Beyond the Hype: AI-accelerated news creation and delivery

Nearly a year and a half has passed since ChatGPT burst onto the digital media scene sparking excitement and alarm in equal measure.

In that time the strengths and weaknesses of GenAI technologies have come more closely into focus. While interesting theoretical debates have sprung up on the nature of AI and its future potential, the practical implications for the news-media sector have become much clearer.

In particular we are now in a good position to answer two questions:

• In what ways can GenAI tools accelerate the production and delivery of news content?
• How should work be divided between human and AI agents in order to optimize both the quality as well as the quantity of the news content produced?

These two questions were the focus of a recent webinar hosted by platform producers Eidosmedia with the participation of CMO & CPO Massimo Barsotti and Mario Garcia, Senior Advisor on News Design at the University of Columbia School of Journalism.

How much can you automate?

In a recent keynote address, professor of computational journalism Nick Diakopoulos identified ten levels of progressive automation of the news creation process, from fully manual to entirely computer controlled.

He then asked the model, ChatGPT 4 itself, to identify how many of an extensive list of journalistic tasks from idea generation to audience analytics could be partly or fully automated while maintaining editorial standards.

For the majority of tasks the model’s assessment was that it could assist with the process, but not in a fully autonomous mode. In most operations involvement of a human author was still required.

Part 1 – AI applications in practice

The need for human involvement is confirmed by looking at the ways GenAI assistance has been incorporated into real-life newsroom workflows over the last year.

Accelerated authoring

Accelerated authoring tools aim to increase the productivity of the individual journalist while maintaining high standards of quality and accuracy.

The webinar looked at examples of AI-assisted authoring that have recently been incorporated into Eidosmedia platforms. They included AI Companion , an integrated ‘toolset’ accessed from within the authoring workspace by a drop-down menu offering actions from ‘Suggest a suitable headline and summary’ to ‘Shorten this paragraph’ or ‘Translate this quote into English’.

Clicking on a command generates the appropriate prompt, sends it to the AI model and inserts the result into the story text. Integrating the function into the authoring UI in this way avoids the distraction of having to move into another workspace, allowing journalists to rapidly access the AI functions without interrupting their authoring focus.

Each action is carried out by a call to an external AI model and is highly customizable: the user can associate the external AI model of their choice with each type of command. It is also extensible – other AI-based services, such as automatic blurring of faces in photos, can be quickly added to the user menus.

Writing Assistant – intelligent editorial review

Unlike the Companion which uses external AI models, the Writing Asssistant is driven by an internal AI engine that learns and adapts its responses to the users and content domain of the newsroom operation. It automates many aspects of manual editorial review, checking spelling, punctuation, grammar, style and accessibility.

All of the editorial checks are context and syntax sensitive, providing more intelligent and nuanced recommendations than a static grammar or style guide.

Automatic print page layout

The most fully automated solution among the examples of AI implementations presented during the webinar was automatic print page layout.

The manual layout of news stories into print pages is the most labor-intensive process remaining in modern news operations and a major source of operating costs.

Eidosmedia’s page automation solution is built onto its existing multi-edition management environment. The story and media data of a print section is exported to an external AI engine supplied by Sophi.io. After a short interval, the computed layout parameters are returned and the stories are laid out correctly in the pages of the section. A ten-page sequence typically takes under two minutes.

In production

The solution, which has just entered production at German regional daily Nordwest Zeitung, (see case study) significantly speeds up the preparation of print editions, freeing up page editors to carry out more creative work and quality control.

As well as the productivity benefits, the solution has also significantly reduced the effort to create special inserts and supplements, allowing the paper to cover special events and news themes more rapidly and flexibly.

The page automation solution is currently being rolled out at two other news groups in Germany.

Part 2 – best practices

AI as a ‘crane’ for heavy lifting

Following these practical examples, in the second part of the Webinar, news designer Mario Garcia laid out a framework for human-AI collaboration that optimizes the strengths of each partner in the relationship.

He invited the participants to think of AI technology like the crane that allows workers to construct a tall building. By doing the heavy lifting, the crane enormously accelerates the work – but the design of the building and the scheduling of the work is something that the crane cannot do – this is work for humans.

In the same way, he sees the role of AI in the editorial space as carrying out those time-consuming routine tasks – tagging, extracting, summarizing, scheduling – to leave human journalists free to do the kind of creative work that leads to engaging content.

The sandwich model

In particular he emphasized that every AI-assisted task should have what might be called a ‘sandwich’ structure – with a human at the beginning and the end and the AI in the middle.

• Human
• AI
• Human

That is, the human initiates the task, whether it’s asking for a summary or style check, the AI performs the task and the human then carries out the essential job of reviewing the result.

Looking ahead

The webinar confirmed the valuable role the AI technologies can play in improving the productivity of news operations through suitable integration with user interfaces and workflow – a role that is likely to grow in importance as the capabilities of AI models are improved and refined.

At the same time it confirmed the centrality of the human subject in the authoring and editing process. Only by ensuring that the critical judgement of the author is fully engaged in specifying the task and assessing the outcome, can we ensure that the dramatic gains in productivity enabled by the use of generative AI are not obtained at the cost of editorial quality.

Find out more about the productivity benefits of automated creation and delivery in news-media operations.