Global Perspectives on Openness and Trust in AI

Contents

Executive Summary

This panel discussion reframes "openness" in AI governance beyond technical definitions (open-source models, shared weights) to encompass broader sociotechnical characteristics including accountability, democracy, transparency, and community participation. Panelists from the US government, France, India, and global media argue that genuine openness requires distributed power and agency, and that middle-income and Global South countries must develop alternative approaches to AI development rather than adopting or competing within existing corporate-led ecosystems.

Key Takeaways

Openness ≠ Technical Transparency: Openness requires sociotechnical dimensions—accountability, democratic participation, power redistribution, transparency in governance, not just code sharing. Models like Llama 2 shared weights but lack the democratic infrastructure of true open-source.
Middle Powers & Coalitions Strategy: Rather than competing head-to-head with US/China, medium-income countries can form ad-hoc coalitions, pool compute/data resources, and support domestic startups via public digital infrastructure and open-source leverage to build digital sovereignty.
Community & Friction as Features: Including communities in AI development (Tahiku Media model) requires building trust through education, transparency, and co-design—processes that create "friction" but enable genuine agency, consent, and value distribution, not just faster adoption.
Beware Adoption Narratives: For Global South countries, the narrative that "openness/AI adoption solves inequality" risks misdirecting resources and turning populations into data/labor extraction sources. Governance, control, and strategic priorities must precede adoption.
Individual Agency Amid Inevitability: Consumers aren't helpless; alternatives exist (refuse use, choose open tools, join protests, use litigation). Third-party labeling systems (like those in fashion/food) could enable informed AI choices. Governance requires enforcers (competition, labor, privacy authorities) at the table alongside technologists.

Key Topics Covered

Redefining "Openness" — Moving beyond technical open-source to include accountability, transparency, and democratic participation
US AI Policy & Governance — Shift from Biden administration's gradient approach to current administration's binary stance; use of trade, tariffs, and export controls as de facto AI regulation outside democratic oversight
Geopolitical Realignment — China's use of open-source as a competitive lever; emergence of "middle powers" (France, Germany, India, Canada, Japan) as organizing principle replacing US/China binary
Global South & Market Dependence — Risks of framing openness as adoption pathway; labor exploitation and data extraction concerns; need for data sovereignty and control
Competition & Antitrust — Market concentration across AI value chain; ecosystem lock-in; IP protection vs. innovation; enforcement mechanisms in absence of traditional regulation
Community-Driven AI — Examples of participatory model development (BLOOM, Tahiku Media speech recognition); small AI vs. large language model paradigm
Labor & Data Ethics — Intellectual labor exploitation; lack of compensation for training data; data workers' rights
Corporate Capture of Language — Critique of "sophisticated corporate speak" appropriating inclusion/diversity language to lock in closed platforms

Key Points & Insights

Openness as Power Shift: The Biden administration conceptualized openness as a "gradient" reflecting deeper principles from open-source software (accountability, shared infrastructure, community modification rights), whereas the current administration treats it as a binary technical achievement. True openness requires redistributing power.
Geopolitical Leverage of Open-Source: China's strategic adoption of open-source (via DeepSeek) demonstrates that open models function as equalizers for challengers to catch up with incumbent powers—similar to Android's role in smartphones. This is a tool other middle powers can leverage.
Governance Through Non-Democratic Means: The US is using tariffs, export controls, semiconductor restrictions, and immigration policy (H-1B visa costs ~$100K per worker) to steer AI development—heavier-handed than regulation but bypassing democratic public notice and comment processes that formal rulemaking requires.
Access, Not Just Adoption: For Global South countries, "openness" framed primarily as adoption pathway risks directing resources toward technology adoption rather than solving structural issues (health, education, sovereignty). Without control over the entire AI stack, populations risk exploitation as data/labor sources.
Competition as Sovereignty Tool: India's Competition Commission identifies systemic risks—self-preferencing, tying/bundling, ecosystem lock-in, targeted price discrimination—across digital markets and AI. Competition policy is essential for maintaining contestable markets and preventing dominant firms from foreclosing innovation.
Participatory Model Development Works: BLOOM (1,000+ researchers, 70 countries, open data governance) and Tahiku Media (community-driven Māori speech recognition with consent-based data collection) demonstrate that participatory openness scales differently—not as single monopoly distribution but as many communities building models for local contexts.
Scale Reframing: "Scale" under Silicon Valley logic means monopoly distribution (one entity, many users). True scale means diverse industries/communities each developing contextual models. Most industries are "data poor"; distributing AI requires application-specific, small models rather than universal large language models.
Infrastructure Opacity: Data centers and cloud infrastructure layer is explicitly closed to community input—NDAs prevent elected officials from disclosing terms, excluding democratic deliberation from foundational AI infrastructure decisions.
Gender & Inclusion Gaps: The panel noted it was the summit's only all-female panel; gender underrepresentation mirrors broader inclusion rhetoric that conflates "inclusion" with market adoption/consumption rather than power and voice in design/governance. Lunar New Year and Ramadan timing also reduced Global South representation.
Open-Washing Risk: Genuine openness must be distinguished from corporate appropriation of inclusion/diversity language to market closed platforms. Competition authorities need refined analytical tools to assess whether "openness" genuinely lowers barriers or maintains underlying dependencies.

Notable Quotes or Statements

Alondra Nelson (Biden Admin, OSTP): "Openness is not a binary it's either open or not open... the broader understanding of openness that comes out of open-source software it was about shifting power. It was about forms of accountability."
On Current US Policy: "It may not be regulatory in the sense of formal rule making... but it is certainly hyper regulatory... and unfortunately I think anti-democratic relative to the status quo" [lacking democratic input mechanisms].
On Middle Powers: "These middle economies can do a lot of things... ad hoc coalitions of the willing... [that] can be useful in the evolution of governance" (French official).
Asa Yadav (India): "Openness as a driver of adoption is actually quite a dangerous frame for global south countries because it moves attention from where we might need to invest our resources... we are not here to do the labor to test bed models built elsewhere."
On Scale Reframe (Karen Hao): "Scale doesn't mean they distribute to everyone... they are the sole distributor... that's a monopoly. What we want is different communities all around the world each developing models by and for them at scale."
Karen Hao on Corporate Language: "It's so interesting to observe corporate speak... they have adopted the language of inclusion diversity empowering marginalized communities to talk about ultimately selling their technology... and locking in their closed platforms."
On Individual Agency: "There are a thousand different touch points for how you can interact with the AI supply chain... In each of those touch points you can choose whether to resist or adopt or be neutral."

Speakers & Organizations Mentioned

Role/Title	Name	Organization
Panel Moderator	Amba Kak	AI Now Institute
Former Deputy Director, OSTP	Alondra Nelson	White House Office of Science and Technology Policy (Biden Admin)
French Special Envoy for AI	[Name not clearly stated]	French Government / AI Action Summit (Paris)
Deputy Chair/Commissioner	Asa Yadav	Competition Commission of India
Chair	Chairperson Cory	Competition Commission of India
Journalist & Author	Karen Hao	Author of Empire of AI
Co-host	[Mentioned]	API Institute (Bangalore)
Partners	Amlan Muanti, Sanja Mishra, Iksho Virat	Summit organizing team

Other Entities Mentioned:

Tahiku Media (New Zealand nonprofit radio, Māori language)
BLOOM Project (1,000+ researchers, 70 countries, 250 institutions)
DeepSeek (Chinese open-source AI model)
Mistral, Kohir, Sakaki (European/Japanese startups using open-source)
Mozilla Foundation Deep Speech (open-source model)
Munich Security Conference, Davos, Paris AI Action Summit (venues)
Mark Carney, Emmanuel Macron (speakers referenced)

Technical Concepts & Resources

Concept/Tool	Description	Context
Open-Source vs. Openness	Technical sharing (code/weights) vs. sociotechnical practice (accountability, transparency, democratic participation, infrastructure sharing)	Core reframing of the panel
Llama 2 / Llama 3	Meta models; criticized as "not really open source" despite shared weights; lack broader democratic infrastructure	Example of "open-washing"
BLOOM LLM	Large language model co-created by 1,000+ researchers across 70 countries; featured better data governance, curation, transparency, value attribution to data sources	Example of participatory openness at scale
Tahiku Media AI Speech Recognition	Community-driven speech recognition for Māori language revitalization; consentful data donation (inspired by Mozilla Deep Speech); community co-design of applications	Example of smaller-scale, high-agency AI development
Mozilla Foundation Deep Speech	Open-source speech recognition trained on consentful data donations	Foundational technology for Tahiku Media project
Market Study: AI & Competition	India's Competition Commission report (released October 25); identified concentration risks, ecosystem lock-in, price discrimination, opacity, exclusive partnerships across AI value chain	Policy/governance tool
Digital Public Infrastructure (DPI)	India's approach via digital identity, digital payments; proposed extension to compute platforms, datasets, small language models	Governance lever for middle powers
Stargate Project	US government/private investment announced by Trump administration; message of US market dominance	Geopolitical context
H-1B Visa Cost	~$100K per high-tech worker (cited as 10-20x typical salary); used as lever to restrict AI talent flows	Trade/immigration control mechanism
Cooperative Models	One-member-one-vote governance (mentioned AMU co-op); alternative to hierarchical corporate structures	Governance/organizational model
Data Protection By Design	Research at University of Chicago and elsewhere on protecting public data (images, text, websites) from AI training through technical obfuscation	Labor/IP protection approach

Assessment of Quality & Gaps:

This transcript captures a substantive policy-oriented discussion with specific names, institutions, and examples. However, some limitations:

Speaker identification: Some panelists' full names/titles are partially unclear (French official not fully named; "Chairperson Cory" attribution); transcription errors ("Alandre" vs. standard spelling; "Iksho Virat" may be spelled differently)
Incomplete citations: Books/reports referenced ("Empire of AI", India's Competition Commission report of Oct. 25) could benefit from full publication details
Tonal elements: The transcript captures real-time panel dynamics (note-taking, selfies, applause) showing community engagement but less relevant to substantive summary

The discussion prioritizes political economy, governance, and power dynamics over technical AI capabilities—a deliberate framing choice evident in opening statements.