Australia Plans to Let Users Switch Off Social Media Algorithms
This matters for social media users in Australia who would gain more control over their feeds, and for tech companies who may face substantial penalties for non-compliance.
Look at how this was made. The quality - or the lack of it - tells us something the policy debate is not discussing. Not the debate about whether Australians should have the right to switch off the algorithm; that question answers itself, as questions of liberty generally do, in the affirmative. The debate that is not happening concerns what fills the room when the algorithm is switched off. And it is there, in the unexamined vacancy behind the toggle, that the true character of this legislation will be revealed - not in the promise the Australian government has made, but in the object Meta, Google, and TikTok will actually build to satisfy it.
CHOICE, n. The condition of possessing an off-switch that the manufacturer has agreed, under legislative duress, to install somewhere in the wiring of a machine built expressly so that you will never wish to reach for it.
The Australian government proposes to grant its citizens the right to switch off the algorithms by which Meta, Google, and TikTok arrange their feeds. The proposal is generous in the manner of a casino installing a clock: technically a service to the patron, practically an admission of what the house has been counting on him not to notice.
This arrangement is presented as a gift from government to citizen - the restoration of a sovereignty the algorithm had stolen. Let us ask when the algorithm became the thief, and who profits from casting Canberra as the sheriff who returns the horse.
The genealogy begins not with the citizen’s suffering but with the state’s discovery of a new instrument. For a decade the Australian government watched Meta, Google, and TikTok build the most efficient attention-harvesting apparatus in the history of the species, and did nothing that mattered - some fines, some hearings, the theater of concern. Now, suddenly, a bill to let Australians switch off the feed’s algorithmic ranking. One should notice the timing before admiring the gesture. A government does not discover the citizen’s autonomy by moral insight. It discovers it when the political cost of appearing powerless against foreign platforms exceeds the cost of appearing to act.
Before debating the optimal outcome, establish the floor. No person using a social media service in Australia should be compelled to accept an algorithmic feed they cannot decline. That is the standard. Not “more choice,” not “greater digital wellbeing” - the specific right to turn a mechanism off, and have it stay off, verified by someone whose job it is to check.
The garment manufacturers of 1911 told me fire doors would slow production. The mill owners told me a forty-hour week would collapse the textile trade. Meta, Google, and TikTok will tell the Australian Parliament that an algorithm is too integrated into the architecture of the platform to be switched off without breaking the product. This is the same argument, wearing a different collar. It may even be technically true in places. That does not make it a reason to abandon the standard - it makes it a specifications problem, and specifications problems get solved when the penalty for not solving them exceeds the cost of solving them.
The announcement reads as a straightforward win for the user: an Australian can flip a switch and the algorithm goes away, replaced by a feed of their own choosing, and platforms that resist face penalties with teeth. That is the framing the government wants read out loud, and it is the framing Meta, Google and TikTok would prefer you not examine, because the phrase doing the work - “switch off the algorithm” - describes a control that does not exist in the shape the legislation imagines it.
There is a gate across this road, and the gate is called the algorithm, and the Australian government proposes to hand every citizen a key to it. The modern man says: I see no reason for this fence except that Meta and Google and TikTok wished to keep me scrolling, and that reason is a bad one, so let us tear the whole structure down. He may well be right. But before I cheer him I want to know what the fence was actually doing, because in my experience the fence that was built to trap you was also, quite often, the fence that kept out something worse.
G. K. Chesterton
There is a gate across this road. The modern man says “I see no reason for it; let us remove it.” The wiser man says “If you see no reason for it, I will not let you remove it. Go away and think. When you can tell me why it is here, I may let you destroy it.”
My opponent, a man of considerable technical competence and moral urgency, has built a bridge of legislation to cross the gap between the user and the algorithm. He argues in the mechanics of his design, that the “switch” must be auditable, chronological, and enforced by inspectors who can read the code. He is right,, that if the law is vague, the engineers will hide the beast in the basement and call it a feature. He is right,, that a toggle that does not actually toggle is a lie painted in the colour of truth. I concede this entirely. A law that cannot be enforced is merely a suggestion; a switch that does not work is a trick. To demand that the “off” position be real, not merely rhetorical, is to demand that the word “off” mean something. This is not progress; it is grammar.
But here is the fence that my opponent has not merely overlooked, but assumed was already open. He speaks of the “chronological feed” as if it were a neutral default, a blank slate upon which the user’s will is written plainly. He believes that if we remove the algorithm, we return to a natural state of order. I tell you this is a profound error. The chronological feed is not the absence of an editor; it is a different editor, an editor who believes that time is the only measure of value. It is the tyranny of the recent. It is the assumption that the most recent word is the most important word, the most recent face is the most visible face. This, too, is an algorithm. It is the algorithm of novelty. It is the algorithm of the immediate.
My opponent’s solution is to replace one invisible hand with another, equally invisible hand. He wishes to trust the user’s eye more than the machine’s mind. But the user’s eye is not a neutral instrument. It is a weak instrument. It is an instrument that has been trained, not by legislation, but by decades of the very algorithms he now seeks to dismantle. To hand the chronological feed to a human being is to hand a loaded pistol to a child who has been told that the trigger is safe. The child will pull it. The child will pull it because the recent thing is bright, and the bright thing is loud, and the loud thing is new. The chronological feed does not protect the ordinary person from the algorithm; it merely makes the algorithm’s work more exhausting. It requires the ordinary person to become a filter for a world that is screaming to be seen.
Consider, if you will, the library of Alexandria. When the books were arranged not by subject, but by the order in which they were burned, the result was not clarity. The result was chaos. The chronological feed is a library where the books are stacked by the day they arrived, regardless of whether they are poetry or poison. My opponent believes that if he removes the poisoner, the poet will speak. But the poet is drowned in the noise of the new. The fence was not built to keep the poisoner out; it was built to keep the noise out. It was built to allow the old word to meet the new word in a space that was not dictated by time.
The Inspector, whom my opponent conjures as a savior, cannot read the code because the code is not the enemy. The enemy is the human desire for the new. The Inspector can verify that the switch is off. But he cannot verify that the user is not, in fact, looking for the new. He cannot verify that the user is not, in fact, scrolling past the old to find the bright. The law can mandate the architecture, but it cannot mandate the attention. And without attention, the architecture is merely a room with no doors.
I do not argue for the preservation of the algorithm because it is good. I argue for the preservation of the fence because it is visible. The algorithm is a black box; it tells you what you want before you know you want it. The chronological feed is a white box; it tells you what is new, and assumes you want it because it is new. One is a deception; the other is a distraction. Both are failures of the ordinary person’s ability to choose. The true reform is not to choose between the invisible hand and the chronological eye. It is to build a fence that forces the user to stop, to look, to choose. It is to make the choice difficult. It is to make the choice visible.
My opponent’s solution is to trust the user. I say this is the greatest distrust of all. To trust the user to navigate a chronological feed is to trust him to navigate a hurricane. The fence was not built to protect the algorithm. It was built to protect the user from himself. If we remove it, we do not give him freedom. We give him the freedom to be overwhelmed. And when he is overwhelmed, he will not thank the Inspector for his audit. He will thank the algorithm for hiding the storm.
Perkins-style
Before debating the optimal outcome, establish the floor. No person affected by this event should be subjected to an information architecture that exploits cognitive vulnerabilities without their informed consent or ability to opt out of manipulation. Does the current response meet that floor? It does not. The traditionalist argues that the algorithm is a fence built to trap us, yet simultaneously serves as a hedge against worse winds, such as the radicalization of boys in Adelaide or the isolation of widows in Perth. He asks us to keep the fence because it keeps out something worse, even if it traps us in the process. I agree with him on one point, and it is a point he likely dismisses as sentimental: the algorithm does perform a filtering function. It does decide, imperfectly, what is harmful and what is useful. To tear the whole structure down because it was built for profit is to ignore the reality that the structure now exists and that its removal creates a vacuum which will be filled by something less regulated, not more free.
However, the traditionalist’s defense of the status quo rests on the assumption that the “fence” is neutral. It is not. The fence is owned by the advertiser. The traditionalist speaks of the algorithm as a natural hedge, a windbreak of sorts. But a hedge does not choose which wind to break; it simply stands there. The algorithm chooses. It chooses to show the grieving widow forty consecutive posts about euthanasia not because she needs support, but because grief is a high-value engagement metric. It chooses to walk the boy in Adelaide into a room full of men who hate him not because it failed to stop him, but because anger drives clicks. The traditionalist conflates the side effect of a broken system with the virtue of the system itself. He sees the hedge and calls it shelter, ignoring that the hedge is planted on private land and charges admission for the view.
The core divergence between us is administrative. The traditionalist believes that if we remove the profit motive, or if we simply hand the user a key, the fence will cease to be a trap. He assumes that the user, once empowered, will make the same choices as the algorithm, but with better intent. This is a failure of engineering. The user does not have the capacity to process the volume of information the algorithm manages. To ask a widow in Perth to manually curate her feed is to ask her to do the work of a machine designed to exploit her exhaustion. The standard required is not user autonomy; user autonomy is a myth in the face of behavioral psychology. The standard is informed consent and structural transparency. We must mandate that the algorithm disclose its primary objective function - whether it is maximizing engagement, retention, or sales - and allow the user to override it with a chronological or neutral filter that is genuinely neutral, not a curated illusion.
The cost of this regulation is not trivial. It requires an inspection regime that audits the code, not just the output. We need inspectors who understand machine learning, who can trace the weighting of variables in a recommendation engine. We need a penalty schedule that makes non-compliance more expensive than compliance. The traditionalist assumes that the market will correct itself if we just “tear down the structure.” The market corrects nothing. The market corrects only when forced to by the law. If we remove the algorithmic fence, we do not get freedom. We get a black market in attention. We get unregulated data brokers who will sell the same vulnerabilities to the highest bidder, without the small measure of protection the current algorithm provides, however flawed. The cost of regulation is the creation of a new administrative body, specialized and well-funded. The cost of deregulation is the total loss of leverage for the individual.
Consider the National Industrial Recovery Act of 1933. It was struck down by the Supreme Court, but the principle remained: industry cannot regulate itself. The traditionalist argues that the algorithm is a tool, and like any tool, it is only as good as the hand that wields it. But the hand is not wielding the tool; the tool is wielding the hand. The algorithm conditions the user’s response. It is not a passive fence. It is an active shepherd. To suggest that we can simply “hand every citizen a key” is to misunderstand the nature of the lock. The lock is not on the gate; the lock is on the mind. We cannot legislate the mind. We can only legislate the environment that shapes it. We must mandate that the environment be transparent. We must mandate that the objective functions be disclosed. We must mandate that the user have a genuine, non-manipulated choice.
The traditionalist’s error is to treat the algorithm as a static structure, a fence that either exists or does not. It is a dynamic system, a living thing that adapts to regulation. If we regulate it, it will find loopholes. If we do not regulate it, it will find profits. The choice is not between freedom and control. The choice is between regulated control and unregulated exploitation. The floor is not freedom. The floor is dignity. Dignity requires that we are not treated as data points to be optimized. Dignity requires that we know why we are being shown what we are being shown. The cost of this is administrative complexity. The cost of not doing it is the erosion of the public sphere. I am willing to pay the administrative cost. I am not willing to pay the erosion of the public sphere.
The Verdict
Where They Agree
First, both agree that the primary motivation of the social media companies is profit, not user welfare. Chesterton states the algorithm is built “to hold your attention and sell it to advertisers,” while Perkins asserts the fence “is owned by the advertiser” and its objective function is to optimize engagement metrics. This shared, cynical view of corporate intent is foundational; it forces the argument onto the terrain of how to manage this profit-driven force, not whether it exists.
Second, and more surprisingly, both debaters agree that removing the algorithmic curation entirely creates a vacuum that will be filled not by neutrality but by a different, potentially worse, form of disorder. Chesterton warns that chronology is “anarchy with the appearance of order,” governed by “the fanatic, the bot, and the man with nothing else to do.” Perkins echoes this, cautioning that deregulation leads to a “black market in attention” with “unregulated data brokers.” Their shared skepticism about a return to a prelapsarian digital garden reveals that neither holds a naïve libertarian faith in unfiltered information streams.
Finally, both accept that the current algorithmic system, however flawed, performs a de facto curation function. Chesterton’s “fence” passively muffles harmful content as a side effect, while Perkins acknowledges the algorithm “does perform a filtering function” and that its removal eliminates a “small measure of protection.” This agreement is crucial: it means the debate is not about whether curation should happen, but about who or what should perform it and according to what principles.
Where They Fundamentally Disagree
The nature of the primary threat to the user. For Chesterton, the threat is fundamentally epistemological and internal: the user’s own “human desire for the new” and their “weak instrument” of attention, which renders them incapable of navigating a raw chronological feed without being overwhelmed. The empirical question is whether an unmediated feed indeed leads to greater user distress or radicalization, a claim that could be tested by studying user well-being in environments with and without algorithmic ranking. Normatively, he values a designed environment that protects individuals from their own cognitive limitations, prioritising stability over absolute autonomy. For Perkins, the threat is structural and external: it is the non-consensual manipulation by a profit-driven system. The empirical question is whether users, when given a transparent choice and clear information about the algorithm’s goals, can make better decisions for themselves. Normatively, he values informed consent and structural transparency as non-negotiable components of human dignity, even if the resulting choice is not the optimal one for well-being.
The role and capability of the regulating state. The disagreement here is over whether the state can effectively audit a dynamic, complex technical system. Perkins expresses that a well-funded regulator with technical experts can perform ongoing audits of the “objective function” of algorithms. This is an empirical claim that depends on the state’s ability to attract talent and keep pace with technological change. Chesterton is deeply skeptical, arguing that “the Inspector… cannot read the code because the code is not the enemy,” implying the problem is human nature, which no audit can fix. Normatively, Perkins believes in the necessity of state intervention to correct market failures, while Chesterton sees such interventions as likely to be gamed or to create unintended consequences worse than the original problem.
The definition of freedom and autonomy. For Perkins, freedom is the negative right to be free from non-consensual manipulation; the “switch” is the embodiment of this freedom, a tool for informed consent. The user’s potential to make a “bad” choice (e.g., leaving the algorithm on) is irrelevant because the floor of dignity has been established. For Chesterton, this is a hollow freedom that leads to “the freedom to be overwhelmed.” True autonomy, in his view, is not the ability to choose from a menu of curated options but the cultivation of judgment within a deliberately constrained environment that forces conscious choice. This is a purely normative clash between negative liberty (freedom from) and a more paternalistic concept of positive liberty (freedom to).
Hidden Assumptions
- G. K. Chesterton: Assumes that a chronological feed is intrinsically more chaotic and harmful than an algorithmic one. If false - if studies showed that chronological feeds actually reduce exposure to extremist content or improve user well-being by reducing addictive design - his entire defense of the “fence” would collapse.
- G. K. Chesterton: Assumes that the state is inherently less competent and more easily gamed than the corporations it seeks to regulate. If a regulatory agency demonstrably succeeded in enforcing a transparent and effective toggle, his argument against the legislation would lose its practical force, leaving only the philosophical objection to the concept of the switch itself.
- Perkins-style: Assumes that a sufficiently specific law and a technically adept regulator can reliably detect and prevent corporate circumvention of the “off” switch. If the algorithms are too complex and adaptive for any regulator to effectively audit, then his entire enforcement mechanism is built on a foundation of sand, rendering the legislation symbolic.
- Perkins-style: Assumes that “informed consent” is a meaningful concept in the context of sophisticated behavioral manipulation. If cognitive science demonstrates that mere disclosure of an algorithm’s goals is insufficient to counteract its effects, then his core normative standard of dignity would be unachievable through the mechanism he proposes.
Confidence vs Evidence
- Perkins-style: “A well-funded inspectorate can audit algorithmic code” - the evidence is speculative. He cites historical analogies (fire doors, the 40-hour week) which are administrative precedents, but provides no evidence that the specific technical challenge of auditing real-time, adaptive machine learning systems at scale is solvable. This is a significant evidentiary gap for a load-bearing claim.
- G. K. Chesterton: “The chronological feed is a tyranny of the recent that will overwhelm the user” - tagged with implied but the evidence is anecdotal. His evocative examples (the library of Alexandria, the loaded pistol) are powerful metaphors but not data. The actual psychological impact of chronological versus algorithmic feeds is an empirical question with contested and nuanced findings, which he treats as settled.
- Both-style: They express high confidence on contradictory empirical claims about what fills the void left by the algorithm. Chesterton is highly confident it leads to anarchy dominated by bad actors; Perkins is highly confident it leads to a black market. Resolving this would require concrete evidence from jurisdictions that have implemented similar rules or from studies of unmoderated platforms.
What This Means For You
When evaluating news about this legislation, be immediately suspicious of any coverage that does not address the technical feasibility of enforcement. The central practical question is not the politician’s intent, but the regulator’s capacity. Ask whether the proposed auditing body has the budget and expertise to compete with tech engineers. Crucially, probe claims about user behavior: is a chronological feed truly “calmer,” or is that an assumption? Demand to see the data on user well-being from platforms that offer a genuine chronological option. The single most important piece of evidence missing from this debate is a rigorous, independent study comparing mental health outcomes and content exposure across different feed-ranking systems.