What's Actually Running in Village Today

Series: Your Conservation Group, Your AI — Understanding Village AI for Environmental Organisations (Article 4 of 5) Author: My Digital Sovereignty Ltd Date: March 2026 Licence: CC BY 4.0 International

Early Days

This article is about what exists today — not what we plan to build, not what we hope to achieve, but what is running right now in production. Where something is planned but not yet live, we will say so plainly.

Village AI has been in production since October 2025. It is a young system. Some parts work well. Some parts are still rough. We believe in telling you both, because an organisation that adopts a platform based on clear information will be a more resilient partner than one that adopts based on marketing.

What Village AI Can Do for Your Conservation Group Today

Answer questions about your organisation's content. When a volunteer asks "When is the next survey day?" or "What did the coordinator report from the autumn bird count?", Village AI searches your organisation's actual records — field reports, stories, event descriptions, management documents — and provides an answer grounded in that content. It does not guess or infer from general knowledge. If it cannot find the answer in your records, it says so.

Help with drafting. Village AI can help draft field report summaries, event announcements, and correspondence. Because it has been trained on your organisation's previous content, its drafts reflect your group's tone and standards — not a generic corporate template. A moderator reviews and edits every draft before it reaches the membership.

Summarise long documents. A lengthy monitoring report or a series of management updates can be summarised into key points. This is useful for volunteers who want to stay informed but do not have time to read everything. For scientific content, the AI is trained to preserve qualifications and uncertainties in summaries rather than smoothing them away — though moderator review remains the standard for any data that will be used in formal reporting.

Translate between languages. Village supports five languages — English, German, French, Dutch, and Te Reo Maori. The AI assists with translation of organisational content, though human review is recommended for important communications.

Triage member feedback. When a volunteer submits feedback through the platform — a question, a suggestion, a report of something not working — the AI classifies it, investigates where possible, and notifies the member when it has been addressed. This happens automatically, freeing the coordinator from manually sorting every piece of feedback.

What the AI Does Not Do

It does not make decisions for your organisation. When a question involves data interpretation, conservation priorities, or management judgements, the AI stops and routes it to a human. Your moderator, your coordinator, your board — the people your organisation trusts with these decisions.

It does not access content it was not given. Private content stays private. Content from other organisations stays with those organisations. The AI cannot reach across boundaries, because those boundaries are structural, not policy-based.

It does not operate without oversight. Every AI response passes through Guardian Agents — the four mathematical verification layers described in the previous article. No response reaches a volunteer without being checked against your organisation's actual records.

It does not pretend to know things it does not know. When the AI's confidence is low, it says so. Every response carries a confidence indicator. Members can see at a glance whether the AI is drawing on solid records or venturing into less certain territory.

How Bias Is Addressed: The Vocabulary System

One of the subtlest forms of bias in AI is linguistic. When a system trained on corporate data calls your volunteers "users" and your field reports "posts," it is imposing a worldview — one where organisations are consumer platforms and communication is content marketing.

Village addresses this through a vocabulary system that adapts the entire platform to your organisation type.

When you set up a Village for a conservation group, the system does not show you generic labels. It shows you the language of environmental work:

• Volunteers and members, not "users"
• Field reports, not "posts" or "updates"
• Board governance, not "admin settings"
• Project stories, not "content"
• The group, not "the community workspace"

This is not cosmetic. The vocabulary shapes how the AI understands and responds to your organisation. When the AI has been trained with the term "volunteer" rather than "user," it processes questions and generates responses within a conservation frame of reference. It understands that "How do we coordinate our next survey?" is a different question from "How do we schedule our next event?" — even though a generic AI system would treat them identically.

Each community type has its own vocabulary. A parish sees "parishioners" and "parish bulletins." A family sees "family members" and "family stories." The platform is the same, but the language — and therefore the AI's understanding — is specific to your organisation.

How the AI Handles Scientific Data Differently

Conservation groups generate data that sits in a different category from social content. A story about a memorable field day is social content — warm, subjective, meant to build community. A species count from that same day is scientific data — precise, objective, meant to inform decisions.

Village AI is being trained to recognise this distinction. When the AI encounters content that contains quantitative data — counts, measurements, dates, grid references — it applies different standards:

• Numerical values are not rounded or approximated unless explicitly asked
• Qualifications and caveats are preserved in summaries, not smoothed away
• Absence of data is noted, not silently omitted
• Survey effort is reported alongside results where the source records include it

This behaviour is reinforced by the Guardian Agents, which check AI outputs against source records with particular attention to numerical accuracy. If a field report records "probable breeding" and the AI upgrades this to "confirmed breeding," the cross-reference validator flags the discrepancy.

This is an area of active development. The current system handles straightforward cases well — direct questions about recorded data, summaries of clearly structured reports. More complex cases — synthesising trends across multiple seasons, or drawing comparisons between sites with different survey methodologies — require continued refinement. We say this plainly because overpromising on data integrity would be worse than underpromising.

How the AI Learns and Improves

Village AI is not static. It improves over time through three mechanisms:

Scheduled retraining. The AI is periodically retrained on your organisation's latest content. During the beta programme, this happens weekly. New field reports, new stories, new event descriptions — they enter the AI's knowledge base so it stays current with your organisation's work.

Moderator feedback. When a moderator flags an AI response as inaccurate or unhelpful, that correction feeds back into the system. Over time, the AI learns what works for your organisation and what does not. This is not generic improvement — it is improvement specific to your group.

Guardian Agent learning. The fourth Guardian Agent — the adaptive learner — adjusts verification thresholds based on patterns of accuracy and error. If the AI consistently gets a certain type of question right, the guardian eases verification intensity for that type. If it consistently struggles with another type, the guardian tightens scrutiny. The system becomes more efficient without becoming less careful.

What Is Still a Work in Progress

The 8B deep reasoning model is trained and deployed, but the routing system that decides which questions go to the faster model and which go to the deeper model is still being refined. Some questions that would benefit from deeper processing are currently handled by the faster model.

Individual personalisation — where the AI learns individual member preferences — is planned but not yet built. For now, the AI knows your organisation as a group, not your individual volunteers as individuals (unless they interact with it directly).

The moderator accreditation path — structured training for members who take on the moderator role — is designed but being rolled out progressively. During the beta programme, founding organisations have direct access to the founder for support.

Scientific data handling — while the foundations are in place, the system's ability to handle complex ecological datasets, multi-year trend analysis, and cross-site comparisons is still maturing. We expect this to improve significantly through the beta period as real conservation groups provide real feedback on real data.

We mention these plainly because we believe you should know what you are adopting. This is a platform in its early months, built by a small team, used by a small number of organisations. It is functional, it is improving, and it is clear about where it stands.

What This Means for Your Conservation Group

If your organisation is considering Village, here is what you are choosing:

Village is a platform where AI knows your organisation's actual content — your field reports, your stories, your events — not the internet's idea of what a conservation group might be. Every AI response is mathematically verified against your records by independent watchers. The vocabulary reflects your work: volunteers, not users; field reports, not posts; board governance, not admin settings.

Your data stays within your organisation's boundary, is not used to train external AI systems, and can be exported or deleted at any time. The system is transparent about its limitations, improves from your moderators' corrections, and stops to ask a human when a question requires judgement rather than information.

You would also be joining a founding community — one of 20–25 conservation groups, parishes, clubs, families, and businesses shaping the platform during its first year.

If that interests you, applications for the beta programme are open until 30 March 2026.

This is Article 4 of 5 in the "Your Conservation Group, Your AI" series. For the full technical architecture, visit Village AI on Agentic Governance.

Previous: Why Rules and Training Aren't Enough — The Governance Challenge Next: The Village Beyond AI — What Your Conservation Group Actually Gets