Researchers: Australia's 'off switch' for the algorithm is the wrong fix

A welcome reform, a narrow public message

On September 8, the Australian government released an exposure draft of its proposed Digital Duty of Care. According to the authors, it is a welcome and necessary reform that requires online services to proactively identify foreseeable risks of harm and take reasonable steps to address them. They describe it as a positive step towards protecting children in particular.

The draft bill also requires platforms to provide "user empowerment tools", which are yet to be specified. Branded "My Feed, My Way", the proposal would give users aged 16+ more control over whether their primary social media feed shows algorithmically recommended content or only material from accounts they follow, in chronological or reverse chronological order.

The authors say media outlets and commentators, encouraged by the government's messaging, seized on the simplest idea: that Australians could turn off 'the algorithm' for the first time. They ask why that idea became the focus of the public case for reform.

Why algorithms cannot simply be switched off

The researchers argue that 'the algorithm' has become shorthand for the much larger machinery of platform power. Harmful recommendations, predatory advertising, addictive design, political polarization and creator visibility are all framed as its fault. The concerns are real, they write, but algorithms are not 'bad guys' in themselves. What matters is what they are optimized for, whose interests they serve and what effects they produce at scale.

They state that there is no single algorithm to switch on or off. Recommender systems help people navigate an abundance of content, and platforms also use algorithms to moderate content, rank search results, target advertising and detect spam or inauthentic activity. These systems underpin safety work too.

The authors point to YouTube as an example of different optimization. They say the company has worked routinely with a group of international experts to feed evidence-based guidance into tweaks of its recommender systems for minors' accounts, deprioritizing low quality content that may be harmful at volume. They add that, since the social media age restrictions were implemented, Australian under-16s can no longer benefit from this wellbeing-oriented optimization, including on parent-supervised teen accounts.

Chronological feeds are not free of algorithms

The authors note that many platforms already offer tools to modify, disable or clear the algorithms shaping feeds, though these are often hard to find, use or understand. They stress that a chronological or 'following' feed still applies rules about what is included, excluded and ordered.

A choice between two platform-defined modes offers only limited agency, they argue. Citing their own research on advertising transparency, they say information and controls do not necessarily make platform systems transparent or give users more power over them. Alternatives could include feeds that prioritize a user's most loved accounts, let users nominate valued sources, or offer more direct control over ranking objectives. They cite Bluesky's "marketplace of feeds", where users can design and share curated feeds.

They also warn that content must exist before it can be recommended. Young people without accounts can still reach designated platforms while logged out, which creates a "cold start" problem for recommenders. If misogynistic or violent material is readily available before the system knows much about a user, they write, removing personalization does not resolve the underlying problem.

What the duty of care actually asks for

The authors say the Digital Duty of Care itself recognizes this broader problem. For children especially, it reaches beyond recommender systems to the material a service makes available, the design features through which users meet it, and the wider systems and processes that create risk. In their view it is considerably more sophisticated than the debate about on-off switches suggests.

They also see a tension in the government's position. Presenting 'the algorithm' as something users need protection from sits awkwardly with requiring platforms to anticipate and mitigate harms, which logically calls for more sophisticated algorithmic approaches, not fewer. At platform scale, spotting spam, abuse, coordinated harmful activity or repeated exposure to content such as eating-disorder material cannot be done manually.

For users and policymakers, the message is that the detail of the bill, and how the still-unspecified user empowerment tools are defined, matters more than the slogan.