The Albanese Government is taking the next step following its world leading social media age ban to ensure Australians have more choice online, with draft legislation for the Digital Duty of Care released today for targeted consultation.These new laws will put the onus on digital service providers to provide a safe online environment, setting minimum standards for the features and tools they use.This will include an Australian first, “My Feed, My Way,” initiative which will give Australians control over their social media algorithms.
Statistical models are not technically algorithms; they are a heuristic which “approximates” a solution probabilistically. They use algorithms to function and operate on data (like everything a computer does) but they are not algorithms themselves.
The main thing they have to do is clearly define that results derived by statistical modeling are opt-in, set limits on what personal data can be used to present content, and ensure there are significant penalties for crippling the experience of everyone who hasn’t opted-in, which all are guaranteed to do to force users to opt-in.
They should define a bare minimum baseline of data points that should be made available to users to control the content they want to see using non-statistical algorithms (e.g. time, location, content type or popularity, friends/subs/followed only or ones they recommend themselves, etc). They should also define that opt-in/out (enabling and disabling statistical model results) should be a part of the main filter UI that uses the aforementioned data points; a simple toggle at the top of their page/feed, permanently visible and near-instantly applicable; otherwise they’re guaranteed to hide it away and make opt-in/out a major friction point.
Basically you need to treat them like they’re sociopathic criminals who will never act in good faith voluntarily, because all multinationals are lead by criminally corrupt sociopaths.
Which makes “opt out of algorithms” even more wrong.
They should define a bare minimum baseline of data points that should be made available to users to control the content they want to see
I’m not a fan of specifying too much here. Many platforms don’t track location while other fields that we can’t think of in advance might be useful. Finding a more general rule that still gets the intent across might be hard but I’d rather have it a bit under-specified than suddenly requiring a social media site to track my location every time I post because someone wasn’t careful enough when writing the text of a law.
Statistical models are not technically algorithms; they are a heuristic which “approximates” a solution probabilistically. They use algorithms to function and operate on data (like everything a computer does) but they are not algorithms themselves.
The main thing they have to do is clearly define that results derived by statistical modeling are opt-in, set limits on what personal data can be used to present content, and ensure there are significant penalties for crippling the experience of everyone who hasn’t opted-in, which all are guaranteed to do to force users to opt-in.
They should define a bare minimum baseline of data points that should be made available to users to control the content they want to see using non-statistical algorithms (e.g. time, location, content type or popularity, friends/subs/followed only or ones they recommend themselves, etc). They should also define that opt-in/out (enabling and disabling statistical model results) should be a part of the main filter UI that uses the aforementioned data points; a simple toggle at the top of their page/feed, permanently visible and near-instantly applicable; otherwise they’re guaranteed to hide it away and make opt-in/out a major friction point.
Basically you need to treat them like they’re sociopathic criminals who will never act in good faith voluntarily, because all multinationals are lead by criminally corrupt sociopaths.
it’s more likely that the government is going to fumble the legislation
Which makes “opt out of algorithms” even more wrong.
I’m not a fan of specifying too much here. Many platforms don’t track location while other fields that we can’t think of in advance might be useful. Finding a more general rule that still gets the intent across might be hard but I’d rather have it a bit under-specified than suddenly requiring a social media site to track my location every time I post because someone wasn’t careful enough when writing the text of a law.
Other than that, I agree 100%.