The Politics of Knowledge in the AI Age: Empowering Human Agency through Standards, Human Rights, and the Role of Corporations

João Brant argues that if AI is left to its own devices, it will reinforce existing divides and extract value produced in developing countries without fair compensation. AI governance must therefore ensure that the economic benefits generated by AI systems are proportional to the contributions made by users and companies in each country.

Brant explicitly states his concern that AI services, if unregulated, will reinforce extraction of value produced especially in developing countries , and asserts that AI governance must ensure economic benefits are proportional to contributions made by users and companies in each country .

Brant contends that information, culture, and knowledge — the main inputs of generative AI — are often left out of economic discussions about AI. Journalism, artistic production, and science must be adequately compensated to prevent predatory exploitation and to sustain the virtuous cycle of creation that benefits society as a whole.

Brant notes that data as information, culture, and knowledge has a great impact on the economy of AI and should be given value , and that journalism, artistic production, and science must be adequately compensated to avoid predatory exploitation and the disruption of a virtual cycle of creation that benefits the whole of society .

Brant argues that the primary key performance indicator for digital platforms — maximising user engagement — is part of their business model but has negative consequences for public knowledge and the information ecosystem. When engagement maximisation is the dominant metric, the system is driven away from protecting knowledge and information integrity.

Brant states that maximisation of engagement is probably the main KPI for platforms and digital services , and that when taken as the key performance indicator, it drives the system in a different direction from protecting knowledge and information integrity . He argues that safeguards and protections for the public interest must be reflected in standards .

Brant argues that decades of defending non-liability for digital platforms regarding third-party content has produced negative externalities. Brazil's Supreme Court found this system partially unconstitutional and introduced a temporary arrangement based on shared responsibility for paid and boosted content and a duty of care for serious criminal offences.

Brant references Brazil's Supreme Court ruling that the non-liability system is not capable of protecting human rights entirely, finding it partially unconstitutional and introducing a new temporary arrangement . This arrangement includes shared responsibility for paid and boosted content and a duty of care or due diligence approach regarding serious criminal offences .

Brant contends that platforms which actively interfere with how content is distributed should bear greater responsibility than those that merely host content neutrally. He calls for courage in discussing liability, acknowledging both the positive effects of the non-liability approach and its negative consequences, in order to protect the information and knowledge ecosystem.

Brant distinguishes between neutral hosting services and those that drive the direction of content consumption, arguing that when platforms impact and interfere with how content is distributed, responsibility should be greater . He calls for a balanced approach to liability that protects the information and knowledge ecosystem within the digital realm .

Brant argues that AI companies have been mining and training their systems on copyrighted material without compensating rights holders, gaining value without fair remuneration. He advocates for copyright clauses specifically addressing AI mining and training to preserve the virtuous cycle of human creation and prevent society from becoming overly dependent on machine-generated content.

Brant describes how AI companies have been taking and training and mining their systems without compensation for rights holders , and states that Brazil is defending a copyright clause for mining and training copyrighted material within its AI bill . He emphasises the need to keep the virtual circle of creation and especially creation by human beings .

Brant argues that governments should be more engaged in supporting the informational environment, including funding initiatives that promote information integrity and digital public goods. He acknowledges the challenge of doing so without introducing bias or affecting the diversity and pluralism of content.

Brant references Brazil's contribution to a UNESCO-held fund for information integrity and climate change, which distributes support to different groups including Wikimedia . He also notes the challenges of domestic support due to the need to start from taxation of digital platforms, which is currently a contentious topic .

Isabel Ebert argues that technical standards, despite appearing neutral, function as governance infrastructures that shape the architecture of the information environment. In ecosystems dominated by a small number of players, these standards can unintentionally reinforce power asymmetries.

Ebert states that standards shape the architecture of the information environment and can unintentionally reinforce power asymmetries, in particular in ecosystems dominated by a small number of players . She argues that human rights standards would try to debunk this black box by requiring more transparency, inclusive participation, and public interest representation in standard-setting bodies .

Ebert argues that technical standards are not merely technical but function as governance infrastructures with significant political implications. They define data formats, identity systems, interoperability rules, and algorithmic parameters, and in doing so determine how knowledge is distributed and who controls knowledge flows.

Ebert explains that both standards of conduct in corporate governance and technical standards need to work together to define data formats, identity systems, interoperability rules, and algorithmic parameters , and that those choices determine how knowledge is distributed and who controls knowledge flows .

Ebert asserts that human agency and control must be the central anchor of AI governance. Technical standards increasingly determine who can access information and how content is filtered and amplified, and these choices have human rights impacts, including on cognitive liberty and freedom of thought.

Ebert states that human agency and control must be the key anchor of AI governance , and that technical standards increasingly determine who can access information, how identities are anticipated in terms of content provenance, and how content is filtered and amplified . She argues that human rights can help use standards to protect cognitive liberty, avoid manipulative design patterns, and ensure freedom of thought .

Ebert argues that human rights due diligence is key to ensuring that the impacts on people are mitigated in relation to technical design choices. Companies must support open, interoperable, and transparent standards that allow for understanding of information asymmetries and empower users to uphold their agency.

Ebert references the OECD's Responsible AI Due Diligence Guidance, which translates standards of conduct and marries them to technical standards, bridging community gaps . She states that human rights due diligence is key to ensure that impacts on people are mitigated when it comes to technical design choices , and that companies must support open, interoperable, and transparent standards that empower users .

Ebert contends that standards governing AI and automated systems must be designed to be contestable and explainable, ensuring that individuals retain meaningful agency. Standards that are opaque or that override individual agency in automated systems are incompatible with human rights principles.

Ebert explicitly states that standards need to be contestable, explainable, and cannot be used to override individuals' agency in automated systems .

Ebert argues that standard-setting processes currently dominated by industry and government must be broadened to include civil society, academia, affected communities, and global majority actors. Without such inclusive participation, standards risk encoding narrow commercial priorities or geopolitical interests rather than serving the public interest.

Ebert states that standard-setting processes need to include civil society, academia, affected communities, global majority actors, and not only industries and government, which is the current status quo . She notes that public interest participation needs to ensure that standards do not encode narrow commercial priorities or geopolitical interests and that they ensure plurality in the knowledge ecosystem .

Ebert argues that transparency in standard-setting processes is essential for democratic legitimacy. This includes open documentation, public comment periods, and accessible meeting records, which together enable democratic participation and ensure that technical infrastructures support pluralism and reliable access to information.

Ebert states that transparency in standard-setting, including open documentation, public comment periods, and accessible meeting records, is essential for democratic legitimacy, and that democratic participation can be enabled if technical infrastructures support pluralism, plurality of voices, transparency, and reliable access to information .

Ebert raises concern about whether society is locked into large language models as the dominant approach to AI-driven knowledge production, and suggests that alternative systems and approaches should be explored. She expresses hope that a more pluralistic and freedom-enabling information ecosystem can be restored, drawing on the early ideals of the internet.

Ebert questions whether we are stuck with LLMs as a sort of information-producing approach or might be able to develop different systems and approaches . She draws on the early days of the internet, referencing Tim Berners-Lee and Wendy Hall, noting that the internet we currently have is not the internet they imagined , and expresses optimism that a pluralistic, freedom-of-the-mind ecosystem can be restored .

Joël Christoph explains that the politics of knowledge is a contest over three decisions: what gets recorded, what gets trusted, and what gets found. Historically these powers were split among different institutions, but AI risks concentrating all three in private hands at unprecedented scale, with most systems lacking transparency or participatory governance.

Christoph describes the politics of knowledge as a contest over what gets recorded, what gets trusted, and what gets found, each relating to control . He notes that AI risks concentrating all three, unlike previous information systems such as libraries, school curricula, and TV and radio broadcasts . He observes that choices are now predominantly made inside digital systems developed in the private sector at enormous scale across borders, and most are not fully transparent or participatory .

Christoph argues that thousands of languages are poorly served by digital standards, meaning that whole bodies of knowledge are nearly invisible to AI systems. This is not because such knowledge lacks value, but because no standard has made it machine-readable — a political outcome that cannot be reduced to a technical gap.

Christoph states that whether a particular human language is well-supported in digital standards can decide what speakers can search, what is understood by others, and what appears in the data that trains AI . He notes that thousands of languages are relatively poorly served , and that whole bodies of knowledge are close to invisible to some of these systems, not because they lack value, but because no standard has made them machine-readable - a political outcome that cannot be reduced to simply a technical gap .

Christoph argues that technical standards are particularly important relative to law for three reasons: they are durable and difficult to reverse once industry builds on them; they encode the interests of those present when they are written, often large corporations and well-resourced governments; and they operate beneath usual democratic scrutiny, unlike parliaments, yet affect billions of people.

Christoph identifies three properties that make standards particularly important relative to law: their durability, as once an industry builds on a particular standard, changing course is slow and costly ; the distribution of power, as those in the room when the standard is written can encode their interests against those not present ; and that standards operate beneath usual democratic scrutiny, with technical committees making decisions affecting billions of citizens .

Christoph argues that while we tend to assume major decisions about knowledge and freedom are made in courts and parliaments, they are increasingly decided upstream in technical communities, through defaults, private corporations, and expected formats. Standards set the terms within which laws, markets, and rights will operate.

Christoph states that while standards will not solve everything on their own, they will set the terms within which laws, markets, and rights will operate . He argues that we tend to assume the big decisions about knowledge and freedom are made in courts and parliaments, but more and more they are decided upstream and earlier, in technical communities, through defaults, private corporations, and the formats that are expected .

Christoph argues that a human rights-based approach to standardisation requires three things in practice: bringing rights holders and affected communities into the design stage; building the ability to question and appeal into the standard itself from the start; and aligning incentives so that the rights-respecting option is also the easiest and cheapest path for companies.

Christoph outlines a human rights-based approach to standardisation that includes bringing rights holders and affected communities into the design stage , building the ability to question and appeal into the standard itself from the start , and aligning incentives so that the rights-respecting option is also the easiest and cheapest path for companies . He references the work of the ITU and OHCHR in bringing a Human Rights Review into standards processes as an example of this approach .

Jenna Omassi argues that TikTok's commitment to human rights is embedded across its overarching policies and frameworks. These include community guidelines grounded in human rights standards, privacy policies, guidelines on handling government removal requests, and transparency reporting, all anchored in the UN Guiding Principles on Business and Human Rights.

Omassi explains that TikTok's community principles, which sit on top of its community guidelines, are based in human rights standards and ensure that community guidelines are led by international human rights commitments including the UN guiding principles . She also references guidelines on government removal requests developed with freedom of expression and privacy in mind , privacy policies applicable in every region , and quarterly community guidelines enforcement reports and transparency reports .

Omassi argues that responsibility for human rights is shared between governments and companies such as TikTok. TikTok has established a cross-functional human rights working group that coordinates and collaborates on human rights initiatives and incorporates the voices of the multi-stakeholder community and multilateral institutions.

Omassi states that responsibility for human rights is shared, with governments having a duty to protect human rights and TikTok and other businesses sharing that same duty . She describes TikTok's cross-functional human rights working group, which coordinates and collaborates on human rights initiatives, and notes that she sits on this group representing public policy views and ensuring that multi-stakeholder and multilateral voices are brought into the conversation .

Omassi argues that TikTok has taken concrete steps to address synthetic media through automatic AI labelling implemented in 2023, which has been used tens of millions of times. TikTok also participates in the C2PA initiative to identify and label synthetic media created on other platforms and brought onto TikTok.

Omassi states that TikTok was one of the first companies to implement an AI label upon content creation in 2023, used tens of millions of times to date . She also notes TikTok's engagement in cross-industry initiatives like C2PA to ensure that content created on other platforms and brought onto TikTok can be identified as synthetic media and labelled . Additionally, TikTok launched a $2 million AI literacy fund following the AI for Good Summit .

Omassi presents TikTok's approach to election integrity as a case study in applying human rights principles practically. TikTok removes harmful misinformation about civic and electoral processes, uses IFCN-accredited fact-checkers, bans political ads, and partners with election authorities and experts to counter emerging threats.

Omassi describes TikTok's election integrity measures, including removing harmful misinformation about civic and electoral processes , using more than 20 IFCN-accredited fact-checkers to assess the accuracy of claims and labelling unverified claims , banning political ads and not allowing politicians, governments, or political parties to monetise on the platform , and partnering with election authorities and experts to counter emerging threats .

Omassi signals that TikTok is beginning to consider how it engages in international standards processes, including at the ITU and other settings. TikTok recognises that its voice can contribute meaningfully to the multi-stakeholder space and expects to become more involved over time.

Omassi states that TikTok is beginning to consider how it engages in international standards processes and that one can expect to see more of TikTok slowly but surely, both at the ITU and in other settings, where it believes it can make a meaningful contribution to the multi-stakeholder space .

Omassi argues that the multi-stakeholder setting for deliberation forms the foundation for human rights-based governance in advancing shared goals of promoting information integrity, community engagement, and human rights. TikTok supports this model through its growing engagement in forums such as WSIS and IGF.

Omassi states that the multi-stakeholder setting for deliberation forms the foundation for human rights-based governance in advancing shared goals of promoting information integrity, inspiring community engagement, and promoting human rights . She references TikTok's growing involvement in the WSIS space and excitement about the permanent IGF mandate and the upcoming IGF in Nairobi .

Jacobo Castellanos argues that standardising content provenance through initiatives like the C2PA enables the authentication of images and videos, which is critical for accountability and justice in human rights documentation. However, proper governance is required to prevent misuse of these standards.

Castellanos explains that Witness has been working on provenance - the source and history of content - as a mechanism to authenticate images and videos for accountability and justice . He describes Witness's journey from developing its own niche tool with the Guardian Project to joining the C2PA to standardise provenance . He notes that his Google Pixel phone now automatically captures provenance information, providing cryptographic evidence of authenticity .

Castellanos argues that despite the C2PA's progress in incorporating human rights considerations, it can still be misused as a surveillance tool, particularly in authoritarian contexts. Governance innovation is needed, including civil society oversight, oversight mechanisms, and protocols to address risks when the standard is misused.

Castellanos references a report published by Witness called 'C2PA and the Surveillance Risks', which argues that despite everything that has been done, the C2PA can be misused and turned into a tool for surveillance . He argues that governance innovation is needed, including bringing in civil society, oversight mechanisms, and protocols so that when the C2PA turns into a surveillance tool, the right governance mechanisms are in place .

Castellanos acknowledges that watermarking and provenance tools, while valuable for authentication, risk enabling the targeting of individuals in authoritarian regimes. Technical backstops and privacy-respecting design are necessary safeguards to mitigate this risk.

Responding to an audience question about authoritarian regimes, Castellanos acknowledges that as a human rights organisation working globally, Witness wants to ensure there are backstops or red lines established in standardisation and potentially in international law . He notes that the C2PA's conformance programme provides a gate at which misuse can be controlled , and that there are ways to ensure privacy-respecting watermarks, with technical backstops available even if they cannot be guaranteed in authoritarian regimes .

Castellanos acknowledges that the C2PA, despite being a private coalition controlled by private companies, has made significant progress in incorporating human rights considerations, including publishing a harm assessment alongside its specifications. However, governance innovation is still needed to address remaining risks.

Castellanos states that the C2PA deserves recognition for incorporating human rights considerations, including a harm assessment published alongside the specs . However, he notes that even this has not been enough, referencing the newly published report on C2PA and surveillance risks . He argues that governance innovation is needed, including oversight mechanisms and protocols to address risks when the standard is misused .

Castellanos argues that legislative initiatives in the EU, the United States, and elsewhere are increasingly referencing standards like the C2PA for regulatory compliance on AI transparency and provenance, without ensuring that there is public interest oversight. Standards used for regulatory compliance must be subject to public control.

Castellanos notes that the EU, the United States, and various states have legislative initiatives that have been passed or are being discussed that require provenance to tackle transparency issues related to AI . He argues that while provenance is needed to authenticate content, it must come with the right governance and standardisation processes, otherwise legislation points to standards like the C2PA without ensuring public interest and oversight . He concludes that standards used for regulatory compliance must also be in the public interest and under public control .

An audience member raises the concern that oral histories and non-written forms of knowledge risk being excluded from AI training corpora, and asks how this can be addressed. The question also raises copyright issues that may bar knowledge from being included in corpora, and asks how digital public goods such as Wikipedia and Wikidata can be supported.

The audience member asks how oral histories of things that are not written are dealt with, and what happens with copyright issues when copyright can bar knowledge from being included in a corpus . They also ask how digital public goods such as Wikipedia and Wikidata could be supported to strengthen the international structure .

Olivier Alais argues that AI systems increasingly determine what people see online, what appears credible, and what disappears, and that these effects are not merely technical but are deeply social and political in nature. He frames this as a fundamental challenge for governance and society.

Alais states that AI is shaping how knowledge is produced, distributed, and trusted, and that what appears credible and what disappears is increasingly shaped by infrastructure, standards, platforms, and algorithms . He explicitly notes that these systems are technical but their effects are deeply social and deeply political .

Alais defines the politics of knowledge as a set of questions about control over information flows and access, and about maintaining human agency at the centre of an AI-reshaped environment. This framing positions the governance of AI and technical standards as fundamentally political rather than merely technical matters.

Alais explicitly defines the politics of knowledge as concerning who controls the flow of information, who can access it, and how human agency is kept at the centre as AI reshapes its environment .

Alais argues that technical standards are not merely neutral technical instruments but are central to the broader conversation about who controls knowledge and how. He also highlights that companies are essential actors in this space, both shaping standards and implementing them in real-world services.

Alais states that technical standards are central to the conversation because they are not neutral, and that the role of companies is also essential because they help shape standards, implement them in real-world services, and operate the infrastructure through which people access knowledge .

Alais frames the responsibility of large platforms by highlighting their significant influence over what content people encounter online. He poses the question of how platforms translate human rights principles into concrete design choices, including in recommendation systems, transparency, user agency, and access to diverse information.

Alais notes that companies such as TikTok play an important role in shaping what people see, discover, and engage with online, and asks how TikTok works to implement human rights principles across the design of its platform, including recommendation systems, and how it works to strengthen transparency, user agency, safety, and meaningful access to diverse information .

Alais frames the challenge of content provenance standards as a dual one: they can be valuable tools for combating misinformation and disinformation, but they must also be designed to anticipate misuse and to protect vulnerable communities, including those documenting human rights abuses. This framing highlights the need for governance alongside technical design.

Alais asks how standards for digital content provenance can help address myths and disinformation whilst also anticipating possible misuse, protecting vulnerable communities, and preserving the safety, credibility, and agency of people documenting human rights abuses .

Alais argues that AI governance and technical standards should be evaluated against the standard of whether they support democratic access to knowledge, human agency, and non-discrimination. He frames the central governance challenge as preventing these systems from reinforcing existing power asymmetries.

Alais poses the question of how AI governance and technical standards can help ensure that digital technologies support democratic access to knowledge, human agency, and non-discrimination rather than reinforcing existing power asymmetries .