Is the ‘Albo-rithm’ making bad recommendations?

By Nathan Powell, Co-Founder and Chief Strategy & Product Officer, Fabulate

The Australian Government has found its latest villain as it seeks to tame the internet. That villain? The algorithm.

Under its proposed new Digital Duty of Care laws, Australians could be able to switch off algorithmically recommended content and instead see posts only from accounts they have chosen to follow.

On the surface, giving people more control over their feeds seems entirely reasonable, and I don’t have a problem with that. People should understand why they’re seeing something and have meaningful choices about how their experience is shaped.

Where I struggle is with the idea that recommendations themselves are the problem we’re trying to solve.

Misogyny, violence and racism didn't suddenly appear the first time someone opened a social media app. They existed long before in our homes, our schools, our workplaces and communities, long before any of us had a For You page. They sadly are a reflection of our society

What social media has done is make them more visible, connect people around them and, in some cases, amplify them at an enormous scale. Platforms absolutely have a responsibility for that amplification because, if you’ve built and monetised a system capable of recommending content to billions of people, you can’t wash your hands of what that system repeatedly chooses to promote.

But if the problem is misogynistic content, let’s deal with misogynistic content. If it’s material promoting racism, violence or extremism, deal with that material and hold platforms accountable when their systems continue to distribute it.

Don’t confuse the content with the technology used to discover it.

An algorithm is ultimately trying to solve a fairly mundane problem. Vastly more content is available than any of us could possibly consume, so platforms need some way to decide what to show us. Some of the best things we find online are things we never knew to search for, and recommendations make that possible.

The problem begins when that same system discovers that outrage keeps us watching, so it gives us more outrage, then something slightly more extreme and eventually something more extreme again.

For many people, discovery is inseparable from social media itself, and the algorithm's serendipity is part of the experience.

It’s also worth pointing out that Australians aren’t currently trapped inside algorithmic feeds with no way out. Instagram already offers a Following feed, while YouTube allows users to disable the watch history that powers personalised Home recommendations. We’re not introducing consumer choice into an ecosystem where none existed before; we’re potentially making that choice more prominent and consistent across platforms.

What I don’t buy is the idea that an internet without recommendation would somehow be a safer internet. It might become less useful and considerably less engaging, but no one is proposing a war on toxic content on social.

Social media today isn’t Facebook of 2009. We don’t open TikTok, Instagram, or YouTube only to see things from people we already know. We go there to discover a creator, a song, a restaurant, a product or an idea we didn’t know existed five minutes earlier.

Recommendation has also democratised who gets heard, often in unwieldy ways . A creator with 5,000 followers can make something exceptional and reach half a million people in hours, while a small business can find customers without a multinational's media budget. Distribution is no longer determined entirely by how large an audience you already own.

Algorithms can create rabbit holes, and we should absolutely be concerned when recommendation systems repeatedly push people further down harmful ones, but again this is about the content, not the mechanisms of discovery.

Algorithms are doorways that introduce us to people, interests, communities, and ideas we didn’t know to search for in the first place.

That’s not a flaw in the modern internet; it’s an integral function that makes the internet work.

We also shouldn’t assume that removing recommendations creates some inherently safer version of social media. If somebody chooses to follow 50 accounts publishing misogynistic content, removing the algorithm doesn’t make the misogyny disappear; it just means they selected the sources themselves and won’t see much else.

There is no neutral feed. A chronological feed reflects who you’ve chosen to follow, a recommended feed reflects decisions about relevance, and a moderated feed reflects decisions about what is acceptable.

The better question isn’t whether algorithms are good or bad, but what we’re comfortable letting them amplify and what responsibility platforms should carry when they repeatedly push people toward demonstrably harmful material.

That’s where I think regulation should be at its strongest.

Platforms should have a responsibility to act when their systems repeatedly surface harmful material, while giving consumers meaningful controls and greater transparency over how recommendation works. And where particular categories of content cross a clearly defined threshold of harm, the focus should be on that content and how it is distributed, rather than recommendation itself.

But don’t throw out discovery in the process.

Because even if we built the safest recommendation system imaginable tomorrow, we’d still face a much harder problem.

Algorithms shine a light on society, and sometimes we don’t particularly like what they show us. Platforms determine what their systems optimise for and must take responsibility for the consequences, but they didn’t invent the attitudes those systems discover.

Turning down the amplifier doesn’t remove what was being amplified. If problematic content is the problem, then tackle content. If we’re genuinely concerned about young men being drawn towards misogyny, changing what appears in their feeds can be part of the response, but surely the conversation can’t end at the algorithm.