GENERATIVE AI has rapidly disrupted the news media industry, permanently changing consumer behaviours and how news is produced and distributed.
But “every challenge is also an opportunity,” noted Gert Ysebaert, president of International News Media Association (INMA) and chief executive officer of Mediahuis, in his opening speech for the INMA Media Subscriptions Summit 2025 in Amsterdam.
Aleksandra Przegalinska, professor at Kozminski University and Harvard University, who is known as the leading futurist in AI, shared how media industry leaders can capitalise on the opportunities afforded by AI.
GenAI technologies have enormous potential to augment productivity by increasing efficiency and the quality of work, she said, provided the technologies are tailored to the specific needs of each business.
A revolution in accessibility
The launch of OpenAI’s ChatGPT in November 2022 caused a “revolution in accessibility,” Przegalinska said, as users can now interact with the technology using natural language instead of code.
In a survey conducted by Przegalinska and her colleagues, 60% of Gen Z respondents said they use GenAI models.
Despite the large uptake, GenAI has caused widespread turbulence in the media industry, blurring the traditional distinction between “human creates, AI generates”.
- Digital Transformation: Time to consider editorial restructuring
- 7 concrete applications of ChatGPT in the newsroom you can start using tomorrow
- In modern digital news operations, editors need to drive the success
- Adapting to the digital age: How newsrooms have transformed in 30 years
Keep Reading
“We are now starting to find the best use cases for these algorithms,” Przegalinska said.
While the industry is adapting to the technology, it has not been completely “naturalised and become obvious” like the internet.
AI development is accelerating and becoming a geopolitical force by states that see it as providing a competitive edge.
The future of AI
A pragmatic shift has been observed worldwide in the last year, with greater pluralism in the number and types of GenAI models and businesses focussing on the real benefits and costs of using the technology in the way they operate.
Competition between new models has grown over the past two years, with Mistral from France and DeepSeek from China, but Przegalinska still describes the market as a “battle of the giants.”
After the release of the Draghi report on EU’s competitiveness, Europe has worked to improve its competitiveness by presenting its own initiatives and funding — such as the Open-EU LLM that bridges different datasets and languages from across Europe.
Looking “under the hood” at the types of genAI now available, Przegalinska sees six categories emerging: Large language models (LLMs), video, sound, diffusion, agentic and physical AI.
Agentic AI consists of operational plug-in systems that sit on top of an LLM and go beyond it to perform tasks such as Web browsing.
They are the most important category for the future development of AI, Przegalinska said.
The implementation of GenAI in business operations is still in the early phase and is dependent on infrastructural expansion to deal with the energy load of AI.
Additionally, GenAI companies are experimenting with subscription-based services.
OpenAI’s superior models, which cost as much as US$20 000 per month, are not performing well, as most people are still becoming accustomed to the openly available models.
The threshold of payment that people will accept for such models is still under discussion.
Open-source models are gaining traction in Europe, and small- to medium-size companies tend to use smaller AI models and tailor them with training data to perform specialised tasks.
Challenges with AI
GenAI models often produce responses to prompts that are “wrong in a very banal sense,” Przegalinska said.
She highlighted the Stanford Smallville experiment, where scientists left non-playing characters run by AI to their own devices in a computer game and observed gossiping, egotistical behaviour and laziness within a couple of days.
This exemplifies how “blind” models still emulate social behaviours learned through training data, with both desired and undesired effects.
Przegalinska stressed that developers and users must be careful with these unintended consequences.
AI developers also need to be mindful of copyright issues, which have hit sound AI models particularly hard because they use licensed content to train models.
The New York Times has filed a lawsuit against OpenAI for content violation, and the European Union is also discussing introducing new legislation to prevent GenAI from breaching copyright.
Take-home message
Przegalinska’s key message was that businesses and organisations must make a conscious choice about how to use AI to maximise the value it adds to the work.
Human-centred collaborative AI that can “adapt to us, instead of us adapting to AI” should be the way forward.
She used the example of writing her book with the help of an AI critic, which responded to a prompt that it didn’t like the book.
The result was a solid and helpful criticism for the authors.
They also created a “book Bot” — deepfaked clones of the authors available on the Web site as an engagement tool for readers.
While there is no “killer app” that performs all possible functions, there is a plethora of AI systems suited for specific tasks.
The next “big thing,” according to Przegalinska, will be AI that helps us on a more nuanced level in different ways for each business
sector.
