can be freely copied, modified, and deployed without centralized control.

There is no single solution that resolves all of these tensions.

Further information

Future of AI

Reports and large-scale perspectives

Industry and research reports provide overviews of AI development, such as the The AI Index Report by Stanford. These reports track trends in investment, performance, and adoption.

Expert and industry perspectives

Different people describe the future of AI in different ways. Examples include industry leaders, researchers, and engineers. These perspectives are not neutral. They reflect different assumptions about risk, control, and progress.

Some commonly discussed directions include:

These developments affect how information is created, accessed, and managed.

In-class activity: AI futures in practice

Find a recent talk, podcast, or interview where someone discusses AI trends or issues.

You may use suggested examples or choose your own.

As you listen, focus on:

what kind of trends is being described
what problems are emphasized or ignored
what assumptions are made about users, institutions, or society

Think about:

whether this perspective seems realistic
how it might relate to your field