Australia must learn to hold AI by the handle, not the blade

Australian Prime Minister Anthony Albanese’s recent speech on AI marked an important shift in Australia’s national conversation. He argued that we cannot stop AI, and that we shouldn’t want to. Our task, he said, is to shape how it develops, and to ensure that it serves Australia’s interests.

He’s right.

This is precisely the debate Australia needs to have. But we won’t resolve it by choosing between uncritical enthusiasm and paralysing fear. We need to learn, collectively, how to hold this technology by the handle rather than by the blade.

Every parent knows that, at some point, they must introduce their children to knives.

Typically, they begin with plastic implements, before graduating to blunt, round-ended butter knives: cutlery that might graze the skin but is unlikely to cut it. Over time, as the child develops both respect for the knife’s potential and skill in using it, the parent introduces sharper tools. The child learns about a knife’s many uses. They also learn that carelessness has its consequences. Even an experienced adult can cut themselves.

So it is with many technologies. They can appear magical and terrifying to the uninitiated and can be genuinely dangerous in untrained hands. But to the thoughtful, appropriately cautious learner, and eventually the competent practitioner, they become invaluable tools.

This is obviously the case with AI.

I have been struck by the fear that AI can evoke, including among highly educated professionals. Some of that fear reflects personal discomfort: AI is disrupting familiar roles, established expertise and assumptions about the value of our work.

But it would be a mistake to dismiss these concerns as simply self-interest or resistance to change. Many people, including those building this technology, have recognised the recurring pattern of technological disruption and are asking a sincere question: is this technology different? Is AI a step too far, a headlong plunge into a dystopian future from which we may not be able to retreat?

The concerns around AI are valid. It is crucial that they be addressed. I believe they can be, albeit not perfectly.

1. AI infrastructure consumes substantial energy and water. While this matters, many other major technologies (e.g. metal smelting, concrete production, air travel) are similarly resource intensive. The answer is not to stop using AI while continuing to accept the environmental costs of other major industries. It is, as the current debate around the world is addressing, to require data centres to underwrite new energy generation and battery storage, use water efficiently and pay the full cost of the infrastructure they require.
2. AI could cause economic, social and regional disruption and deepen inequality.  Previous waves of technological change have had a similar effect, particularly when their consequences were poorly managed. We should recognise that the benefits and costs will likely be unequally shared – including geographically - to an extent that is unacceptable to many communities. Governments, businesses and educational institutions must invest deliberately in skills, workforce transitions and ways to share the productivity gains and mange inequitable costs.
3. AI raises fundamental questions about creative ownership. As has been established by IP law for centuries, creators of content - writers, artists, musicians for example - should retain meaningful ownership and control over how their work is used and how its value is recognised. As an unimagined technology, AI may require that we refresh our well-established IP laws. 
4. AI could diminish the meaning people derive from work. This may be the most existential concern, at least to those of us in professions that are likely to experience the transition most directly. Meaningful work provides people with identity, purpose, and connection. We do not yet know whether AI will diminish those things or allow more people to escape repetitive work and focus on what is distinctly human. The outcome will depend less on the technology itself than on the choices we make about its use. Unlike concern 5 below, with appropriate attention, we can largely address this concern.
5. AI could distort our shared reality. It can produce convincing falsehoods about matters of profound importance, leading even experts to be deceived. Responding to this will require, at the very least, regulation, technical ingenuity, media literacy, and clearer accountabilities regarding who wears the risk of the deception. It will also involve an ongoing contest between deception and detection, as is always the case between bad and good. Nonetheless, on this concern, my confidence is lower than on the others.

These concerns around AI are real. Our choice is whether we approach the knife with respect, pick it up by the handle, and learn to wield it skilfully and safely, or else choose to become “knife illiterate” and become dependent on the countries and companies who choose to do so.

Every nation will need to develop this capability at three levels: individually, organisationally, and societally. National capability is the handle: the infrastructure, institutions, skills and confidence that allow us to use AI productively and on our own terms.

We will not get everything right. A strategy designed to eliminate every possible failure would also eliminate experimentation, learning and much of the potential benefit. Our aim should be to render failure containable, reversible, and instructive, not systemic or catastrophic.

To manage the costs and risks, and sustain the trust of citizens, we must actively shape both the development and application of AI.

Government needs the capacity to understand, buy, evaluate and use AI. Government cannot credibly regulate or evaluate technology that it does not understand. It must become an intelligent purchaser, user and steward of AI.

Traditional procurement processes struggle when model capability and prices can change between the preparation of a tender and the signing of a contract. Procurement teams need faster ways to compare capability, cost, security, portability and performance.

Australia should commission national AI benchmarks and evaluations to support this work. Popular benchmarks and leaderboards usually measure general model performance. They rarely test how well systems operate under Australian laws, language conventions, public-sector processes or service conditions.

A national program could define transparent tests for priority uses. Initial benchmarks could cover emergency response, multilingual service delivery, Australian regulatory compliance, administrative decision support and culturally appropriate services.

Government agencies could then test products against common criteria rather than rely on vendor claims or overseas rankings. Public benchmarks would also give global suppliers and Australian researchers clear performance targets.

This is a practical form of sovereign capability. Australia can define what good AI looks like here without paying to build a frontier model from scratch.

Regulation should follow the consequences of failure. Healthcare, employment, credit, education, essential services and government decisions warrant stronger controls because errors can affect a person’s safety, rights or livelihood.

People should know when AI has materially influenced a decision. They should receive an intelligible explanation, have access to human review and retain the right to appeal. The organisation providing the service must remain accountable. It cannot transfer accountability to a model developer, an algorithm or a disclaimer. 

Lower-risk applications need room for experimentation. Regulatory sandboxes can allow government, businesses and community representatives to test systems together, examine problems and determine whether existing laws remain suitable.

Regulators should also use AI in their own operations. It can triage complaints, find needles in large haystacks of qualitative information, conduct routine checks and give businesses guidance tailored to their circumstances. The objective is many useful trials and few serious failures. Achieving that balance may require changes to laws and processes built on the assumption that every regulatory decision is manual.

Australia should attract AI infrastructure and set firm conditions for its development. We have large areas suitable for industrial development, strong solar and wind resources, high-quality research institutions and a stable legal system. AI infrastructure can create skilled work in construction, engineering, energy and data-centre operations.

New data centres should fund the power generation, firming, grid connections and water infrastructure they require. They should strengthen the energy system rather than raise prices for households and other businesses.

These new data centres should also deliver clear local benefits, with communities involved meaningfully in decisions about siting, resource use and long-term impacts, and with transparent arrangements that ensure economic, social and environmental value is shared rather than extracted.

Hosting servers, however, creates a narrow form of advantage. Australian businesses, researchers, governments and community organisations need affordable access to computing power. They also need secure data-sharing arrangements, trusted digital identity and standards that allow systems to work together.

Australia does not need to produce every chip, data centre and model itself. Sovereign capability means retaining options: the ability to choose between providers, move workloads, host sensitive systems locally and keep essential services operating when an overseas provider fails.

Health, aged care, disability services, education, public administration and public safety face rising demand and persistent workforce shortages. They depend on human relationships. AI is not likely to fully replace a teacher’s judgement, a clinician’s expertise, or a carer’s empathy, but it can complement them. It can reduce the work that takes them away from people by summarising case histories, preparing draft documentation, translating service information, retrieving relevant evidence and identifying cases that require closer review.

Buying software licences will not achieve this. Governments and service providers need to redesign workflows around what people and AI each do best, involve frontline staff and service users, establish baseline measures, run controlled trials and publish the results.

Education may be the hardest part of the response. AI can recall information, draft text and perform routine analysis. Assessments based mainly on those tasks no longer provide reliable evidence of a student’s capability.

Schools, universities and accrediting bodies need to place more weight on judgement, problem definition, ethical reasoning, practical application and decisions under uncertainty. Assessment should include authentic projects, oral defences and demonstrations of applied skill. Students should explain how they reached a conclusion, which sources and tools they used, and why their answer deserves confidence.

Education must also guard against cognitive atrophy. AI should extend a student’s thinking rather than allow the student to avoid it. Teachers need clear guidance and evidence about when AI supports learning and when it weakens the development of knowledge and reasoning.

One principle should apply across education and work: we can delegate tasks to AI, but we cannot delegate accountability. Anyone who submits AI-assisted work should understand it, be able to defend it, explain it and be prepared to stand behind it.

Australia is making enormous investments in data centres, chips and energy generation. We now need comparable ambition in the enabling institutions around them: infrastructure, standards, proportionate regulation, education, workforce transition, sovereign capability, and the capacity and data holdings of government. Australia does not need to predict the future perfectly, nor build the world’s most advanced model. We need the capability to choose, test, adapt, and act on our own terms.

The Prime Minister’s speech provides a welcome statement of intent. The harder work is now to turn that intent into institutions, no-regrets investments, and practical action. Guardrails tell us where the blade is. Infrastructure, institutions, skills and standards give us a grip. That is the handle Australia must create and the knife it must learn to wield. 

Get in touch to discuss how your organisation can build the capabilities, confidence, and safeguards required to use AI responsibly and on its own terms.

Connect with Tim Orton on LinkedIn.