Community-Governed Language AI Stewardship Adviser

The mandate

The council repeatedly confronts a question that cannot be reduced to model accuracy: under whose authority may speech, text, cultural context and corrections from distinct Indigenous language communities be collected, transformed, retained and used in public-service AI? Existing proposals assume one consent and access model, even though communities differ in governance, restricted knowledge, revitalisation priorities and preferred forms of benefit.

Three working days per month are reserved for this appointment: one session with community-designated representatives, one technical or public-service challenge meeting, and one day for reading, individual conversations and written advice. Four joint governance-council meetings are included; sensitive ad-hoc questions receive acknowledgement within one business day and a considered response within four, subject to the relevant community protocol.

The term lasts eight months and closes when the council considers the first operating stewardship charter and pilot decision. Any extension must be proposed jointly by both co-chairs, supported by participating language authorities and approved for a stated unresolved question; delayed engineering or procurement does not automatically prolong the appointment.

The adviser holds no line authority, cultural authority, ownership claim, executive responsibility or right to approve data use on another party’s behalf. Participating communities retain their own decisions, public-service officers own statutory and delivery duties, and technical leaders own implementation; advice may shape options but cannot substitute for consent, governance or accountable approval.

Other public-interest, academic and language-technology appointments may continue if disclosed and cadence is protected. Undeclared work for a bidding platform provider, control of a competing dataset, a commercial claim over derived resources, or access to one community’s restricted material through another mandate creates a conflict requiring recusal and potentially an end to the term.

Why the board wants this voice

Community representatives bring authority and language knowledge, while the technical team understands model construction, yet the council lacks someone who has built enforceable governance across both domains. Earlier papers treated consent as an intake event rather than a continuing set of rights over access, correction, withdrawal and benefit. The co-chairs want an experienced translator of governance into operating controls, not a proxy spokesperson for any community.

What you will own

• Press sponsors to recognise each participating language authority separately and record who can permit collection, transformation, evaluation, release, correction and withdrawal.
• Test proposed data journeys against cultural restrictions, provenance loss, model memorisation, access inheritance, vendor reuse and the practical reversibility of consent.
• Shape stewardship terms covering purpose, custody, derived artefacts, benefit, attribution, restricted access, retention, deletion and dispute escalation.
• Challenge evaluation plans to include fluent community-designated reviewers, dialect context, harmful output, service usefulness and uncertainty rather than generic aggregate scores.
• Probe procurement and hosting choices for subcontractor access, cross-border processing, model-provider improvement rights and credible return or destruction of material.
• Guide the council through pilot stop, narrow, continue and expand criteria while ensuring that absence of data is not treated as permission or technical deficiency.
• Produce an advisory record of choices, dissent, unresolved authority, expiring permissions and controls the council should test after the term.

Candidate qualifications

• Built or governed a language, speech, archive or AI programme in which Indigenous or community-designated authorities controlled material decisions.
• Converted consent, restricted-knowledge and benefit principles into enforceable technical access, provenance, retention and supplier controls.
• Designed language-model evaluation with fluent community reviewers and can show how their findings altered data, product or release choices.
• Resolved a disagreement between institutional delivery pressure and community authority without treating consultation as retrospective approval.
• Advised a multi-party governance council where public duties, intellectual property, cultural obligations and machine-learning dependencies intersected.
• Can evidence trusted long-term relationships while clearly stating that prior work does not confer authority to represent a new community.

Non-negotiables

• Can attend four Sydney council meetings and travel to three invited community sessions under the hosts’ agreed timing, conduct and information protocols.
• Will disclose dataset interests, vendor work, research funding, archive access and prior community relationships before receiving any protected material.
• Accepts that a participating authority may decline, narrow or withdraw use and that the adviser cannot negotiate consent privately.
• No protected recording, transcript, model output or governance paper may enter personal tools, external AI services or unrelated research.

1. 49 words maximum. Describe one community-governed language or data project and the decision right that changed its technical design.
2. 49 words maximum. Which present funding, dataset, vendor or community relationship would this council need to assess for conflict?
3. 49 words maximum. How would you test whether a proposed withdrawal right remains operational after model training and derived-data creation?