Pluralistic: 30 May 2020

Today's links

Australia caves on "robodebt" (permalink)

The Australian government used an automated system to detect "welfare overpayments" from the Centrelink Agency. The system tormented Australia's poorest, most vulnerable people for years by insisting that they had been overpayed and demanding repayment with interest.

These "robodebts" became notorious, but successive Australian governments insisted that all was well, any problems were isolated incidents, and the tales of torment and misery were just moaning from workshy shirkers who'd been bathing in government money.Now, the government has admitted that at least 373,000 people were billed for AUD721,000,000 that they didn't owe, and has promised to refund them. The victims of this automated system had been targeted through a "discredited income-averaging scheme."

Government services minister Stuart Robert, who helped create the robodebt crisis in 2015, did not apologise, insisting that the system "was developed to make identifying welfare overpayments more efficient."

Neither Robert nor the Scott Morrison regime would say whether the people who'd been defrauded of $721m would be paid interest on the money that had been taken off them under false pretenses. Nor would the government confirm that it was scrapping the system.

Leaks seen by The Guardian reveal that the government's own assessment of the programme is that it is only viable because people must find years-old payslips in order to challenge robodebt notices – if the system was required to establish debts on its own, it would collapse.

Administrative arbitrators working for the Australian state repeatedly told the government that the scheme is illegal. Nevertheless, the government has pressed on, doubling down, even after it was forced to admit that it had stolen money from hundreds of thousands of people.

It's easy to understand why. Australia is an oligarchic, neoliberal place of increasing inequality, housing insecurity, and waves of climate emergencies that threaten the habitability of much of the continent's landmass.

To preserve the status quo in the face of these catastrophic threats, Australian elites need to make scapegoats of others – to find ways to get working people to fight amongst themselves, rather than turning on the few at the top.

They've been mashing the "scapegoat aboriginal people button" for centuries and its efficacy started to wane. The "scapegoat asylum seekers" button started to wear out after a few decades too.

Now it's "scapegoat disabled people, poor people, single parents and other people on benefits."

In that logic, all welfare payments are overpayments, because the Australian state owes nothing to those people, who should have the good graces to dig holes, climb in and pull the dirt in after them.

Their stubbourn refusal to stop needing food, shelter and care is just evidence of their moral failings.

Seen in that light, there's no reason to ever let up on robodebts.

Locus Award shortlist (permalink)

The shortlist for Locus Magazine's annual Locus Awards for science fiction, fantasy and horror is out, and it's full out outstanding reads from 2019.

Some of my favorites, with reviews:

The Grand Dark, Richard Kadrey (Best Horror Novel):

Catfishing on CatNet, Naomi Kritzer (Best YA Novel):

Magic for Liars, Sarah Gailey (Best First Novel):

Infinite Detail, Tim Maughan (Best First Novel):

Congrats to all the nominees!

Why I haven't written about CDA 230 (permalink)

People have asked me why I haven't written about Trump's executive order on social media and CDA 230. Here's why.

As everyone who understands the law knows, this will not survive contact with the judiciary. It's unconstitutional and incoherent and just stupid.

It's as if Trump declared up to be down, and then threatened FAA sanctions against anyone caught standing on the ground. This will doubtless inflict pain and chaos, but the first judge that hears the case will tell him to knock it off and stop being an idiot.

The real purpose – tissue thin, totally obvious – is to get us to stop paying attention to white nationalism, pandemic genocide, 101,000 dead, and corruption and start talking about whether up is down.

And life is short.

We've got things pretty good, all things considered. My family is solvent and healthy (for now) and I'm working and writing, but like everyone, I'm in a pretty brittle place – both very sad and very scared, and often very angry. It's not a good way to be.

And writing explainers about why up isn't down, watching my mentions fill with cultists who are one purple shroud away from a pudding cup, insisting that down is actually up, all of that, it fills me with despair.

It's not that I don't see the point of it. It's not to change minds in the death cult, it's to ensure that the vast majority of people who have sensibly never bothered to fill their brains with the minutiae of Clinton-era internet laws don't get misled.

The point of setting out bait for the Zombies From the Planet of Motivated Reasoning isn't to talk to the zombies, it's to play to the gallery. It's to fill the data void.

The people who are filling that void are doing important work, but some days, I just can't.

If you want to know why up is not down, here's some reading:

Sarah Jeong (always excellent, now more than ever):

On Techdirt, Mike Masnick:

And most comprehensively, my EFF colleagues Aaron Mackey and David Greene:

Thank you all, you brave void-fillers, for your service in the up-and-down wars. I'll see you on the other side.


Bus drivers refuse to take arrested protesters to jail (permalink)

As US police forces express solidarity with their murderous colleagues by effecting mass arrests of protesters, they have a logistical challenge: how do you cart away all the people you've arrested?

One traditional solution: commandeer a city bus, fill it with prisoners, and have the bus driver take them to the police station for processing.

But this requires the cooperation of bus drivers, and in most cities, the drivers are both racialized and unionized, and that means they don't want to help arrest protesters, and they don't have to.

So in Minneapolis, drivers are telling cops they're on their own.

New York drivers, too.

“ATU members live with similar fears on a daily basis. ATU members face racism daily. Our members live in and work in neighborhoods where actions like this happen, and where this took place, now watched in horror across the globe,” ATU Local 1005 said in a statement.

AI has been stagnating for a decade (permalink)

MIT CS PhD student Davis Blalock undertook a review of papers documenting progress in "pruning" for machine learning (a critical means of eking out performance improvements), and found little progress for a decade, despite the authors' claims.

Rather, the improvements could be chalked up to variations in measurement – the papers' authors were choosing to benchmark their systems using metrics that made them look good, but when these benchmarks were normalized, the improvements largely vanished.

His work confirms the 2019 ACM SIGIR meta-analysis that confirmed that the field's "high-water mark … was actually set in 2009."

Writing in Science, Matthew Hutson documents a series of these analyses that find little support for AI hype. Rather, the progress in AI seems to be an artifact of publishing bias from scientific journals.

This is a well-understood problem: journals like to document dramatic improvements, not incremental ones. So researchers who create new ML systems that perform existing tasks more efficiently are published with fanfare.

Researchers who tweak existing systems to match those performance gains languish in obscurity. Academic career advancement is predicated on publication, so research agendas are distorted to match publication bias, producing the illusion of a run of spectacular breakthroughs.

To maintain the illusion, researchers shy from comparisons with existing systems, creating bespoke benchmarks that cast their work in a good light. And since the field is growing fast, there aren't enough skilled reviewers to flag this stats-juking prior to publication.

These methodological issues are not limited to AI journals. As Ben Goldacre's 2012 book "Bad Pharma" documents, the most widely prescribed class of drugs is statins, and statins are (were?) not tested head to head.

Rather, each new statin was compared to placebos. That means that there were several statins on the market, and yet we didn't know which one performed best, and this is the most widely prescribed class of drugs!

To their credit, many scholarly and scientific publishers are taking steps to fight publication bias.

My favorite is "registered reports" where researchers approach journals before they experiment.

The journals pre-commit to publishing the outcomes based on the salience of the research agenda and the rigor of their methodology – rather than basing their publication on how spectacular the finding is.

(Image: Cryteria, CC BY; Gartner, modified)

This day in history (permalink)

#5yrsago The PATRIOT Act is uglier than you thought, and what to do about it

#1yrago How the "prosperity gospel" convinces poor people to give everything to grifty millionaire preachers

#1yrago Wealth is correlated with greed, dishonesty and cheating — are these effects or a causes?

#1yrago New Amazon patent application reveals "solution" to missed Alexa instructions: always on recording

#1yrago Now that Uber and Lyft are public, their inevitable financial collapse is much clearer

Colophon (permalink)

Today's top sources: Asher Wolf (, (, Naked Capitalism (

Currently writing: My next novel, "The Lost Cause," a post-GND novel about truth and reconciliation. Yesterday's progress: 522 words (21565 total).

Currently reading: Adventures of a Dwergish Girl, Daniel Pinkwater

Latest podcast: Someone Comes to Town, Someone Leaves Town (part 03)

Upcoming appearances: Discussion with Nnedi Okorafor, Torcon, June 14

Upcoming books: "Poesy the Monster Slayer" (Jul 2020), a picture book about monsters, bedtime, gender, and kicking ass. Pre-order here: Get a personalized, signed copy here:

"Attack Surface": The third Little Brother book, Oct 20, 2020.

"Little Brother/Homeland": A reissue omnibus edition with a new introduction by Edward Snowden:

This work licensed under a Creative Commons Attribution 4.0 license. That means you can use it any way you like, including commerically, provided that you attribute it to me, Cory Doctorow, and include a link to

Quotations and images are not included in this license; they are included either under a limitation or exception to copyright, or on the basis of a separate license. Please exercise caution.

How to get Pluralistic:

Blog (no ads, tracking, or data-collection):

Newsletter (no ads, tracking, or data-collection):

Mastodon (no ads, tracking, or data-collection):

Twitter (mass-scale, unrestricted, third-party surveillance and advertising):

Tumblr (mass-scale, unrestricted, third-party surveillance and advertising):

When life gives you SARS, you make sarsaparilla -Joey "Accordion Guy" DeVilla

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.