New Zealand’s Stuff media group has joined other leading news organisations around the world in restricting Open AI from using its content to power artificial intelligence tool Chat GPT.

A growing number of media companies globally have taken action to block access to Open AI bots from crawling and scraping content from their news sites.

Open AI is behind the most well-known and fastest-growing artificial intelligence chatbots, Chat GPT, released late 2022.

“The scraping of any content from Stuff or its news masthead sites for commercial gain has always been against our policy,” says Stuff CEO Laura Maxwell. “But it is important in this new era of Generative AI that we take further steps to protect our intellectual property.”

Generative Artificial Intelligence (Gen AI) is the name given to technologies that use vast amounts of information scraped from the internet to train large language models (LLMs).

This enables them to generate seemingly original answers — in text, visuals or other media — to queries based on mathematically predicting the most likely right answer to a prompt or dialogue.

Some of the most well-known Gen AI tools include Open AI’s ChatGPT and Dall-E, and Google’s Bard.

Surge of unease
There has been a surge of unease from news organisations, artists, writers and other creators of original content that their work has already been harvested without permission, knowledge or compensation by Open AI or other tech companies seeking to build new commercial products through Gen AI technology.

“High quality, accurate and credible journalism is of great value to these businesses, yet the business model of journalism has been significantly weakened as a result of their growth off the back of that work,” said Maxwell.

“The news industry must learn from the mistakes of the past, namely what happened in the era of search engines and social media, where global tech giants were able to build businesses of previously unimaginable scale and influence off the back of the original work of others.

“We recognise the value of our work to Open AI and others, and also the huge risk that these new tools pose to our existence if we do not protect our IP now.”

There is also increasing concern these tools will exacerbate the spread of disinformation and misinformation globally.

“Content produced by journalists here and around the world is the cornerstone of what makes these Gen AI tools valuable to the user,” Maxwell said.

“Without it, the models would be left to train on a sea of dross, misinformation and unverified information on the internet — and increasingly that will become the information that has itself been already generated by AI.

Risk of ‘eating itself’
“There is a risk the whole thing will end up eating itself.”

Stuff and other news companies have been able to block Open AI’s access to their content because its web crawler, GPTBot, is identifiable.

But not all crawlers are clearly labelled.

Stuff has also updated its site terms and conditions to expressly bar the use of its content to train AI models owned by any other company, as well as any other unauthorised use of its content for commercial use.

Earlier this year The Washington Post published a tool that detailed all major New Zealand news websites were already being used by OpenAI.

OpenAI has entered into negotiations with some news organisations in the United States, notably Associated Press, to license their content to train ChatGPT.

So far these agreements have not been widespread although a number of news companies globally are seeking licensing arrangements.

Maxwell said Stuff was looking forward to holding conversations around licensing its content in due course.

Article by AsiaPacificReport.nz

I was concerned when the story broke last month about inappropriate subediting by RNZ staff of ‘wirecopy’ from international sources such as Reuters. The wire-tampering story broke with particular reference to stories about the war in Ukraine; and, at least for that story, it needs to be understood that Aotearoa New Zealand is an aligned party to that military conflict, so certain sensitivities will apply.

I was then concerned when RNZ chief executive Paul Thomson called the RNZ subedits “pro-Kremlin garbage”. For background see Mediawatch: Further fallout as RNZ takes out the ‘Kremlin garbage’, Evening Report, 18 June 2023. For a senior professional communicator, the RNZ CE set a particularly bad example.

Subsequently, RNZ has undertaken an audit of stories published on its website, so its possible to check out the bias of the sub-edits. It turns out that there is a clear anti-Washington rather than pro-Kremlin sub-editorial line. A number of the stories brought to light – and corrected – relate to Latin America; in addition to stories featuring Ukraine, China, Taiwan, Israel and Ireland. (I have heard it said that the sub-editor in question is not only pro-Kremlin, but also has a disposition towards anti-democratic regimes. I cannot agree; I would assess the sub-editor in question to be an old-fashioned democratic left-winger who, in Cold War times, might once have had some pro-Soviet sympathies.)

Before looking at specific themes of the sub-edits, I present the following quote (8’20”) from Mediawatch, RNZ’s Russiagate, 14 June 2023.

The programme features Hayden Donnell talking to RNZ’s Anna Thomas about the purpose of subediting a “pre-subbed” wired story from an international news agency: “It’s already gone through a pretty robust process at Reuters or AP or wherever you’re sourcing it from. Most of the time it’ll just require an editor formatting it to in-house style, maybe removing some Americanisms, cutting it to length, and plonking it on the website.”

And then: “What can [ie should] you edit with wirecopy? Even if you agree with this person’s edit, the heart of the issue is that you cannot take copy and make substantive changes to its meaning. But you can add context, you can delete sections for length, you can insert relevant local information or quotes. If you cannot make any changes at all, that’s untenable.” [I have sub-edited bits of this second quotation to shorten it, to remove repetition, and to make it more like written rather than spoken language.]

The problem is that even very small additions, deletions and substitutions can subtly alter the meaning of a text. That’s of course a problem here, and it is clear that there has been an intent to steer the meaning in an anti-Washington direction. By way of contrast, disinterested subediting will be like a ‘random walk’ [a statistical concept] meaning that, on average, altered meanings are unbiased. Subeditors who are close to an issue may display unconscious bias, whereas outsourced subeditors (including robotic subeditors) who are distant from the issues in a text may be unbiased but ‘noisy’; such subeditors will on average make more mistakes, and will struggle to appreciate nuance in a text.

While the problem subeditor in question was clearly inserting an anti-Washington bias, his defence may well have been that he [other media stories refer to ‘he’] was correcting a pro-Washington bias in the material he was working on. Certainly, in any Goodies versus Baddies narrative – inherent in war stories – academic or journalist disinterest is largely absent from most stories; these are narrational contexts where a person who is not overtly on one side is too easily characterised as being on the other side. As the question goes: ‘Which side are you on?‘.

Editorial biases are commonly worse than sub-editorial tampering. These in particular involve the decision whether or not to run a story. While these are often dictated by the fast-moving news cycle – meaning that stories about Covid19, for example, were biased towards the beginning of that pandemic, and created an ‘exceptionalism’ towards that disease at the expense of contextual discussion and other health risks – they also reflect self-censorship (partly but not only because of the fear of the wrath of authorities or other power-brokers).

Another form of bias arises in the need to create headlines which will draw readers to a story; a bias compounded by the fact today that most stories have ‘click-bait’ headlines (hyperlinks) which are even more sensational and less qualified than the actual headlines to the stories.

‘Loaded’ Language

Consider this story: Organization of American States head ‘one of worst in history’ – Ebrard. The changes made and then unmade are listed at the end of the story.  With respect to former Bolivian president Evo Morales, the mischievous subeditor replaced “resigned under pressure” with “resigned and fled under threat”. Both versions are essentially true, though the original (and restored) version may have understated the danger Morales faced; or perhaps the modified version overstated the danger.

We also see, in that story, the clause “a presidential vote that the OAS said was rigged” was changed (and unchanged) to “a presidential vote that the OAS claimed was rigged”. This leads to the issue of the degree to which some synonyms are more ‘loaded’ or ‘accusative’ than others. (Note here that if the original story had used the word ‘claimed’, there would have been no issue; the question is the motive of the subeditor in making the change.)

Aside

A common sort of story takes the form ‘A abuses B’, where ‘to abuse’ means any action that is in some sense ‘bad’. Consider this story, about the November 1975 regime change in Australia (commonly known there as ‘The Dismissal’). The allegation is of Washington involvement in precipitating this particular political crisis.

This is an A abuses B story, where (in this case) ‘A’ is the American CIA, ‘abuses’ means ‘dismisses’, and ‘B’ means ‘the elected government of Australia’. The story at its most disinterested level is [passive voice] that ‘The Government was dismissed’. In the active voice, the most neutral version is that ‘sources said that the Dismissal was instigated by the CIA’. The next level would be ‘sources claimed that the Dismissal was instigated by the CIA’. Up another notch would be ‘The CIA allegedly instigated the Dismissal’, or [passive voice] ‘The CIA was accused of instigating the Dismissal’. Finally, the most overt form is the unqualified ‘The CIA instigated the Dismissal’.

In the various stories we read and hear, many which are in the ‘A abuses B’ form, we will encounter the full linguistic range from neutral (‘something bad happened’) to the presentation of an accusation as a fact. Actually, the way a story is narrated is ‘rhetoric’; and neutral rhetoric can be a way to intentionally downplay something, just as accusative rhetoric upplays that same story.

Back to the Main Narrative

We see this in this RNZ story, Two activists involved in land dispute killed in Brazil: police, in which the restored headline is in the passive voice and the word ‘said’ is only implied. The inappropriately sub-edited version is in the active voice with the abused named without ‘alleged’ as a qualification: ‘Death squad shoots dead two Brazilian land activists’.

This story Chile passes bill to boost taxes on rich, spur investment, small business shows that the subversive  RNZ sub-editor is coming from a somewhat conventional left-wing perspective, and not from an autocratic ‘far-right’ Russian perspective. People who are anti-inequality don’t usually regard Russia these days as an exemplar country.

This story for a while contained an inserted and unqualified allegation of a “2014 US-based coup”:Serbia accuses Ukraine and unnamed EU country of Air Serbia bomb hoaxes. It’s an example that shows the anti-Washington stance of the sub-editor. Indeed, articles like these are not the correct place to debate the extent of United States’ involvement (or otherwise) in the regime-change event in Ukraine in February 2014.

In this story we see the explicit anti-Washington subeditorial stance with respect to China over Taiwan, and also the more neutral word ‘says’ preferred by the subeditor over the word ‘worries’ with respect to Japan: South Korea’s president seeks closer Tokyo ties after latest North Korea missile launch. Yet this story’s subediting uses the rhetorical word ‘blunders’ with respect to China, not exactly an endorsement of Beijing.

Finally

I would regard Paul Thomson’s use of the rhetorical word ‘garbage’ to be more problematic than the sub-editors’ word ‘blunders’. Garbage is ‘waste’, not ‘lies’. Waste is a reality of life that should be regarded as normatively neutral, not wicked. In ecology and sustainable economics, waste is indeed a ‘good’, not a ‘bad’; an input as well as an output. It is not professional to oppose bad rhetoric with worse rhetoric.

And, I wonder if the mischievous subeditor has a point in interpreting much of the copy that came his way as having its own bias. If the generative AI chatbot ChatGPT was trained only on copy acceptable to today’s western authorities and power-brokers, would the bot’s outputs really be any more truthful than the ‘pro-Kremlin garbage’ that a frustrated socialist RNZ minion was (for a brief while) turning out?

*******

Keith Rankin (keith at rankin dot nz), trained as an economic historian, is a retired lecturer in Economics and Statistics. He lives in Auckland, New Zealand.

When New Zealand’s opposition National Party’s transport spokesperson, Simeon Brown, questioned the logic of bilingual traffic signs, he seemed to echo his leader Christopher Luxon’s earlier misgivings about the now prevalent use of te reo Māori in government departments.

Genuine concern or political signalling in an election year? After all, Luxon himself has expressed interest in learning te reo, and also encouraged its use when he was CEO of Air New Zealand.

He even sought to trademark “Kia Ora” as the title of the airline’s in-flight magazine.

And for his part, Brown has no problem with Māori place names on road signs. His concern is that important messaging about safety or directions should be readily understood. “Signs need to be clear,” he said.

“We all speak English, and they should be in English.” Adding more words, he believes, is simply confusing.

It’s important to take Brown at his word, then, with a new selection of proposed bilingual signs now out for public consultation. Given the National Party’s enthusiastic embrace of AI to generate pre-election advertising imagery, one obvious place to start is with ChatGPT, which tells us:

Bilingual traffic signs, which display information in two or more languages, are generally not considered a driver hazard. In fact, bilingual signage is often implemented to improve safety and ensure that drivers of different language backgrounds can understand and follow the traffic regulations.

ChatGPT also suggests that by providing information about speed limits, directions and warnings, bilingual traffic signs “accommodate diverse communities and promote road safety for all drivers”.

‘They should be in English’: National to ditch te reo Māori traffic signs https://t.co/7FGYyQDrPu

— RNZ News (@rnz_news) May 26, 2023

Safety and culture
With mounting concern over AI’s potential existential threat to human survival, however, it’s probably best we don’t take the bot’s word for it.

Fortunately, government transport agency Waka Kotahi has already examined the use of bilingual traffic signs in 19 countries across the Americas, Asia, Europe and the Middle East. It’s 2021 report states:

The use of bilingual traffic signage is common around the world and considered “standard” in the European Union. Culture, safety and commerce appear to be the primary impetuses behind bilingual signage.

Given Brown’s explicit preference for the use of English, it’s instructive that in the UK itself, the Welsh, Ulster Scots and Scots Gaelic languages appear alongside English on road signs in Wales, Northern Ireland and Scotland.

More to the point, on the basis of the evidence it reviewed, Waka Kotahi concluded that — providing other important design considerations are attended to — bilingual traffic signs can both improve safety and respond to cultural aspirations:

In regions of Aotearoa New Zealand where people of Māori descent are over-represented in vehicle crash statistics, or where they represent a large proportion of the local population, bilingual traffic signage may impart benefits in terms of reducing harm on our road network.

‘One people’
Politically, however, the problem with a debate over bilingual road signs is that it quickly becomes another skirmish in the culture wars — echoing the common catchcry of those opposed to greater biculturalism in Aotearoa New Zealand: “We are one people”.

It’s a loaded phrase, originally attributed to the Crown’s representative Lieutenant Governor William Hobson, who supposedly said “he iwi tahi tātou” (we are one people) at the signing of the Treaty of Waitangi in 1840.

Whether or not he said any such thing is up for debate. William Colenso, who was at Waitangi on the day and who reported Hobson’s words, thought he had.

But Colenso’s account was published 50 years after the events in question (and just nine years before he died aged 89).

Either way, the assertion has since come to be favoured by those to whom the notion of cultural homogeneity appeals. It’s a common response to the increasing public visibility of te ao Māori (the Māori world).

But being “one people” means other things become singular too: one law, one science, one language, one system. In other words, a non-Māori system, the one many of us take for granted as simply the way things are.

Any suggestion that system might incorporate or coexist with aspects of other systems — indeed might benefit from them — tends to come up against the kind of resistance we see to such things as bilingual road signs.

Fretful sleepers
The discomfort many New Zealanders still feel with the use of te reo Māori in public settings brings to mind Bill Pearson’s famous 1952 essay, Fretful Sleepers.

In it, Pearson reflects on the anxiety that can seep unbidden into the lives of those who would like to live in a “wishfully untroubled world”, but who nonetheless sense things are not quite right out here on the margins of the globe.

Pearson lived in a very different New Zealand. But he had his finger on the same fear and defensiveness that can cause people to fret about the little things (like bilingual signs) when there are so many more consequential things to disrupt our sleep.

Anyway, Simeon Brown and his fellow fretful sleepers appear to be on the wrong side of history. Evidence suggests most New Zealanders would like to see more te reo Māori in their lives, not less.

Two-thirds would like te reo taught as a core subject in primary schools, and 56 percent think “signage should be in both te reo Māori and English”.

If the experience in other parts of the world is anything to go by, bilingual signage will be just another milestone on the road a majority seem happy to be on.

Dr Richard Shaw, Professor of Politics, Massey University.  This article is republished from The Conversation under a Creative Commons licence. Read the original article.

Article by AsiaPacificReport.nz

I wasn’t good at French in my final year of high school. My classmates had five years of language studies behind them. I had three. As a result of my woeful grip on the language, I wrote a terribly bad essay in my final French exam.

The more I read of ChatGPT’s output, the more I am reminded of my final French essay. I could not express the complex ideas I wrote in my English essays, so instead, I repeated the question a lot and clumped together words and phrases that sounded like they kind of went together. There was no logical thread, no cogent argument.

It was a bit like the perplexing, digressive, buzzword-rich oratory stylings of Donald Trump.

I have been a university lecturer, tutor and marker for coming on two decades now and late last year a student submitted an essay that I sent off to the university integrity team, explaining that it was “bad in a new and different way”.

According to Turnitin (our detection software), it wasn’t plagiarised. It didn’t read like it had been written in another language and run through Google Translate. The grammar was impeccable but there were glaring non-sequiturs and it danced around the question, which it repeated several times, but didn’t actually answer.

I didn’t hear back from the integrity people. They probably didn’t know what to do about it and may have been busy, as it was the end of the teaching year. I had also said it wasn’t urgent, as it had failed against my marking key, meaning the student, whose marks had been poor all along, would have to repeat the unit anyway.

New teaching year
A couple of weeks later ChatGPT was made available to the public, joining the dozen or so other AI writers available to people who want AI to string together their sentences.

Now, heading back into a new teaching year, having spent the summer chatting with ChatGPT, I am in conversations with my colleagues about how we should proceed. I teach journalism and my colleagues are from a range of arts and communications disciplines.

Collectively our feelings are mixed, but I’m looking forward to letting my students know about this leap forward in communications technology.

I plan to explain it in the context of the other leaps and lurches I’ve lived through.

This won’t be the first to make swathes of workers redundant. I remember the angst in my industry about digital typesetting usurping the compositors and typesetters, replacing vast numbers of them with far fewer graphic designers.

ChatGPT will undoubtedly take some jobs, but it’s the donkey work of the writing professions. It frees us up to do the innovative fun stuff. Also, while ChatGPT is big and shiny, we’ve known that AI writing is on its way for a long time.

In 2018, Noam Lemelshtrich Latar summed up the progress in our field to date in his book Robot Journalism: Can human journalism survive? He documented the many workplaces already using AI writing software and concluded that there was still work to be done. There still is.

Essay capacity underwhelming
Much of the media racket over ChatGPT this summer has been about its capacity to write essays, and so I have read several essays it has written, and I can happily report that I am underwhelmed by them, but also fascinated by the challenge we face in getting better at describing the ways in which they are bad.

This task is part of the mission humanity more broadly is facing in figuring out what it is that people can do that robots can’t. If robots/AI writers are going to do the donkey-work writing in workplaces, that is not something we need to be training graduates to do.

Graduates need to be able to do things an AI language model can’t, and they need to be able to articulate their skill sets.

So, I will be generating AI content in my classrooms and we are going to set to work pulling it apart, in search of its failings and foibles. We’ll do this together and learn about it and ourselves as we go.

Another big theme in the media hype has been ChatGPT’s ability to “do the marking for us”. This, in my opinion, is rubbish. Sure, you can copy-and-paste some text into ChatGPT and ask it for a comment and a grade, but every university I know of demands more of the markers than a simple comment and grade.

If only it was that simple. But, no. We have to describe the specific criteria every piece of work will be assessed against, and the expectations ascribed to each criterion that will result in the award of a specific number of marks. This forms a table called a rubric, which is embedded in our unit websites and getting the assignments and rubrics out of that software and into ChatGPT would take longer than the tight time allocation we get to mark each piece.

Besides the software we mark in is already replete with time-saving tricks, like a record function so you can speak rather than type feedback and the ability to save commonly used comments.

‘Getting to know students’
In addition, failing to read the assignments would inhibit the “getting to know your students” process that marking their work facilitates, and so I imagine it to be the sort of drain-circling behaviour used by failing teachers on their way out of the profession — as student assessment of teachers who cheat in their marking is going to be on par with teacher assessment of students who cheat in their assessments.

Cheating is a key word here. While ChatGPT is new, universities have longstanding policies and charters that use words like “honesty and fairness” in relation to academic integrity. These are being underscored and highlighted in preparation for the start of semester and hyperlinked to paragraphs about AI writing.

Honest use of ChatGPT will involve disclosure about how it was used, and what measures have been taken to verify its content and iron out its wrinkles. It then joins the swath of online tools we encourage our students to use to prepare them for the professions they’ll enter when they graduate.

For my first year students these will be professions that have adjusted to the existence of AI language models, and so their new graduate brilliance will need to be focused on things only humans can do to make the world a better place. This is how I’m going to frame it in my classes, when our next semester starts.

Dr Kayt Davies is a lecturer in journalism at Curtin University. She is a contributor to Pacific Journalism Review. The article was first published in The West Australian and is republished by Asia Pacific Report with the author’s permission.

Article by AsiaPacificReport.nz