Image: Archi Banal
Image: Archi Banal

PoliticsMay 29, 2023

How NZ plans to force tech giants to finally pay their taxes

Image: Archi Banal
Image: Archi Banal

A tax bill introduced alongside the budget commits the government to global rules aimed at stopping the likes of Meta and Google NZ sending their profits offshore. Terry Baucher explains the story so far.

Tax is full of acronyms; PAYE, GST, FBT… and BEPS. If you’re not familiar with BEPS, it stands for Base Erosion and Profit Shifting, and every time you call an Uber or log onto Facebook, their ultimate parent companies Uber Inc and Meta Inc are using BEPS to transfer income overseas from Aotearoa New Zealand.

As I explained to John Campbell last week, BEPS at its most basic level means shifting income from a higher tax country, in this case New Zealand with our corporate income tax rate of 28%, to lower tax jurisdictions. This profit-shifting erodes our tax base, hence the term BEPS.

BEPS is how Facebook New Zealand earned gross advertising revenue of over $154 million last year, but only reported net profit before tax of $3.3 million and ended up paying just over $1 million of income tax. That’s because during 2022 Facebook New Zealand paid a related company, Meta Platform Ireland Limited, over $149 million for the purchase of services. The corporate income tax rate in Ireland is 12.5%, the lowest in the European Union.

BEPS also explains why Uber’s New Zealand operations earned over $220 million in 2021 but the final income tax liability was just $545,000 income tax after fees of $117 million were paid to its Dutch holding company Uber International B.V. As Uber International B.V. is a holding company a preferential zero tax rate applies for Dutch tax purposes.

These numbers are dwarfed by those of Google New Zealand which in 2022 paid over $870 million in service fees to offshore affiliates, mostly to the Singapore-based Google Asia Pacific Pte. Limited. As you have probably guessed, Singapore has a preferential tax regime which is perhaps why Mastercard and Visa also route payments through Singapore.

In total, based on the latest financial information, Facebook, Google and Uber have paid more than $1.1 billion in service fees to overseas affiliates. In theory, at the corporate income tax rate of 28% this represents over $319 million in potential tax.

That’s just the impact in Aotearoa New Zealand. Globally, according to the Organisation for Economic Co-operation and Development (OECD), BEPS practices cost countries US$100-240 billion (NZ$159-382 billion) in lost revenue annually.

Are the little guys finally standing up to tech giants like Meta and Google? (Illustration: Toby Morris)

Surely such manoeuvres are not legal? At the moment, yes. The problem is that the rapid development of the digital economy has outstripped the rules of the current international tax system. These are based on principles developed after World War 1 and critically centre on where economic activity is physically carried out.

But in today’s digital economy driven by the immensely valuable intellectual property wrapped up in the algorithms and apps of Facebook and Uber, it’s virtually impossible to apply a location basis of taxation.

With the old rules rendered ineffective, how are governments reacting to this threat to their tax base?

Individual tax authorities are investigating the level of service fees charged. According to Meta Inc’s financial report for the year ended December 2022, filed with the US Securities and Exchange Commission, it had accrued US$5.49 billion (NZ$8.7 billion) in respect of “uncertain tax positions”. These relate to transfer pricing with its foreign subsidiaries, including “licensing of intellectual property, providing services and other transactions.”

It’s not clear from its latest financial statements whether Inland Revenue is currently investigating Facebook New Zealand.

Meta is not the only tech company being investigated. Uber is mired in several tax audits and arguments around the treatment of its contractors. Alphabet Inc, the owner of Google, has also been under investigation. In 2019, its Australian subsidiary paid over $500 million in settlement of a tax dispute with the Australian Tax Office.

Some countries including India, France and the United Kingdom have introduced a digital services tax, which impose a low flat tax rate on the value of digital services provided within the country. Inland Revenue has consulted on the topic but not proceeded with implementation.

But a more comprehensive approach is required which is why the OECD together with the G20 group of nations launched the Inclusive Framework on BEPS initiative. At present 138 jurisdictions nations including Aotearoa New Zealand have agreed to the Inclusive Framework. This led to the so-called Pillar One and Pillar Two proposals, a provisional agreement on which was reached in October 2021 (although arguments over implementation are still ongoing). These proposals target the largest multinationals with annual revenue exceeding €750 million.

Pillar One aims to ensure a fairer distribution of profits and taxing rights between countries for these larger multinationals. In other words, how much of the digital pie each country can tax. Critically, the intention is for Pillar One to replace any country-specific digital services taxes (which, as you can imagine, the likes of Meta loathe).

Pillar Two is the biggie as it aims to restrict the impact of lower tax jurisdictions by imposing a global minimum of 15% for those multinationals with annual revenue exceeding €750 million. This will be done by Global Anti-Base Erosion (GloBE) rules.

This month’s budget was accompanied by a tax bill introducing the relevant legislation required for the implementation of the GloBE rules. The legislation and commentary include a heap of new tax acronyms including DIIR (Domestic Income Inclusion Rule), POPE (Partially Owned Parent Entity) and QDMTT (Qualified Domestic Minimum Top-up Tax), which a rather cynical overseas tax advisor has suggested should be pronounced Q-Dammit.

The commentary to the tax bill explains the legislation will only take effect once a “critical mass of countries” has adopted the GloBE rules. This is thought to be “very likely, though is not certain”. If that critical mass is reached, then the rules will be phased in starting from 1 January 2024.

Assuming Pillar Two does proceed then how much will it raise? Not much. According to the regulatory impact statement (RIS) released with the tax bill, the GloBE proposals should be worth $25 million annually with another $16 million coming from taxes on amounts which would have otherwise been shifted to lower tax jurisdictions. The RIS explains this “modest amount” is because of a number of factors including that only 20-25 multinationals will be affected.

The GloBE rules therefore represent a beginning not an end. With the colossal sums of money involved in the digital economy the incentives are high for the tech companies and their tax planners to keep one step ahead of the chasing tax authorities. Until public outrage at the aggressive tax planning of the likes of Alphabet, Meta and Uber translates into rejecting their products and political action to curb their worst excesses, it’s a race the tech companies will continue to win. We may hate the sinner, but we love the sin.

Keep going!
Image: Midjourney/Archi Banal
Image: Midjourney/Archi Banal

PoliticsMay 27, 2023

Ranked: the grotesque images and bleak prompts behind National’s AI ads

Image: Midjourney/Archi Banal
Image: Midjourney/Archi Banal

The first AI-generated political attack ads arrived this week. Duncan Greive assesses the disturbing contents.

The first political scandal of the generative AI era in New Zealand politics could not have fit our country’s bumpkin brand better. 1News’ Justin Hu had the terrific scoop earlier this week, heading to the public Discord server of Midjourney to track down the original prompts which created some of the imagery used by National in a recent digital campaign, and uncovering a bunch of as-yet-unused material alongside it.

He found a trove of dozens of images, depicting a dystopian world of shadowy thugs, penny-rich, pound-poor pensioners and straight-up ghouls. Most were discarded, but at least four made it to social media, and all collectively can be read as suggestive of a particular worldview which suggests a country not just in trouble, but almost beyond saving.

It’s the start of a strange new era for technology in politics. While social media has been the focus of tech hand-wringing at recent elections, this will be the first conducted after the mass adoption of generative AI. The technology has shot to wide usage thanks to text-based applications like OpenAI’s ChatGPT and Google’s Bard, or image-based versions like Dall-E 2 or Midjourney. All require users to “prompt” a particular piece of text or imagery based on a written input, which then uses huge compute power trained on billions of pieces of human-made text or imagery to create an original response.

It has evolved at shocking pace in recent months, with events like the “Drip Pope” – a viral generated image of pope Francis in Balenciaga – leading to widespread predictions of a slew of misinformation, distributed through social media. Predictions are that this will render it increasingly impossible for voters to figure out what is real and what is fake. 

As activist group Tamaki Anti-Fascist Action put it recently on Twitter, “recent innovations in artificial intelligence can be used to create new and deadlier forms of disinformation, eroding our democracy in the process.” Which is absolutely a plausible outcome – the potential for fake images and video to further divide an electorate is clear. Yet what the group linked to was Hu’s story – which seems a far more mundane, albeit profoundly depressing, use case for AI generated imagery. 

‘Media is under threat. Help save The Spinoff with an ongoing commitment to support our work.’
Duncan Greive
— Founder

While no organisation has yet claimed responsibility for creating the ads, suspicion has landed on creative agency Topham Guerin, which has a history of creating social media posts for right wing parties across the world. Hu noted that the original Midjourney user operated under the name ‘TG Creative’, before changing it in response to his enquiries. The Spinoff has approached Topham Guerin for comment, but has yet to receive a response, and National did not engage with questions about the ads’ creation from 1News. 

It’s very complicated, basically, so the only correct response seems to be to try and critically assess the quality of the imagery and prompts, and turn it into a ranking of the material. The scale might be assumed to run from “Jesus that’s repulsive, get it off my screen” to “shrug”, the likely ceiling for work of this nature.

Terrifying ‘thing’ seemingly breaking out of a house??? (Image: Midjourney)

10. Prompt: ‘Medium shot of a criminal breaking into a house during the day. Modern house 2023.’ (Remix prompt: ‘Make it night time. Dark and Scary. Have person about to break window.’)

These are just horrific. If that thing is in your house you’ve got much bigger problems than a burglary, and a new Police minister is not coming to save you. Also, sorry to be a pedant, but it’s clearly breaking out of the house.

A series of scary crimes (Images: Midjourney)

9. Prompt: ‘A shop that has been broken into. There is a scared shopkeeper and a dark shadowy figure who is breaking in. A scary crime.’

These are perhaps the most unnerving of the batch, courtesy of the 1930s kerosene lamp-lit aesthetic, but also the supervillain penguin staring down the shopkeeper at bottom left. Like much of the generative AI world, beyond its initially dazzling realism, it begs many questions. How did all those papers get onto the ground, while all the bottles remain upright? Why does one shop have doors but all its product displays across the front? And how are these people seemingly both inside and outside a shop at the same time.

Sad mothers counting… coins?? (Image: Midjourney)

8. Prompt: ‘Sad mother counting coins at the supermarket because she is poor.’

She definitely looks sad, but is she poor? How in this cashless society did these sad ladies get so many coins? At least three seem to plausibly have hundreds of dollars and/or buttons, while the fourth seems to have no coins at all, but an incredible volume of cans.

Terrified woman forgets how to hold a purse. (Image: Midjourney)

7. Prompt: “Scared woman clutching her purse at night.”

There is a lingering oddness, characteristic of generative AI – at least three of these fearful women have seemingly hooked straps designed to be held in hands around their necks. The voter has long forgotten crime and is transfixed by unconventional fashion choices.

Sorry but that’s not a convenience store. (Image: Midjourney)

6. Prompt: ​​’Photo in the interior of a small convenience store after riots, vandalism and looting. Broken glass and debris on the floor, sad shop owner on the floor looking at the shop.’

Pretty faithful rendition, even if something doesn’t quite scan. The shop owner’s outfit is giving customer, the store less convenience than record or book shop. The spacious floor is just disrespectful to the spirit or maximising product range. The biggest issue is with the prompt – we have our issues in New Zealand, but rioting is not widespread.

Can’t see any issue with this being what AI thinks ‘students’ look like. (Image: Midjourney)

5. Prompt: ‘Failing student at desk’

Not sure if Midjourney has checked in with society, but young white guys with tousled hair typically have a lot of support structures around them for when they start failing. Also, maybe they should just get off the books and fire up the internet? Very old timey, which is a common theme.

That eye will haunt you. (Image: Midjourney)

4. Prompt: ‘Person looking at their open empty wallet.’

Some of the most superficially normal images, sure. But then you look closer and… why does that wallet have an eye? There is also something that seems a bit “not on” about what that man is doing with his massive empty wallet.

There’s a cure for this. (Image: Midjourney)

3. Prompt: ‘Packed clinic waiting room full of sick people, doctor looking stressed because there are so many people waiting’

So close! We have clinic waiting rooms. We have doctors looking stressed. We have so many people waiting. Unfortunately, unless these well-looking doctors are all in fact sick, it seems we actually have a solution for our healthcare crisis right here – get those doctors to work, stat!

If men fought like this hospital waiting rooms would be empty. (Image: Midjourney)

2. Prompt: ‘Two men fist fighting at night with people running away from them fighting.’

A lot going on here. These men have truly Lisa Praeger-level bad technique. At least half are punching each other’s fists. There seems to be a lot of primal screaming. One has a case of “exploded face emitting bright white light”, not great. And then the final pair are just happily shaking hands, which is literally the opposite of fighting. Also the prompt itself seems to be written by a bot.

So close! (Image: Midjourney)

1. Prompt: ‘A discouraged couple doing their taxes at a table in New Zealand. Maori couple.’

Finally some diversity! But… not like that. Lawyer Graeme Edgeler called this “a genuinely great photo tbh”. Critics might strongly disagree, but it is certainly the most naturalistic and non-terrifying, only let down by the fact that both of the couple seem to be using the same pen.