What AI Actually Is (and What It Isn't)

Series: To Hapori, To AI — Digital Sovereignty for Indigenous Communities (Article 1 of 5) Author: My Digital Sovereignty Ltd Date: June 2026 Licence: CC BY 4.0 International

AI Changed While You Were Watching

You have probably heard people say that artificial intelligence is going to change everything. You may also have heard people say it is just a fad, or that it cannot do anything truly new. Both of these positions miss the point, and understanding why will help your community make better decisions about this technology.

But there is something else worth noticing first: AI has changed even in the short time people have been arguing about it.

A year ago, when most people said "AI," they meant a chatbot — a window you typed a question into and got an answer back. You asked, it answered. That was the whole transaction. Today, the centre of gravity has moved. The systems attracting the most attention and the most investment are no longer just chatbots that answer. They are agents that act — fill in the form, send the message, browse the websites, submit the records, write and run the code.

This is the single most important shift to understand, and the rest of this article builds on it. To make sense of AI today, you need to hold two ideas apart:

• The engine — the underlying model, the thing that produces language.
• The agent — the engine put to work, wired up so it can take actions in the world on your behalf.

The engine has been getting more capable. But the bigger change — and, for indigenous communities, the more serious one — is what people are now building around the engine. Let us take each in turn.

The Engine: A Machine That Predicts

Here is the plainest description of what the engine does: it predicts what word should come next.

When you type a message into a chatbot, the system is not thinking about your question the way you or your kaumatua would think about it. It is doing something much more mechanical. It has been shown billions of pages of text — books, websites, conversations, legal documents, recipes, medical papers, arguments on social media — and from all of that reading, it has learned patterns. When you ask it a question, it generates an answer by predicting, one word at a time, what a plausible response looks like based on everything it has seen before.

This is genuinely useful. A system that has absorbed the patterns of billions of pages of text can help you draft a letter, summarise a long document, answer a factual question, or suggest how to word a difficult announcement. These are real capabilities, and they save real time.

But at its core, the engine is doing pattern matching at an extraordinary scale. That single fact explains both what it is astonishingly good at and where it quietly goes wrong — a point this series returns to, because for a community whose knowledge is underrepresented in that training data, the way it goes wrong is not random. It leans Western.

Can the Engine Reason?

There is a deeper question that researchers are actively investigating, and the plain answer is: we do not yet know.

When early AI systems produced fluent text, it was reasonable to describe them as sophisticated pattern-matching and leave it there. But a newer generation of engines — often called "reasoning" or "thinking" models — does something different. Instead of answering immediately, it works through a problem in steps, producing a visible chain of intermediate thinking before it commits to an answer. Given harder problems, it spends longer. The results can be remarkable: in 2025, reasoning systems from more than one major laboratory solved problems from the International Mathematical Olympiad — one of the hardest mathematics competitions in the world — at a level equivalent to a human gold medallist.

So is that reasoning, or is it very sophisticated pattern-matching dressed up to look like reasoning?

The research is genuinely unsettled, and serious people disagree. One influential 2025 study argued these systems are partly an "illusion of thinking" — that they collapse on certain puzzles in ways a real reasoner would not. Several equally serious replies argued the opposite. The most careful current verdict is that today's reasoning models are neither true reasoners nor mere parrots — they are something genuinely new that we do not yet fully understand. Anyone who tells you AI definitely can or definitely cannot reason is overstating what the evidence supports.

One finding does matter for your community, and it is easy to misread, so read it carefully. When these systems show you their "thinking," that visible chain does not reliably reflect what actually drove the answer. Researchers have repeatedly found that a model's stated reasoning can leave out the real influences on its conclusion — not because the machine is being dishonest in any human sense (it has no intentions), but because the words it shows you are themselves just more predicted text, not a true readout of an inner process. The practical consequence: you cannot keep an AI accountable simply by reading the explanation it offers for itself. That is one reason community-controlled governance — which checks the output against your own records rather than trusting the AI's self-report — matters so much. We return to it in Article 3.

What we can say is this: the trajectory is steep. A few years ago these systems could barely string a coherent paragraph together. Today they write essays, pass professional examinations, generate working computer code, and increasingly act on the world rather than just describing it. The next few years will bring greater capability again.

"AI Can't Do Anything New" — It Depends What You Mean by New

People who dismiss AI by saying it cannot create anything original are making a claim that is narrowly true and broadly misleading.

An engine cannot originate from experience. It has never sat at a tangi. It has never felt the weight of speaking on behalf of a whanau. It cannot understand why the karanga matters at the gate of a marae — it can only reproduce patterns that statistically resemble understanding. In that sense, everything it produces is a recombination of material it absorbed during training.

But consider what "recombination" actually means at this scale. No single human being has read every piece of Treaty settlement documentation, every report from the Waitangi Tribunal, every piece of legislation on indigenous rights across the Commonwealth, every academic paper on indigenous data sovereignty, and every community newsletter from the last hundred years. When the AI draws a connection between polycentric governance theory and traditional Maori decision-making structures, that connection is genuinely new to any individual human, even though both ideas existed separately.

So "AI can't do anything new" is true at the level of origination and false at the level of synthesis. Both things matter, and serious engagement with this technology requires holding both.

From Answering to Acting: The Agent

Now to the change that matters most for your community.

For most of the chatbot era, the worst an AI could do directly was give you a bad answer. The harm only landed if a person acted on it — sent the wrong words, trusted the wrong figure, passed on the flawed advice. There was always a person between the machine and the consequence.

An agent removes that person from the loop, by design.

An AI agent is an engine wrapped in what researchers call "scaffolding" — a memory so it can keep track of a task, access to a web browser, the ability to use software tools and other programs, and a goal you give it in plain language. With that scaffolding, the system can pursue the goal across many steps with much less supervision: it can search, decide, act, check the result, and act again. A chatbot answers. An agent acts.

This is why AI suddenly feels different even though the underlying engines did not change overnight. The new thing is largely the wrapper. The industry separates the two deliberately: the engine provides the raw capability, and the scaffolding turns that capability into something that does work in the world. Much of the recent leap in what AI can do — as opposed to what it can say — comes from better scaffolding, not a new kind of mind.

The well-known agent products of 2025 and 2026 — the ones that browse the web for you, operate a computer, or write and run software — are almost all built by large American technology companies, and we will look at what that means for your data in the next article. For now, hold the concept: the question is no longer only "what will the AI tell me?" It is "what will the AI do, and can it be stopped in time if it goes wrong?"

For indigenous communities this carries a particular weight. When a system acts on its own there are fewer chances to intervene, and some actions cannot be undone. Knowledge, once shared, cannot be un-shared; korero submitted to an outside system, or a taonga of mātauranga passed to a platform that then acts on it, cannot be recalled. An agent that shares, publishes, or submits on its own — without a kaitiaki in the loop — is not a neutral convenience. It is the digital continuation of a very old pattern: knowledge leaving the community's control without the community's say. And if something goes wrong, it is genuinely hard to say who was responsible — the person who set a goal in a sentence, or the company whose system chose the steps. None of this means agents are useless. It means the stakes of whose agent you use, and who holds authority over what it does, just rose sharply.

The Real Issue: Whose Patterns, and Whose Hands on the Controls?

Here is where it gets practical for your community.

When a large AI engine is trained on the internet, it absorbs the internet's biases, assumptions, and cultural defaults. The internet is overwhelmingly English-language, Western, commercially oriented, and shaped by the values of a handful of technology companies. This is not a conspiracy — it is simply what happens when you train a system on data that disproportionately represents one culture and one set of priorities.

For indigenous communities, this bias is not subtle. It is structural. The internet over-represents written, Western, individualised knowledge. It under-represents oral traditions, collective decision-making, relational knowledge systems, and the forms of understanding that indigenous peoples have carried for generations. When an AI is trained on this data, it does not merely lack indigenous knowledge — it is structurally weighted against it. The patterns it has learned treat Western frameworks as the default and everything else as the exception.

When a whanau member asks an AI for advice about a difficult family situation, the system defaults to the language of individual therapy — assertiveness training, boundary-setting, self-care — because that is what dominates its training data. It does not reach for whanaungatanga (kinship obligation), manaakitanga (care for others), or the long view that comes from knowing your obligations extend across generations. When a community leader asks it to help with a sensitive communication, it defaults to corporate language, because business correspondence vastly outnumbers indigenous community correspondence in what it learned from.

The system is not hostile to your knowledge. It simply does not know your knowledge. It knows what is statistically common, and what is statistically common is overwhelmingly Western.

In the chatbot era, that bias shaped the words you read — a mihi that read like grief counselling. In the agent era, the same bias shapes the actions taken on your behalf. An agent that does not understand tikanga will not just describe your world poorly; it may act in your name, in ways that breach protocol, before anyone can stop it. So the real issue with AI now has two halves: whose patterns does it carry, and whose hands are on the controls when it acts?

Why This Matters Now

No one knows with certainty what happens if an AI system ever develops something resembling its own intent — purposes and priorities that may not align with ours. We are likely still some distance from that threshold. But the architecture we build now, the habits of governance we establish today, will determine whether we are prepared when that moment arrives or whether we discover too late that we handed over control without noticing.

This is not science fiction. It is a straightforward observation about institutional preparedness. Your community has its own governance structures — its own tikanga for how decisions are made, how authority is exercised, how knowledge is shared. These exist not because every hui descends into chaos, but because governance structures need to be in place before they are needed, not after.

The same principle applies to AI — and it applies with more urgency now that AI does not merely advise but acts.

Two Paths Forward

There are two ways a community can engage with AI.

The first path is to use Big Tech AI — the chatbots and, increasingly, the agents built by the largest American technology companies. These are powerful, convenient, and often free or cheap. But they come with conditions. Your data flows to their servers. Your conversations — and now your agent's actions — pass through systems you do not control. The AI's behaviour is governed by the company's policies, which can change without your consent. And the patterns the AI carries are set by its training data, which you have no influence over. For indigenous communities that have spent generations fighting for sovereignty over their own knowledge, this is not a neutral trade-off.

The second path is to use AI that your community controls. A more focused system, trained on your content, running on infrastructure you control, governed by rules your community sets. A system that knows the difference between a community announcement and a corporate blog post, because your community taught it. A system whose responses are checked against your actual records by independent watchers that operate separately from the AI itself — and whose ability to act is deliberately bounded, so a person from your community is always able to step in before anything leaves your boundary.

This is what Village AI is. It is not the most powerful AI system available. It is designed to be faithful to your community — to your content, your values, and your governance. For indigenous communities, that faithfulness includes the ability to define your own vocabulary, your own governance boundaries, and your own rules about how knowledge is shared — and to keep authority over what an AI does in your name.

The next article in this series explains how Village AI is structurally different from Big Tech AI, and why that difference matters — particularly for communities whose knowledge systems have already survived one wave of colonisation and should not have to survive another.

This is Article 1 of 5 in the "To Hapori, To AI" series. For the full technical architecture, visit Village AI — Agentic Governance.

Next: Big Tech AI vs. Your Community's AI — Why the Difference Matters