Distinguishing between Internet Governance and AI Governance

· 8 min read
By

The UN General Assembly has mandated a Global Dialogue on AI Governance involving governments and all relevant stakeholders on the margins of existing UN meetings. We have had many constructive and nuanced conversations with stakeholders on designing this dialogue. Recently, some stakeholders have suggested the Internet Governance Forum (IGF) as the umbrella meeting for this dialogue. In this short post we outline why this is a suboptimal choice. 

We fully support the IGF and its mandate extension as part of the WSIS +20 review conference. However, as the name indicates, the IGF is an Internet governance conference. It is not an AI governance conference. In our own analysis of options for the modalities of the Global Dialogue on AI Governance, we included the IGF as an analogy from which to draw important lessons, not as a structure in which the AI dialogue could be subsumed. 

The Internet and AI have two distinct technical communities

The Internet is a communication protocol for sending data packets between computers in a standardized way. AI is an information technology that processes inputs and creates outputs.

The Internet Protocol suite consists of handcrafted and well-understood standards. In contrast, AI models are trained on vast amounts of data and can learn to make decisions in complex and hard to understand ways. 

“The” Internet – singular – is a global communication network. AI systems – plural  – can sit at the nodes of a computer network and be accessed via the Internet. However, AI systems can also be deployed on-premise in companies and on consumer devices without the Internet.

Overall, the Internet and AI are different technologies on different layers of the tech stack. To make this more concrete for a governance dialogue:

Institutions

The multistakeholder community of Internet governance is centered around three institutions: the IETF (standards), ICANN (assigning unique names and numbers), and the IGF (broader societal impacts).

AI has no equivalent to the IETF. The IETF is a body consisting of individuals (mostly employees of large tech companies and academia) that standardize the evolution of interoperable communication protocols. While relevant AI standardization efforts exist there is no unified body governing AI development. AI is an information technology; it still relies on Internet standards for telecommunication. OpenAI, Deepmind, Anthropic, Meta, and Deepseek don’t need to use a standardized neural architecture so that their AIs can talk to each other.

AI has no equivalent to ICANN. Again, the Internet is a global communication network. To have globally unique identities, this requires the management of domain names (e.g. “simoninstitute.ch”) and of IP-numbers (e.g. “192.0.2.1”). AI is an information technology. AI services accessed via the Internet still use the domain name system (e.g. “openai.com”). However, there is no need for an equivalent system to coordinate which English word can be used by a specific AI language model.

Communities

The Internet and AI communities meet at different conferences. The IGF is a multistakeholder summit on broader societal implications of the Internet. The closest equivalent community for AI governance could currently be met at the AI for Good Summit or the Global Partnership on AI. It seems that there is limited overlap between the community attending these AI conferences and the Multistakeholder Advisory Group (MAG) of the IGF.

The Internet and AI governance communities have different leaders: Based on public knowledge, Vint Cerf – the American internet pioneer who leads the IGF Leadership Panel – has contributed to Internet standards but has never been involved in training an AI model or published a paper on AI. Vice versa, one of the “godfathers of AI”, Yoshua Bengio, chairs the International Scientific Report on the Safety of Advanced AI because he has made crucial contributions to AI research. However, he has never contributed to Internet standards.

The Internet and AI communities have different jargon: Every epistemic community has its abbreviations and jargon. For example, in the context of Internet governance “autonomous systems” refers to computer networks that have been assigned a common routing policy. In the context of AI, “autonomous systems” refers to systems capable of independent decision-making. This is but one example among many to showcase that AI and the Internet are better off in well-differentiated fora.

Internet and AI governance have overlaps, but so do many other topics

There are relevant overlaps and intersections between Internet governance and AI governance. We support the discussion of issues relevant to both the Internet and AI at the IGF. Similarly, the Global Dialogue on AI Governance will at times have to include Internet-related aspects.

However, there are also aspects of Internet governance that have little to do with AI governance and vice versa. 

A few examples of distinct issues in Internet Governance include:

  • IPv6 transition: Devices need unique addresses to communicate. The older IPv4 standard is running out of addresses, so IPv6 expands the pool.
  • Verification of data routes: The BGP (Border Gateway Protocol) routes data between networks. BGPsec aims to secure these routes to prevent hijacking or misdirection of Internet traffic.
  • Domain name system: Someone has to decide who gets to use top-level domains like “.amazon” This ensures unique, consistent naming on the Internet.
  • Internet shutdowns: Governments or other actors can shut down the Internet at certain points. This raises governance, human rights, and technical resilience questions.

A few examples of distinct issues in AI governance include:

  • Interpretability: Complex AI systems can be “black boxes.” Improving transparency helps users and regulators understand and trust AI decisions.
  • Alignment: AI should act in line with human intentions and ethical values. Misaligned AI could behave in harmful or unintended ways, especially when reinforcement learning is involved.
  • Liability: If an AI system makes a harmful decision, who is legally responsible? This has implications for developers, deployers, and users.
  • Patents: With AI-generated innovation questions arise around whether AIs can legally be “inventors,” and how to handle intellectual property when AI creates new solutions. 

A relevant overlap cannot be a sufficient condition to subsume the Global Dialogue on AI Governance within a differently scoped event. As a general-purpose technology, AI has overlaps with UN events in nearly every domain. For example, the following all have relevant overlaps with AI:

 It’s great to discuss AI within these domains. However, we think the added value of a Global Dialogue on AI Governance would really stem from focusing on aspects that are uniquely relevant to AI and by connecting AI governance discussions across domain siloes.

Both Internet governance and AI governance matter – a haphazard combination would be detrimental to both communities

If the IGF were to make a pivot and sustainably shift towards AI going forward, the Internet governance community would rightly complain that their conference has been hijacked. The stability of the Internet governance regime depends on a delicate balance between its three core institutions: IETF, ICANN and IGF. Merging the agenda of one of them with AI governance could have significant unintended consequences.

AI governance involves a set of national regulatory measures to control diffusion dynamics, whereas Internet governance typically emphasizes a single, global Internet that should be defended against fragmentation. If these dialogues are merged under the IGF, how might these distinct frameworks be reconciled? If the Global Dialogue on AI Governance becomes part of the IGF, is there a need to replace existing members of IGF leadership structures with members with more AI expertise? More fundamentally, if policymakers are unable to distinguish between AI governance and Internet governance, might observers not question whether AI for Good, the WSIS Forum, and the IGF are duplicative, rather than complementary?

We think that the IGF remains a good model of global multistakeholder engagement that can offer lessons for other domains. However, Internet governance and AI governance are not literally the same and we would encourage a complementary rather than a competitive approach between these two domains. Accordingly, the UNGA-mandated Global Dialogue on AI Governance should have focused, expert-driven, and dedicated attention. It should not be a side topic at a conference focused on a different subject.