We are marking the unofficial end of summer here in Canada. But if you work in the AI space at a media company, even if you were spending the warmer days on the beach, you were probably thinking about use cases and what the future may hold as we prepare for 2025. 

Today’s newsletter outlines a few use cases as well as a few thought-provoking scenarios. I hope you find them useful.

Sonali

4 ways news publishers are using GenAI

The New York Times is pleased with the performance of its GenAI ad-targeting experiment BrandMatch. After running a beta test with six brands across tech, finance, luxury, and other categories, the news publication is now inviting other advertisers to try BrandMatch as well.

“BrandMatch works by building personalised targeting segments for every ad campaign, capturing the nuances of a brand and its audience,” a Times spokesperson said. “With this new tool, advertisers can provide their existing marketing briefs, which BrandMatch uses to build personalised ad targeting segments based on relevant articles and the audiences that most engage with those articles.

“BrandMatch provides a solution to a common challenge that marketers face: how to reach specific target audiences as described in their briefs when traditional targeting can only offer a predefined menu of targeting criteria to choose from. The new tool works in a different way than existing ad targeting models, automatically interpreting a brief and matching it with the most relevant context and audiences.”

Meanwhile in Denmark, news magazine Dagbladet’s news magazine Information has built a chat product on top of a new customer database, which has been 10 months in the making.  

“Now I and others in our organisation can chat directly with the underlying tables of data in pure prose,” said Chief Commercial Officer Simon Fancony. “This will save me immense amounts of time and — more importantly — save a lot of time for our already extremely busy data analysts who have spent so much time writing SQL-queries and making dashboards for us non-SQL/data-savvy people to better understand our business in the past. Hopefully, we have now reduced that time significantly.”

The Washington Post has built an AI tool called Haystacker that analysed more than 700 political ads to help with an investigation. 

Haystacker uses the vision capabilities of a large language model. It extracts stills from video files, processes them to on-screen text, and labels the objects present, such as an American flag. The text and visual information are then analysed and verified by reporters. The tool, which took more than a year to build, can be used on any large dataset to find patterns.

And in New Zealand, Stuff Group has created a customised AI tool that helps with the reporting of public documents such as council meeting minutes, submissions, and government reports. Journalists use the Democracy AI tool to scan, prioritise, and report on hyper-local decision-making documents. They still cross-check the content with source documents. 

“Our trials of this GPT tool, which uses official council and government documents as source data to scan for news and decision making, has shown how we can streamline some aspects of the news gathering and writing process, allowing our reporters to focus on polishing and adding depth to the stories,” said Stuff Masthead Publishing Managing Director Joanna Norris. 

The tool can edit stories to varying lengths and formats, and write summaries and headlines, for both print and digital.

At one Stuff publication, The Waikato Times, Editor Jonathan MacKenzie said he was amazed by the speed at which his reporters could work when assisted by AI. 

“We have 11 local authorities in our coverage area so this tool is a win-win for our widely dispersed audience and the newsroom,” he said. “It’s far better for my staff to be out talking to people and digging for stories than stuck behind a desk reading a council agenda.”

Date for the calendar

Wednesday, September 4, 2024: Please join us for a Webinar on insights and inspirations for building GenAI chat products, featuring two prominent names in the business who are doing interesting work: The Washington Post’s Phoebe Connelly and Schibsted’s Martin Schori. Free to INMA members.

AI in journalism: What does the future hold?

What does the future of our information ecosystem look like?

The AI in Journalism Futures project explored how AI could fundamentally transform it over the next five to 15 years. Almost 1,000 people contributed to the project, which used a scenario-planning approach. 

The project outlined five key scenarios, some of which already seem to be underway:  

1. The “machines in the middle” scenario envisions an information ecosystem in which a large portion of journalistic and civic information is gathered, processed, assembled, and distributed via AI. Humans are both the sources and consumers of this information, but AI mediates nearly every process within the information ecosystem, essentially “becoming the newsroom.” 

“This scenario describes an AI information ecosystem that operates largely without a dedicated information producing profession or class — essentially without journalists,” the report said. “This does not necessarily mean that such an ecosystem would operate without editorial oversight, or even without the values and ethical principles of journalism, but that such oversight, values, and principles would be applied via an AI layer between sources of information and consumers of information.”

2. The “power flows to those who know your needs” scenario envisions an information ecosystem in which AI can essentially create any conceivable experience of journalism or information, regardless of format, style, medium, etc., and regardless of the source of information.

The central question is what, specifically, to produce for each individual consumer in every specific consumption situation. 

“Whether such an ecosystem is utopian or dystopian would depend not only on who controls AI but also on who knows what to do with it. This is essentially a situation in which AI knows you better than you know yourself, and therefore it is in your own interests to give up agency to that AI.”

3. The “omniscience for me, noise for you” scenario envisions an information ecosystem in which different individuals and different groups in society experience vastly different information realities because of the different ways in which they engage with AI. One particularly significant situation could be where some people are essentially empowered by AI-assisted information tools, while others are not.

4. The “AI with its own agency and power” scenario envisions an information ecosystem without meaningful human oversight in which very powerful AI systems control the gathering and experience of information for most people. 

“This scenario is not a ‘Terminator’-style takeover of human societies by superintelligent machines, and it does not assume any kind of sentience or consciousness within AI systems. Instead, it describes a more nuanced situation in which people — consumers, engineers, editors, or executives — gradually give up more and more agency to adaptive AI systems until humans no longer control those systems in any meaningful way,” the report said. 

5. The “AI on a leash” scenario envisions an information ecosystem in which the potential impact of AI has been substantially restricted by societies or by the collective action of consumers. 

Some key themes that emerged:

• Personalisation of information, presented as both an end-state and as a driving force leading to other end states. For example, new capabilities to tailor content in multi-modal formats, language, literacy level, etc., according to user preference, could create an environment where information is ubiquitously highly personalised. This hyper personalisation then leads to multiple and often conflicting information realities and filter bubbles, as well as increasing audience fragmentation.
• Information and mis/dis/mal-information at mass scale. 
• The rise of AI agents (intelligent systems that perform autonomous tasks) and AI assistants (a user-facing system that performs tasks, often directed through a conversational interface). 
• The rise of both individual human and machine influencers, creators, personalities, and celebrities as user-facing distribution channels. 
• Inability of legacy news media to adapt to changes. You can almost hear the authors of the report sighing as they write: “There was considerable skepticism about the ability of traditional journalism institutions to adapt successfully to an AI-driven future, and participants tended to envision coming changes in terms of power shifts away from journalists, with little attention to how those changes might increase or decrease value for audiences.”

Worthwhile links

• GenAI and reporting: A reporter used GenAI to write stories; their newspaper did not have an AI policy.
• GenAI and reporting II: This news site uses 14 different bots to produce automated news about events. 
• GenAI and trust in news: Audiences see some positive aspects of using GenAI in news, though there is still plenty of suspicion.
• GenAI and not-for-profit news: It’s used for personalised fundraising campaigns, among other things.  
• GenAI and election disinformation: It’s not particularly worse than it was pre-GenAI.
• GenAI and search: Reddit gets in on the game with AI-powered overviews.
• GenAI and search II: Google AI Overviews are now in six more countries.
• GenAI and news distribution: Perplexity partners with a crypto company to distribute news summaries.
• GenAI and software development: It works well for repetitive tasks and debugging but not for complex tasks.
• GenAI and theft: Steal content first, clean up later, says former Google CEO Eric Schmidt.
• GenAI and attachment: OpenAI warns that some humans are becoming too involved with its bot. It has also been called “addictive intelligence.”
• AI and customer service: A Japanese company is using AI to monitor its customer-facing employees’ smiles and tone of language. 

About this newsletter

Today’s newsletter is written by Sonali Verma, based in Toronto, and lead for the INMA Generative AI Initiative. Sonali will share research, case studies, and thought leadership on the topic of generative AI and how it relates to all areas of news media.

This newsletter is a public face of the Generative AI Initiative by INMA, outlined here. E-mail Sonali at sonali.verma@inma.org or connect with her on INMA’s Slack channel with thoughts, suggestions, and questions.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .