Essays

Comments on the National AI Plan 

14 December 2025

Following are my personal thoughts on the National AI Plan [1]. They focus on three related areas, which I see as gaps in the Plan.

1. The Plan does not attempt to define the term "Artificial Intelligence". It instead assumes that the reader knows what "AI" means. This is a missed opportunity. Providing a definition, even acknowledging that a strict definition is difficult, provides a scoping opportunity. It provides an opportunity to state which aspects or interpretations of AI are in scope or out of scope for this Plan, and most importantly, to state reasons for this scoping in the context of this Plan. It also provides an opportunity to state what the Australian Government is doing to address those aspects and interpretations of AI that are considered to be out of scope for this Plan.

2. The plan appears to focus on large language models and generative AI, and does not explicitly mention other approaches. These other approaches are not obscure. In fact they are fast becoming mainstream. These include embodied intelligence, proof assistance, and non-binary computation. Because the Plan does not specifically address these other approaches, it misses the opportunity to address the societal challenges and opportunities they pose.
   By "embodied intelligence" I mean a wide scope of related approaches, including Internet of things, swarms, autonomous vehicles, and robotics in general. The Plan has missed the opportunity to envision the impact that autonomous agents interacting with the environment will have on agency and safety of the general public, and the challenges that this poses to laws, commerce and education [2]. See Section 3 of [3] and Section 3.2 of [4] for deeper discussions on the role of embodiment in artificial intelligence.
   Increasingly, AI systems are being combined with proof assistants to tackle benchmarks in mathematical reasoning [5]. Examples include Goedel-Prover [6], DeepSeek-Prover [7], AlphaProof [8], and HERMES [9].
   The application of non-binary computation to artificial intelligence is an active field of research. Non-binary computation includes neuromorphic computing [10,11,12] and quantum computing [13].

3. While the Plan addresses data centre capacity, and AI research and training capabilities, it does not explicitly deal with Australia's existing publicly funded high performance computing capability, nor does it explicitly address the expertise and resources needed to conduct fundamental research into AI, and especially into alternative approaches to AI. These two omissions miss the opportunity to explore what is meant by sovereign AI capability, and what resources may be needed and available to Australia to achieve such capability [14].
   Action 1 in the Plan focuses on data centre investment, giving the examples of CDC Data Centres, Firmus, Amazon and Microsoft, but it does not mention Australia's national computing infrastructure organizations such as National Computational Infrastructure, and the Pawsey Supercomputing Research Centre [15]. Both types of infrastructure are needed.
   Action 5 in the Plan focuses on skills, but does not explicitly mention the skills involved in research into the fundamental principles, software and hardware that is needed to support AI. The focus instead appears to be on the skills needed for the adoption and application of AI. Both types of skills are needed [16,17].

References

[1] Australian Government Department of Industry, Science and Resources (2025). National AI Plan.

[2] Vellante, D. (2025) A Strategic Analysis of the Future of AI and Robotics: From Industrial Efficiency to Embodied Intelligence, theCUBE Research.

[3] Mumuni, A., & Mumuni, F. (2025). Large language models for artificial general intelligence (AGI): A survey of foundational principles and approaches. arXiv preprint arXiv:2501.03151.

[4] Zilberman, J., & Tamir, B. (2025). AGI imagined: how is AGI configured by the theories of mind. AI & SOCIETY.

[5] Yang, K., Poesia, G., He, J., Li, W., Lauter, K., Chaudhuri, S., & Song, D. (2024). Formal Mathematical Reasoning: A New Frontier in AI. arXiv preprint arXiv:2412.16075.

[6] Lin, Y., Tang, S., Lyu, B., Yang, Z., Chung, J. H., Zhao, H., ... & Jin, C. (2025). Goedel-Prover-V2: Scaling Formal Theorem Proving with Scaffolded Data Synthesis and Self-Correction. arXiv preprint arXiv:2508.03613.

[7] Ren, Z. Z., Shao, Z., Song, J., Xin, H., Wang, H., Zhao, W., ... & Ruan, C. (2025). DeepSeek-Prover-V2: Advancing Formal Mathematical Reasoning via Reinforcement Learning for Subgoal Decomposition. arXiv preprint arXiv:2504.21801.

[8] Hubert, T., Mehta, R., Sartran, L., Horváth, M. Z., Žužić, G., Wieser, E., ... & Silver, D. (2025). Olympiad-level formal mathematical reasoning with reinforcement learning. Nature.

[9] Ospanov, A., Feng, Z., Sun, J., Bai, H., Shen, X., & Farnia, F. (2025). HERMES: Towards Efficient and Verifiable Mathematical Reasoning in LLMs. arXiv preprint arXiv:2511.18760.

[10] Kudithipudi, D., Schuman, C., Vineyard, C. M., Pandit, T., Merkel, C., Kubendran, R., ... & Furber, S. (2025). Neuromorphic computing at scale. Nature, 637, 8047.

[11] Muir, D. R., & Sheik, S. (2025). The road to commercial success for neuromorphic technologies. Nature Communications, 16(1).

[12] Borra, R., Neuromorphic Computing: Bridging Biological Intelligence and Artificial Intelligence (2024). International Journal of Engineering and Advanced Technology, 14(2).

[13] García Pineda, V., Valencia-Arias, A., López Giraldo, F. E., Zapata-Ochoa, E. A. (2026) Integrating artificial intelligence and quantum computing: A systematic literature review of features and applications, International Journal of Cognitive Computing in Engineering, 7.

[14] Australian Academy of Technological Sciences and Engineering (ATSE). (2024). Made in Australia: Our AI opportunity.

[15] National Computational Infrastructure. (2025). Record Demand Highlights Australia's Growing Need for Supercomputing Power. NCI News.

[16] Reynolds, M. (2018). Training the Next Generation of AI Researchers. Australian Council of Learned Academies (ACOLA).

[17] Panchanathan, S. (2024). Envisioning the future of the AI research ecosystem. PNAS Nexus, 3(2).