Forgive me my pedantry, but if Australia is going to regulate social media algorithms, we should be precise about what an algorithm actually is.
Discussion of the government’s proposed digital duty of care has been framed in the media as giving Australians the ability to ‘turn off’ or ‘opt out of’ social media algorithms.
The alternative is usually presented as the old-fashioned feed with posts from accounts you have chosen to follow, arranged from newest to oldest.
There’s just one problem: that’s an algorithm too. A reverse chronological feed takes a collection of posts and applies a rule – sort by timestamp, newest first – to determine what appears on your screen. That’s an algorithm!
In reality, we aren’t debating whether people should experience social media with or without ‘algorithms’, a term that has become increasingly meaningless. Instead, this is a debate about who gets to pick the algorithms and what sort of content they show us.
Instead of forcing Australians into a crude choice between Big Tech’s engagement-driven black boxes and a raw chronological dump, the digital duty of care should be ambitious enough to let users fully choose how their feeds are curated and by whom.
The government has made much of the choice its My Feed, My Way approach would apparently give Australian users.
As Communications Minister Anika Wells put it, ‘Whether they want to keep the algorithm, whether they want to do things a different way, it should be their choice, not big tech’s choice.’
All great stuff. But the choice on offer is pretty narrow. You either get the platform’s recommendation system or a following-only feed.
People often say they prefer feeds featuring only accounts they have actually chosen to follow, much like how Facebook and other platforms originally operated before more sophisticated news feed sorting was developed.
But observed behaviour and revealed preferences tend to be quite different. Large field experiments on Facebook and Instagram have found that switching users to a reverse chronological feed reduced both time spent and activity on platform.
That by no means is a smoking gun proving people dislike following-only feeds. Lower engagement is not the same thing as lower welfare, and many people believe they spend too much time on social media anyway.
But it does challenge the assumption that a simple following-only feed is what people really want.
Nor is a chronological feed obviously healthier. The same research found greater exposure to content from sources classified as untrustworthy and no detectable reduction in political polarisation.
As the eSafety Commissioner has noted, users often switch back to engagement-optimised feeds despite claiming they don’t like them.
So the government’s approach does deserve some credit. It breaks the default. It requires platforms to offer users a meaningful alternative and, crucially, to respect that choice.
Bluer skies
But if the principle is that Australians should control how their information environment is curated, why stop at two options designed by the platforms themselves?
Bluesky offers a glimpse of a more interesting model. The Twitter-like social network lets users and third-party developers create, install and customise different feed algorithms rather than being locked into a single platform-selected ranking system. One recent study counted 97,719 custom feeds.
I have used Bluesky’s new AI-assisted feed builder to make one tailored to my own information diet: recent posts from Australian journalists, China watchers and others, ranked according to the signals I care about. It may be useful to an audience of one – me. But that’s the point: it’s my feed, my way.
Some of those Australian journalists – and more than 47 federal politicians, by my count – are already there. (One small way the government could help loosen the network effect of X is sharing major breaking announcements on Bluesky 10 minutes before X.)
Even many Bluesky users seem not to realise that the platform has ‘algorithms’ at all. One custom feed exists solely to surface commentary from people who make posts arguing Bluesky ‘doesn’t have an algorithm’ – on a platform that in fact offers users more algorithmic choice than almost any other.
Clearly, there is some education to do. Giving people more algorithms to choose from only works if those choices are intelligible, and if someone can turn the technical business of ranking a feed into a service people understand and trust.
That is essentially the idea behind ‘middleware’. In a landmark 2020 paper, a Stanford working group led by Francis Fukuyama proposed the creation of a ‘middleware’ industry in which independent services could sit between users and dominant platforms and offer competing ways to rank, label and curate the same underlying content.
The researchers were not worried about the harms of any given individual post. They were more concerned with the fact a handful of private companies had accumulated an extraordinarily important editorial function – effectively the infrastructure for deciding which voices were amplified and how billions of people encountered information.
Middleware would prise apart the platform and that editorial power. Instead of choosing only between Meta’s recommendation engine and Meta’s chronological feed, users could choose whose ranking system they trusted – perhaps one built by a news organisation, a university, a civil society group or anyone else capable of offering a useful service.
This wouldn’t be easy. Middleware raises hard questions about data access, privacy safeguards, commercial incentives, liability and governance. It’s certainly not something Canberra could easily bolt onto this legislation tomorrow.
But the government’s reform establishes an important regulatory principle that platforms should not have an absolute right to determine how a user experiences their networks. Stage one is to break the default. Stage two should be to expand the choice.
That could also take some of the heat out of the censorship debate surrounding online safety. Platforms would still be required to deal with illegal material and meet baseline safety obligations.
But for lawful content, more of the political and editorial layer could be left to users, who could build their own feeds or subscribe to those created by people and organisations they trust.
That doesn’t solve all the difficult questions about moderation on social media. Nor does it guarantee people will choose their feeds and content wisely.
It does, however, distribute editorial power rather than simply taking it away from Silicon Valley and giving it to a government agency.
Prime Minister Anthony Albanese says the reform is ‘not about giving government control, it is about giving people control.’ That’s a good start, but let’s actually take that principle seriously.
Don’t stop at giving Australians a choice between Big Tech’s black box algorithm and a simple following feed. Let’s build towards a system in which they can actually choose whose algorithm they trust.
This is a lightly edited version of an article first published in Capital Brief.