What the Financial Times learned from experimenting with AI

Credit: Adam Tinworth, via One Man and his Blog

When the Financial Times (FT) created its AccelerateAI team earlier this year, it was trying to navigate through the noise and explore both opportunities and threats.

"People are tired of talking about whether AI is going to kill us or save humanity," said team leader Liz Lohn, director of product at FT, speaking at the Future of Media Technology conference last Thursday.

Early on, the AccelerateAI team decided that they would never risk readers’ trust. They were ready to experiment but also to pull the plug if something went wrong.

It was not all about generative AI either. According to Lohn, there are two distinct solutions the technology can be used for: creation and discovery of the story, and delivery.

Content creation

The first lesson the team learned was humility, as it became clear that the value of content AI produces is only as good as the tool itself. Can it produce summaries that are both informative and enticing? Or translate that to prompt and create good enough input?

The team spent six weeks training a large language model (LLM) and crafted a prompt for one summary. When they saw the underwhelming result, the newsletter team did not buy in.

"It was the best kind of failure, we’ve learned a lot from it," said Lohn.

Computational news discovery

The next experiment was using AI to extract information from structured datasets and generate new story leads. They decided to look into MPs' earnings and expenses.

The problem with generative AI, however, is still that it makes things up (hallucinates).

"Hallucinations are a bug in the media and a feature in LLMs," said Lohn, adding that generative AI will work better as an "inspire me" tool rather than a content creator.

"As soon as it works, no one calls it AI anymore," she joked.

Sam Gould, AI lead at FT Strategies who joined Lohn on the stage, sees three clear changes in user behaviour.

• products are becoming multimodal
• audiences engage with conversational experiences
• and automation tools are becoming agents

Multimodal products

Video is becoming a more important source of online news, especially with younger groups. AI-powered tools can now create text, images, audio and video, allowing publishers to experiment with different formats and features like internal processes (transcription, translations, creative analysis, data journalism), which all involve processing multiple unstructured data types. One practical example Gould shared was image analysis from live reporting on the ground, where AI was able to instantly translate a street sign written in a different language.

Fancy a chatbot?

While most people shudder at the prospect of a lengthy "conversation" with a robot, it can be an interesting tool for your audience to explore content. Just like short videos, chatbots prove particularly popular with younger audiences.

AI agent

Gould concluded by saying that AI tools are being given increasingly complex tasks, like writing code or creating new apps with minimal input from humans.

So like it or loathe it, AI is changing the publishing industry and we must get ready to make the best of opportunities while mitigating the risks.

Free daily newsletter

If you like our news and feature articles, you can sign up to receive our free daily (Mon-Fri) email newsletter (mobile friendly).