> It would appear that Australia's AI strategy is based upon existing structured data. That was appropriate fuel twenty years ago. Modern AI runs on something entirely different — and we're not creating it. ![[Gemini_Generated_Image_581xc7581xc7581x.png]] There is a deep irony at the heart of Australia's AI data strategy. Twenty years ago, AI ran on petrol. Structured data — clean, formatted, queryable — had to be extracted, refined, and processed before it was usable. It required industrial infrastructure: data engineers, ETL pipelines, warehouses, schemas. Like petroleum, it was expensive to produce, tightly controlled, and available only to institutions with the refining capacity to handle it. The strategy Australia is now pursuing was designed for that era. But the engine changed. Modern AI runs on electricity. LLMs derive much of their capability not from databases, but from large corpora of narrative and relational text — web pages, books, Wikipedia, documentation, the accumulated knowledge of human civilisation encoded in language. That is precisely what makes them powerful. They can read a wiki page about a CRC spinout and understand the relationships, the chronology, the significance. And electricity can come from any source. A government data warehouse. A university repository. A MediaWiki instance built by a retired researcher in Brisbane with no institutional funding. The grid does not care about the scale of the generator. It cares about connection. So the question becomes: if we are transitioning to engines driven by electricity, where does the electricity come from? ## What the Report Gets Right The Australian Government's response to the [Senate Select Committee on Adopting Artificial Intelligence](https://www.industry.gov.au/publications/australian-government-response-senate-select-committee-adopting-artificial-intelligence-ai-report) is not a bad document. It is, in places, an honest one. It acknowledges something that has taken years to get anyone in Canberra to say plainly: *"Data is a strategic national asset and critical driver of modern economies."* Good. That is correct. We are making progress. The response goes further. It commits to unlocking high value datasets for pilot AI use cases. It references consistent data standards, metadata frameworks, trusted data sharing, non-sensitive datasets. It speaks of building locally relevant AI applications and models. The language is careful, considered, and shaped by a single unstated assumption: That data means structured data. Rows and columns. Government registries. Private sector transaction records. Sensor readings. Clinical trials. The kind of data that lives in a database, carries a schema, and can be queried with SQL. ## What the Report Cannot See Nowhere in the Senate response — not once — is there a discussion of *creating* data. The entire strategy is about unlocking what already exists. Open it up. Make it accessible. Build the pipes. This is not new. The [2022 Australian Data Strategy](https://www.finance.gov.au/sites/default/files/2022-10/australian-data-strategy.pdf)— the foundational document that preceded everything since — committed to developing "the right infrastructure to enable the better use of data and underpin the creation of new data assets." The word *creation* appeared. And then vanished. Every strategy document since — the [2023 Data and Digital Government Strategy](https://www.dataanddigital.gov.au/strategy) , the [2025 National AI Plan,](https://www.industry.gov.au/publications/national-ai-plan) the 2026 Senate response — has recycled the same framework: access, sharing, unlocking, governance. Creation dropped out of the conversation before it ever entered it. The data that would allow an Australian AI to understand Australia — to know what this country has built, who built it, who funded it, what it became, and what we allowed to disappear — does not live in a government database. It is not in a private sector data lake. It is not waiting to be unlocked. It does not exist in one place. In many cases, it does not exist at all. The causal chains of Australian innovation — the lineage from a research grant to a spinout to a commercial outcome to an acquisition — are scattered across web pages that disappear when departments restructure, PDFs that are never indexed, reports that are commissioned and then quietly buried, and the memories of researchers who are retiring. This is not structured data. It is narrative knowledge. Relational knowledge. Provenance. ## Who Is This AI For? The Senate report frames AI adoption as an economic opportunity for Australian industry. Business intelligence. Market analysis. Procurement. R&D partnership. Supply chain optimisation. That is the right ambition. Australian SMEs using AI to compete globally — yes. That is what we want. But now ask the question nobody in the report asks: if Australian industry uses AI tools running on American and European fuel, what happens? The AI will do exactly what it is trained to do. An engine connected primarily to foreign generation draws power shaped by foreign priorities. It will surface American suppliers. European researchers. Offshore solutions. Not through malice. Not through conspiracy. Simply through absence. Australian capability is not in the training data, so it does not appear in the results. Australian researchers are invisible. Australian supply chains are invisible. Australian innovations — the ones that were commercialised, the ones that were acquired, the ones that are sitting in university labs waiting for a partner — invisible. We will have handed Australian SMEs a powerful, well-funded, government-endorsed tool that systematically connects them to everywhere except here. An AI that does not know Australia built Zebedee, or AutoHaul, or Hovermap, or the molecular clamp vaccine technology, or forty years of field robotics capability — that AI cannot connect an Australian manufacturer to an Australian solution. It will find an American one instead. And we will call this AI adoption a success. ## The Proof I have been making this argument for years. I have made it in submissions, in articles, in conversations with people who nod and then return to their spreadsheets. So I stopped arguing and started building. Three years ago I created iwiki.au — the [Innovation Wiki of Australia.](https://iwiki.au) Not a database. Not a structured dataset. A knowledge graph. A MediaWiki instance mapping the relationships between historical precedent, policy decisions, funding programs, research activity, spinouts, and commercial outcomes. It does not replicate source material. It traces the connections. It preserves provenance, including time-stamped references to sources that have since disappeared. It is the kind of infrastructure that should have been built institutionally, decades ago. I built it for the common good. I also built it for selfish reasons — because my own innovations had been erased from the institutional record, and I was not prepared to let that happen to the broader ecosystem. And I built it deliberately as a specialised knowledge graph — giving LLMs the relational context and provenance they need to anchor their reasoning without hallucinating. It worked. This month, nearly 300,000 unique visitors. LLMs are now citing iwiki.au. When asked who invented Zebedee — the handheld 3D mobile mapping system that won the Eureka Prize, was listed as one of CSIRO's most significant innovations, and was acquired for $70 million — ChatGPT found the answer not from CSIRO, not from IEEE, not from any government source. From iwiki.au. I asked ChatGPT and it described it as a specialised knowledge graph for Australian robotics and CSIRO history — a corpus that makes Australian innovation history retrievable by machines. One person. Three years. No government funding. No institutional support. ## The Indictment The Australian Government's AI data strategy will unlock existing datasets. It will build metadata frameworks. It will establish data sharing protocols. It will do all of this competently and carefully and it will not solve the problem. Because the problem is not access. The problem is existence. The knowledge infrastructure that would allow an Australian AI to understand Australia — to map its innovation ecosystem, trace its causal chains, surface its capabilities to its own industries — does not exist at the scale required. What exists are fragments. Siloed repositories. Disappeared web pages. Archived PDFs. ...and one wiki built by a researcher (myself) who got angry enough to do something about it. Institutional incentives favour measurable, structured assets. But Wikipedia is not noise. It is one of the most powerful generators on the grid. The connective tissue of human knowledge — relational, narrative, cross-referenced — is what allows an AI to *understand* rather than merely *retrieve*. Australia needs that connective grid, built for Australia, about Australia, by Australians. ## The Policy Contradiction The AI data strategy is not the only place this gap appears. The [Ambitious Australia: Strategic Examination of Research and Development final report](https://www.industry.gov.au/publications/ambitious-australia-strategic-examination-research-and-development-final-report), has recommendation 20, which calls on the National Innovation Council to create a national narrative that persistently demonstrates the benefits of R&D&I to the community. A national innovation narrative. Built how, exactly? From what? You cannot narrate without stories. You cannot have stories without people and projects. And you cannot find the people and projects without the infrastructure to collect, preserve, and connect them. The SERD is calling for power. The AI data strategy is still calling for building petrol stations. And nobody in either document asks where the electricity comes from. This is the same blind spot in two different documents. The AI strategy wants to train locally relevant models on Australian data. The SERD wants to build a national innovation narrative. Both require the same thing: a systematic, persistent, funded effort to document what Australia has built, who built it, and what happened next. Neither document funds it. Neither document mentions it. Neither document appears to know it is missing. ## The Ask This is not a criticism of the people working on Australia's AI strategy. It is a criticism of the frame. The current energy crisis has forced a choice: build more oil refineries in Australia, or accelerate electrification with Australian generation. The answer is clear. Sovereignty comes from generating your own power, not from refining someone else's oil more efficiently. The same choice faces Australian AI. We can double down on structured data pipelines fed by offshore sources — more refined, more accessible, more interoperable. Or we can build the narrative knowledge infrastructure that generates distinctly Australian electricity for the grid. An AI running on American and European narrative data is no more sovereign than a grid running on imported LNG. True sovereign AI requires knowledge infrastructure that reflects Australian capabilities, institutions and experience. Stop building refineries. Start building Australian energy sources for the grid. The Australian Government subsidises households to install solar panels and home batteries. It does this because it understands that a distributed energy grid — many small generators connected to a shared network — is more resilient, more equitable, and more sustainable than a centralised one. The same logic applies to knowledge infrastructure. If Australia subsidises distributed energy generation because it strengthens the grid, why not subsidise distributed knowledge generation for exactly the same reason? The ask: * Expand the definition of strategic data to include narrative knowledge infrastructure: the structured preservation of what Australia has built, who built it, and what happened next. * Fund its creation, not just its access. * Treat distributed knowledge generation as national infrastructure, not archival activity. ## Related Articles [[What Vehicle is the Australian Innovation Ecosystem?]] - I used the metaphor of a vehicle to describe Australia's innovation system. At the time, I identified funding as the fuel. I now believe that was incomplete. It is also fuelled by knowledge. [[The SERD Recommendation Nobody Talks About]] — comments on Recommendation 20 and the missing narrative infrastructure. [[National Data Strategy]] - The Case for Free and Open Collaborative Networks — on the platform business model for knowledge networks, and why they need to be subsidised to survive. ## Postscript In an earlier post [[2026-05-26 My second attempt at EgoPrompting]], I noted that my website — [The Innovation Wiki of Australia ](https://iwiki.au)— had been accessed more than 200,000 times by Chinese bots. On one interpretation, this could be seen as simple scraping or even IP theft. But there is another, more strategic reading: if this material is being used as training data for AI systems (as is widely understood to be the case), then those systems will inevitably develop a deeper understanding of Australia — its innovation ecosystem, institutions, and priorities. In that sense, data flows become a form of soft power. Just as international students studying in Australia shape relationships, understanding, and influence over time, large-scale exposure to Australian public knowledge may also shape how external AI systems interpret and respond to Australia. The question is whether we are comfortable with that asymmetry — where others’ AI systems are trained on our knowledge, while we have limited reciprocal access or control over how that knowledge is incorporated, weighted, and used. Are we unintentionally biasing external AI systems to better understand us, without building equivalent sovereign capability in return?