AI is transforming journalism, but Balkan affiliates, meeting in Podgoricia, Montenegro, have committed to taking up the challenge as an industrial issue - whatever solutions legislators propose.

Sara Ai presents broadcast news in Montenegro . She is an artificial intelligence (AI) propelled chat bot. Although not the most animated broadcaster, scrip fumbles, off piste remarks, and private life misjudgements never trouble her performance. So visiting the Montenegro to deliver training for journalists' union leaders from this country, Serbia and North Macedonia about navigating the issues of AI, I fear accusations of arriving too late.

Miloš Panić who presents radio sports commentary in Serbia is quick to reassure. "Journalists in the region have know almost nothing about AI, particularly those of us working the legacy media" he tells the group of around 20, assembled as part of the TADAM project.

The group express the mix of hopes and anxieties about how AI might affect our trade that are common among journalists. Transcribing, translating, and summarising seem like the miracle solutions for which reporters have been waiting, particularly in areas outside the global linguistic blocks. The potential for deskilling, falling headcounts and a tsunami of confected content that undermines trust in professional news platforms are the worrisome counterweight.

Technical instruction comes from Tomaz Hollaek and Raphael Hernandes from Cambridge University's Leverhulme Centre For The Future Of Intelligence. They challenge what we know about the sources on which the 'large language models' on which AI depends have been 'taught'.

Having established that few of us have much understanding, the academics guide us to the 'model card' or 'system card' documents that should be easily available for any facility based on a large language model. These, we learn, are long, narrative documents setting out in detail what information has been used to 'teach' the AI models that set out the steps taken to ensure that it is fair, accurate, without bias and has not violated copyright laws.

Simple searches are sufficient to turn up, and interrogate, these records.

On a warm afternoon, taking in 32 pages of modestly technical language was a it of a stretch. Knowing that these documents exist and can be dissected would appear to be a minimum expectation of those who might use generative AI in a professional environment, however. Familiarity with them immediately confers a better understanding of the process by which AI magic is performed.

My role was to explain the extraordinary panoply of initiatives currently in play to regulate AI. Until the launch of ChatGPT in November 2022, AI had attracted little attention beyond specialist circles. Indeed, in the years since, say 1997, when IBM's Deep Blue computer beat chess grand master Gary Kasperov there have been periods when AI research was staved of funding and was considered a marginal area of computing research.

ChatGPT changed all that. Within three months it was the most downloaded piece of software in history. Not long after that, tens of thousands of AI tech professionals, including Elon Musk and Steve Wozniak demanded a six month moratorium on development of this unknown technology that 'might prove a threat to humanity itself."

That did not happen but national leaders were spurred to so many initiatives to tame AI that I couldn't fit them all into a training session. They include: President Biden's executive order, the G7's Hiroshima Process, the UK government's Bletchley Declaration , the EU's AI Act, and a recent initiative by the Council of Europe.

Creators' organisations whose focus is EU policy are currently gearing up for a fresh consultation on how the EU's new Act will be implemented.

I pointed out that well intentioned, and potentially useful, as most of these were, none have had much actual impact on the deployment of AI as it affects workers in general and journalists in particular. On the other hand, the strikes mounted in Hollywood by the US Writers Guild, and SAG-AFTRA have immediately and beneficially changed the terms of trade for many of those working in film and television.

However it is produced, every written work used in US film and television will be the responsibility of an actual human writer, who will be paid. Actors will be able to control when AI simulations of their likeness are used and will be paid accordingly. In Tinsel Town at least, AI will not be used to replace human talent.

SAG-AFTRA's lead negotiator, Duncan Crabtree-Ireland's remarks to an awards ceremony were well made. "The result (of our strike) is a ground-breaking agreement. It sets a vision for the coming decades and it takes huge steps to defining how AI becomes a part of our world, putting necessary guardrails in place. It is not only important for us, but it was also important for the rest of the industry, and frankly for everybody".

Technology is moving at great pace, I told my classroom of presidents and general secretaries. Legislation and litigation will always struggle to keep up. An agreement between workers and management can be settled in an afternoon, and varied quickly should further advances necessitate.

By the time we all set off for home, the Balkan unionists seemed enthused. Commitments were made to informing their members about the prospect of AI in the newsroom. And I hope they were persuaded that unions have a role in steering its implementation.

Waiting for my plane, I watched one of Sara AI's broadcasts. It struck me that for all her flawless skin, and unruffled hair, she was only a sock puppet. Technology may have replaced the puppeteer's hand, but her reports are the product of human decisions. For so long as that is true, unions have a job to do organising those decision makers.

Tim Dawson is the IFJ's deputy general secretary

Published

15 October 2024

• Twitter
• LinkedIn
• Facebook