Digital Innovation Digital Publishing Top Stories
7 mins read

“Buckle up for the ride”: What publishers need to know about AI, from Reuters Institute

“Its speed and capabilities are awe-inspiring and frightening at the same time.”

We are on the cusp of a new wave of disruption as artificial intelligence technologies start to impact on the real world, driving greater efficiency and automation on the one hand, but also enabling content to be remixed in surprising and unpredictable ways.

“The next wave of technical innovation is already here”, according to the Journalism, Media, and Technology Trends and Predictions 2023 report by Reuters Institute.

And it’s not talking about the metaverse. Extraordinary advances in artificial intelligence (AI) in 2022 have laid bare more immediate opportunities – and challenges – for journalism. AI offers the chance for publishers (finally) to deliver more personal information and formats, to help deal with channel fragmentation and information overload. 

These new technologies will also bring existential and ethical questions – along with more deep fakes, deep porn, and other synthetic media. Buckle up for the ride.

Nic Newman, Author, Journalism, Media, and Technology Trends and Predictions 2023

Against this backdrop, news organisations that have not yet fully embraced digital will be at a severe disadvantage. According to the report, the next few years will not be defined by how fast we adopt digital, but by how we transform our digital content to meet rapidly changing audience expectations.

Here’s what the report has to say on the impact of AI on publishing, in 2023 and beyond:

Publishers are “quietly integrating AI”

Media companies are quietly integrating AI into their products as a way of delivering more personalised experiences. 

Almost three in ten (28%) say this is now a regular part of their activities, with a further 39% saying they have been conducting experiments in this area. 

New applications such as ChatGPT and DALL-E 2 also illustrate opportunities for production efficiency and the creation of new types of semi-automated content.

AI voices are getting smarter

Artificial intelligence (AI) voices are getting smarter and it is now possible to clone a journalist’s voice with extraordinary accuracy. Aftenposten, one of Norway’s largest news publications, recently cloned the voice of its podcast host using AI technology while News24 in South Africa has also trained its systems with the voice of a popular actor for its news and feature stories.

Examples of audio-based news
Source: Reuters Institute

Breakthrough year for artificial intelligence 

AI chatbots have been widely and justifiably mocked in recent years, but the arrival of ChatGPT, from OpenAI, has transformed the debate. Its speed and capabilities are awe-inspiring and frightening at the same time. While the underlying models have been around for some time, ChatGPT has turned these into an accessible prototype that gives a real sense of where AI may be heading. 

It can tell jokes (but has been trained not to tell racist or sexist ones), come up with plots for a film or book, write computer code, and even summarise the challenges facing local journalism in a few sentences (below).

GPTChat example
Source: Reuters Institute

Biggest technological advance since the invention of the internet?

Some view ChatGPT as one of the biggest technological advances since the invention of the internet and is part of a wider trend called ‘generative AI’ that enables computers to create not just words but also pictures, videos, and even virtual worlds from just a few text prompts.

Here is an image of a journalist, in the style of a Raymond Chandler novel, filing a story from a Pacific Beach using a laptop – created in seconds using the AI tool MidJourney.

Example of AI picture
Journalist files a story from the beach with a cocktail: Rendered by MidJourney | Source: Reuters Institute

Generative AI enables computers not just to make existing processes more efficient, but use a range of existing assets to create something new. For the journalist in that Raymond Chandler novel, this raises existential questions but also opens up a range of new possibilities.

AI in 2023

This year we’ll start to see more of these tools being opened up to creators, journalists, and others, allowing us to create new versions of ourselves, of others, and the world around us. 

Magic avatars will take over timelines 

Lensa is an app that allows you to magic avatars of yourself and remove unwanted objects from any picture with ‘no skills required’. These apps have already been criticised for stealing from artists, using predatory data-sharing practices, and promoting sexualised stereotypes, but that won’t stop them taking over social media timelines this year.

Lensa examples
Left: Nic Newman in different guises (Lensa), Right: Example images from another AI Art app (Facetune) | Source: Reuters Institute

The implications for journalism are not entirely clear but tools like MidJourney and DALL-E are already being used to create illustrative art for articles and blog posts. 

Semafor has used AI to animate the Ukraine war, but similar tech can be used for deep fakes

Examples of use of AI
Source: Reuters Institute

All this is likely to lead to an explosion of automated or semi-automated media in the next few years – for good or ill (the research firm Gartner estimates it will account for 25% of all internet data). It will be easier than ever to create ‘good looking’ and highly plausible multimedia content, but it will also be harder than ever to separate what is real from what is fake, misleading, or doctored.

Other ways in which AI is being used by publishers

In the survey, news executives talk about different ways in which they are using AI technologies such as Machine Learning (ML) and Natural Language Processing (NLP) to make existing production processes more efficient: ‘Summaries, text to speech, and image recognition leading to automated tagging and subtitles’, are some of the examples mentioned by Mathieu Halkes, Head of Product at Schibsted. ‘[There are] more and more use cases we see and apply every day’, he says.

AI transcription

AI transcription tools are now routine in newsrooms, with Danish digital-born outlet Zetland developing a speech-to-text transcription service aimed specifically at journalists and designed to work with smaller languages that big corporate products have failed to support. 

Good Tape is built on top of OpenAI technology and is currently free to use. Meanwhile, in Finland, the public broadcaster Yle has been able to start a service for Ukrainian refugees with news being automatically translated by machine before being checked by a native speaker. During the pandemic, Yle was also able to provide information in Somali, Arabic, Kurdish, and Persian.

Better personalisation and improved content recommendations

Others hope that AI can help deliver better personalisation and improve content recommendations to help increase engagement. The Sophi tool was developed by the Globe and Mail in Canada, where it has automated the vast majority of its web homepages, allowing editors’ time to be used more productively and driving a 17% increase in click-through rates. The product is now being offered to other publishers. AI-driven tools like Sophi are also used to manage social distribution tasks such as headline optimisation and the best time to post.

When it comes to recommendations, around a quarter (23%) of respondents say they are now using AI regularly, with 5% of early adopters making it a big part of what they do.

Automated homepages, new languages, and transcription

News organisation screenshots
Source: Reuters Institute

The Newsroom is a start-up (still in beta) which uses AI to automatically identify and write summaries of the top news stories of the day, as well as summarising background context and providing links to related stories that are clustered by political perspectives. Although AI does the heavy lifting, all copy is checked and if necessary modified by a journalist.

Stories written by AI but checked by humans with automated context

The debates over automation in journalism are not straightforward. Many welcome the capability to make non-journalistic tasks more efficient, but at the same time worry that cheaply produced synthetic media and semi-automated content could further commoditise news and undermine trust. 

One respondent from a leading quality news company argues that in these circumstances human curation becomes an even more important differentiator:

We want to apply AI fundamentally to enhance manual curation not to replace it. All our work in this area is rooted in an understanding of the value manual curation brings to our readers and how it differentiates us from platforms with a host of problems around automation and curation.

AI and publishing: What else might happen?

Broadcast companies embrace virtual presenters

Deep Brain AI, a technology company based in South Korea, creates digital copies – or digital twins – of popular TV news anchors and these now make regular appearances on mainstream channels in Asia. MBN and Arirang in Korea and BTV and CCTV in China are using the technology to help save costs and enhance the presence of the most popular presenters. The company is now looking for customers in the United States where TV companies are under pressure to do more with less. 

Pictures: Deep Brain AI | Source: Reuters Institute

One likely ‘use case’ is for on-demand weather, where an AI model can be created of a popular forecaster including their favourite phrases and expressions, and then updated videos can be created for any location whenever the underlying data change. These models can also be combined with ChatGPT functionality to create a virtual chat bot answering questions about an election, for example, by a political correspondent’s digital twin.

Debate over regulation of AI hots up

As these opportunities become more real, so do the ethical and regulatory dilemmas. Deep fakes have already been used to create non-consensual pornography, commit fraud, and fuel disinformation campaigns. 

Discussion about regulation is ongoing, and the EU is proposing an AI Act that would ban ‘unacceptable’ uses of applications that violate people’s fundamental rights and safety – even if in practice these will be hard to identify and enforce. In the meantime, journalism could take the lead in making its use of AI more transparent. 

We can expect more publishers to publish ethical guidelines covering the key areas which range from photo-improvement or manipulation to transparency and copyright. Open AI is working on digital watermarking responses and better labelling that could also help build trust in positive uses of these technologies.

“On the cusp of a new wave of disruption”

The report underscores that we are on the cusp of a new wave of disruption as artificial intelligence technologies start to impact on the real world, driving greater efficiency and automation on the one hand, but also enabling content to be remixed in surprising and unpredictable ways. 

This will help media companies do more with less, as well as open up opportunities in the creation and distribution of smarter content. But it will also bring new dilemmas about how these powerful technologies can be used in an ethical and transparent way.

Nic Newman, Author, Journalism, Media, and Technology Trends and Predictions 2023

The full report can be downloaded from Reuters Institute:
Journalism, Media, and Technology Trends and Predictions 2023