ATAIN - He Ara Mauri: Asking five large language models to calculate their value

 This session of ATAIN presented by Tanya Ruka (He Herenga Waka - Victoria University of Wellington).

She presents on:

He Ara Mauri: Asking Five Large Language Models to calculate their value

He Ara Mauri is a vitality framework grounded in Indigenous relational ontologies that asks a different question of AI: does this system sustain conditions for life, or degrade them? Applied to five frontier AI systems - ChatGPT, Claude, Gemini, Grok, and Meta AI - each system was asked to evaluate itself and then peer-review the others across planetary, human, and more-than-human domains. The result: every system failed.

Notes taken of presentation:

Began with a introduction to her background - starting from Native Land Digital - maintain one of the most widely accessed indigenous platforms in the world. Their remit includes providing educational resources to be used in classrooms, museums, universities etc. Develop indigenous platform. 

At the moment, funding has ceased but continues with work by passionate people.

Between 207 and 2024, she worked on the Te Aho Tapu Hou project - a Mātauranga Māori led research to create circular / regenerative cottage industries. At Victoria, there is one paper - training Korēro AI - understanding 'value' from an indigenous perspective. What happens when AI is tasked with looking into indigenous value systems.

Introduced the He Ara Mauri framework. Started investigating aspects of data sovereignty and what is indigenous AI. Coined AIRI (Artificial Indigenous relational intelligence) as a term to try to explain through a Native Land data, land and water acknowledgement chatbot.

LLMs build on linear, grab and dash and taking/scraping models. No data is safe. It is an extractive, hierarchical and predictive/control oriented. However, indigenous AI is relational in architecture, bussing from a GNN model (e.g. Deepseek) which are relational and layered. This allows use to be less, drawing more on human knowledge.

Kōrero AI created as a co-designed learning tools to move beyond static land acknowledgement. Instead of providing scripted responses, the chatbot invites users into reflection but asking 3 questions about their relationships to land and water. Across a year Feb 2025 to 2026, a million users have engaged. 

Summarised the process of creating the He Ara Mauri framework for system vitality in socio-technical systems. From this, sought to find out larger LLMs deal with the framework. 5 AI systems evaluated and each evaluated on environmental impacts, human impacts, more than human impacts, power, reciprocity and future consequences. Each system failed as they could not account for ecological consequence, more than human relations, intergenerational responsibility, power concentration and long-term vitality.

The systems could not describe themselves, acknowledge uncertainty, bias or harm, demonstrate responsibility, account of future flourishing and could not explain what they restore, instead of just that they take. Therefore, each system's responses reflected the business model, strategic priorities, and public positioning of the organisation that developed it.

Yet integration across education, public services, etc. is ongoing!!

Proposed some solutions - especially in the education context. Recommends, humans must remain responsible, teach AI critically, protect human knowledge creatin, embed future generations into governance. 

Therefore AI is not simply about efficiencies but we need to understand impacts beyond the now.

Education should be about a community of care and includes stewardship of knowledge, culture, democracy, environmental responsibility, future responsibility and technology should strengthen rather than replace relationships.

A thought provoking presentation bringing us back to the fundamental questions of what makes us human and how AI needs to be carefully handled, not just taken as a tool, but to carefully evaluate where it might fit, how it is used, who and when it is used, and how AI literacy must move beyond just using and evaluating but putting AI through different perspectives. 

Q & A provided good discussion on clarifying the principles presented along with discussion from others on implications and challenges.