Scroll to top

AI in journalism

The difference between generative AI
and rules based AI
How rules based AI and generative AI solve different problems for local newsrooms

Since the arrival of publicly available generative AI (initially as ChatGPT) in late 2022, AI has been one of the hottest topics in the news industry. At United Robots, we firmly believe that the focus of AI in local newsrooms should not be tech, but rather the problems it can solve. Problems like lack of newsroom resources and time as well as lack of unique local content. Here, we’ll discuss what problems our rules based AI solutions solve, versus what generative AI is best at.

Not all AI is created equal…

United Robots build text robots using rules based AI, and we sell the automated content they produce. The raw material is structured, verified data – meaning only facts available in the data set end up in the text. Hallucinations cannot happen because the type of AI we build is journalistically limited by design. The downside of building text robots, compared to using GPT, is that it requires expertise and is a complex process involving programmers, writers (the robots don’t actually create the text segments, people do), data experts and linguists. The upside is that the content generated is safe to publish straight to news sites – it’s immediate and it’s correct.

Use cases rules based AI:

> Hyperlocal content, e g home sales on a neighbourhood level, traffic updates or game reports from local sports. That means each small community gets stories very close to home, relevant to them. 
> 24/7 coverage. Because the content is based on verified data and can be auto-published, it can be used to provide 24/7 instant updates of extreme weather warnings, wildfire alerts, hurricanes and earthquakes.

Everyone is talking about AI – but what are we talking about? 

As tends to happen with buzzwords, any original definition of AI is gradually being replaced by whatever meaning people perceive it to have. Here, we're going to limit ourselves to clearing up some confusion around the types of AI used for text generation. There are two basic models:

• Data-to-text models, (built on rules based AI, see below) which create text based on sets of structured data like sports results or financial data. This type of tech is what many of the first text robots were built on, it’s the model for self-service tools like Wordsmith and Arria, and the data-to-text model is what we use. The key feature here is that it’s data based. In other words, the model includes facts from the data, and no other “facts” in the text. Factual correctness is basically guaranteed.

• Text-to-text models, (built on generative AI, see below) a k a Large Language Models, which use deep learning to create text based on existing text – in the case of GPT-3 on 175 billion parameters of human language drawn from the internet. While these models can create good language, they work from prompts, not from data. Factual correctness cannot be assumed. Because while an LLM is able to refer to all accessible information, and include it – more or less randomly – in a text, it is not able to do fact checking.

In contrast, generative AI based on Large Language Models simply looks for language patterns to create its texts and is inherently unable to distinguish between fact and fiction. It is, however, fast, flexible and creative.

Use cases generative AI:

> Generative AI is excellent for tasks like transcribing, editing, summarizing and ideas generation for example. Anything where speed and creativity are the main features desired.

…but clearly both can create value. 

Most of our publisher partners use a combination of the two types of AI. We’ve mapped out what value(s) different types of AI / automation, in the context of local content, can bring to newsrooms as well as to the end user, the reader, as these two charts illustrate:

Values created by AI/automation for local newsrooms:

NewsroomValues

Values created by AI/automation for local readers:

Automated Content in Local News

Are you interested in learning more about what automated content and rules based AI can do for your newsroom and readers?
Contact us and set up a meeting!

Or read one of our case studies: 
How automated weather warnings are a triple win for Advance Local's newsroomsHow automated annual report articles revive local business journalism

Address

United Robots AB, Djäknegatan 9
211 35 Malmö

United Robots Data Policy

Social

Curious? Get in touch!