\u201cThis is a gamechanger\u201d.<\/p>\n
That\u2019s how Paul Scully, New South Wales minister for planning and public spaces, described<\/a> the state government\u2019s launch of a tender for an artificial intelligence (AI) solution to the housing crisis earlier this month.<\/p>\n The system, which is aimed at cutting red tape and getting more homes built fast, is expected to be functioning by the end of 2025.<\/p>\n It is \u201callowing construction to get under way and new keys into new doors,\u201d Scully added.<\/p>\n The announcement was later endorsed<\/a> by the federal treasurer, Jim Chalmers, as a model for other states and territories to replicate, to \u201cunlock more housing\u201d and \u201cboost productivity across the economy\u201d.<\/p>\n Speeding up building approvals is a key concern of the so-called abundance agenda<\/a> for boosting economic growth.<\/p>\n Those wheels are already in motion elsewhere in Australia. Tasmania is developing an AI policy<\/a>, and South Australia is trialling a small-scale pilot<\/a> for specific dwelling applications to allow users to submit digital architectural drawings to be automatically assessed against prescribed criteria.<\/p>\n But will AI really be a quick fix to Australia\u2019s housing crisis?<\/p>\n Cutting red tape<\/p>\n Housing and AI were both key themes at last month\u2019s productivity roundtable.<\/p>\n In a joint media release<\/a>, the federal minister for housing, Clare O\u2019Neil, and minister for the environment and water, Murray Watt, said easing \u201cthe regulatory burden for builders\u201d is what Australia needs.<\/p>\n They point to the backlog of 26,000 homes currently stuck in assessment under environmental protection laws as a clear choke point. And AI is going to be used to \u201csimplify and speed up assessments and approvals\u201d.<\/p>\n None of this, however, explains AI\u2019s precise role within the complex machinery of the planning system, leaving much to speculation.<\/p>\n Will the role of AI be limited to checking applications for completeness and classifying and validating documents, as Victorian councils<\/a> are already exploring? Or drafting written elements of assessments, as is already the case in the Australian Capital Territory<\/a>?<\/p>\n Or will it go further? Will AI agents, for example, have some autonomy in parts of the assessment process? If so, where exactly will this be? How will it be integrated into existing infrastructure? And most importantly, to what extent will expert judgment be displaced?<\/p>\n A tempting quick fix<\/p>\n Presenting AI as a quick fix for Australia\u2019s housing shortage might be tempting. But it risks distracting from deeper systemic issues<\/a> such as labour market bottlenecks, financial and tax incentives, and shrinking social and affordable housing.<\/p>\n The technology is also quietly reshaping the planning system \u2013 and the role of planners within it \u2013 with serious consequences.<\/p>\n Planning is not just paperwork waiting to be automated. It is judgment exercised in site visits, in listening to stakeholders, and in weighing local context against the broader one.<\/p>\n Stripping that away can make both the system and the people brittle, displacing planners\u2019 expertise and blurring responsibility when things go wrong. And when errors involving AI happen, it can be very hard to trace them, with research showing explainability<\/a> has been the technology\u2019s achilles heel.<\/p>\n The NSW government suggests<\/a> putting a human in charge of the final decision is enough to solve these concerns.<\/p>\n But the machine doesn\u2019t just sit quietly in the corner waiting for the approve button to be pressed. It nudges. It frames. It shapes what gets seen and what gets ignored in different stages of assessment, often in ways that aren\u2019t obvious at all.<\/p>\n For example, highlighting some ecological risks over others can simply tilt an assessor\u2019s briefing, even when local communities might have entirely different concerns. Or when AI ranks one assessment pathway as the \u201cbest fit\u201d based on patterns buried in its training data, the assessor may simply drift toward that option, not realising the scope and direction of their choices have already been narrowed.<\/p>\n Lessons from Robodebt<\/p>\n Centrelink\u2019s online compliance intervention program \u2013 more commonly known as Robodebt \u2013 carries some important lessons here. Sold as a way to make debt recovery more \u201cefficient\u201d, it soon collapsed into a $4.7bn fiasco<\/a>.<\/p>\n In that case, an automated spreadsheet \u2013 not even AI \u2013 harmed thousands of people, triggered a hefty class action and shattered public trust in the government.<\/p>\n If governments now see AI as a tool to reform planning and assessments, they shouldn\u2019t rush in headlong.<\/p>\n The fear of missing out may be real. But the wiser move is to pause and ask first: what problem are we actually trying to solve with AI, and does everyone even agree it\u2019s the real problem?<\/p>\n Only then comes the harder question of how to do it responsibly, without stumbling into the same avoidable consequences as Robodebt.<\/p>\n Responsible innovation offers a roadmap forward<\/p>\n Responsible innovation<\/a> means anticipating risks and unintended consequences early on \u2013 by including and deliberating with those who will use and be affected by the system, proactively looking for the blind spots, and being responsive to the impacts.<\/p>\n There are abundant research case studies, tools and frameworks in the field of responsible innovation that can guide the design, development and deployment of AI systems in planning. But the key is to engage with root causes and unintended consequences<\/a>, and to question the underlying assumptions about the vision and purpose of the AI system.<\/p>\n We can\u2019t afford to ignore the basics of responsible innovation. Otherwise, this so-called \u201cgamechanger\u201d to the housing crisis might find itself sitting alongside Robodebt as yet another cautionary tale of how innovations sold as efficiency gains can so go wrong.<\/p>\n