The Office of AI

The Australian government loves to innovate. As it did with its "world-leading" tobacco control and social media ban (both with predictably bad results), it's now taking the lead in regulating AI through the creation of a federal "Office of AI" to sit within the Department of the Prime Minister and Cabinet.

The new Office is going to "co-ordinate the design of our new Australian Standards", which are basically the "energy, water, safety and other requirements" that every Australian and business operating in Australia will have to obey to use AI. The details of exactly what that means are fairly scant; all we know is that from "early next year", data centres specifically will have a legal obligation to:

"...underwrite their own new power supply, pay their full share of connection costs so energy bills are not impacted, reduce power when needed to strengthen the grid, and be as water efficient as possible."

That's not the worst list of requirements, except for the vague grid-related one. Data centres maintain a high baseline of constant power to ensure 24/7 uptime (making them the perfect customer for an always-on baseload source like coal or nuclear), and can be incentivised to reduce their load during grid peak periods through prices. If Albo understood even basic economics, he wouldn't be talking about needing a heavy-handed mandate; just let market forces work!

But the real devil will be in the details. Maybe this will all amount to nothing; a vague AI framework dressed up as legal "standards" that doesn't really do much at all. However, it could also go the other way and be written to explicitly require AI companies to do things like fund green energy infrastructure they don't need, allowing the government to socialise the cost of its own energy transition while pretending to regulate AI.

There are plenty of other problematic phrases in Albo's speech, too. For example, that data centres should not be allowed to "compete with housing", which is a recipe for giving NIMBYs a federal veto over any large-scale AI infrastructure projects. Similarly, the statement that AI must "support and create good jobs, not replace them", which is one bad regulation away from effectively giving licence to unions and large firms to weaponise the AI regulatory process, thwarting innovation and productivity growth.

What really concerns me is the overarching tone: how Albo wants to "set the terms" and "determine AI's social licence", all before anyone really knows what the technology is even capable of doing! That just reeks of pre-emptive dirigisme, and represents a profound lack of faith in the dispersed, evolutionary process of market discovery, along with a dangerous overconfidence in the ability of policymakers to design complex systems from the top down.

But the fundamental error in Albo's thinking is the belief that a central office can design the incentives, guardrails, and institutions for a general-purpose technology like AI. Institutions work best when they emerge bottom-up from common law, contract, and property rights. Fix the tax code (properly this time, please), streamline planning laws, enforce property rights, reduce energy costs, and then just get out of the way.

Is it really too much to ask for Australia's kakistocratic political class to just learn their place in the world, take a deep breath, sit back, and observe how the countries actually building AI decide to tackle it before boldly regulating where no one has regulated before? It's not like there's a shortage of actual problems that could use some political attention in Australia, rather than imaginary potential problems that may or may not materialise.