The news you are reading may have been written by artificial intelligence

More and more companies are using artificial intelligence to generate media content, including news, to attract customers. Now, we are even seeing artificial intelligence being used for "gamification" of news - creating interactivity related to news content.

Whether good or bad, artificial intelligence is changing the nature of news media. If we want to protect the integrity of this industry, we must make wise choices.

Negative Impact Generated

Imagine you are reading a tragic article about the death of a young sports coach at a famous school in Sydney.

In a box on the right, there is a poll asking you to speculate on the cause of death. The poll is generated by artificial intelligence. It is designed to involve you in the story, as this will make you more likely to respond to the advertisements provided by the voting operator.

This situation is not fictional. The Guardian recently mentioned this in its report on the death of Lilie James.

Under a licensing agreement, Microsoft republished The Guardian's report on its news app and website, Microsoft Start. The poll was conducted based on the content of the article and displayed alongside the article, but The Guardian did not participate in or control it.

If the article was about an upcoming sports match, conducting a poll on possible outcomes would be harmless. However, this example illustrates the range of issues that arise when artificial intelligence is combined with news.

This incident has sparked anger. In a letter to Microsoft President Brad Smith, Anna Bateson, CEO of The Guardian Media Group, described it as "inappropriate use of generated artificial intelligence" and caused "significant reputational damage" to The Guardian and the journalist who wrote the report.

Naturally, the poll was canceled. But it raises the question: why did Microsoft allow it to happen in the first place?

Consequences of Ignoring Common Sense

Research from the University of Texas Center for Media Engagement found that polls and quizzes, among other supplementary news products, do attract readers.

Considering how low the cost of using artificial intelligence for this purpose is, news organizations (as well as companies displaying others' news) seem likely to continue doing so.

The second part of the answer is that there was no "human involvement" or limited human involvement in the Microsoft incident.

The main providers of large language models - the language models that support various artificial intelligence programs - have economic and reputational incentives to ensure that their programs do not cause harm. OpenAI's GPT model and DALL-E, Google's PaLM 2, and Meta's Llama 2 have made significant efforts to ensure that their models do not generate harmful content.

They typically do this through a process called "reinforcement learning," in which humans plan reactions to questions that could lead to harm. However, this does not always prevent the model from generating inappropriate content.

Thousands of AI-Generated Articles Every Week

Generative artificial intelligence is becoming accessible and affordable. This makes it attractive to commercial news businesses that have been suffering revenue losses. As a result, we now see artificial intelligence "writing" news stories, saving companies from paying journalists' salaries.

In June, News Corp Executive Chairman Michael Miller revealed that the company has a small team that uses artificial intelligence to write about 3,000 articles per week.

Essentially, the four-person team ensures that the content is meaningful and does not include "fabrications." They avoid the inclusion of false information fabricated by the model in the articles.

While this news may be accurate, the same tools can be used to generate potentially misleading content that can be disseminated as news, making it almost indistinguishable from articles written by professional journalists.

Since April, an investigation by NewsGuard found that hundreds of websites written in multiple languages are mostly or entirely generated by artificial intelligence to mimic real news websites. Some of these include harmful misinformation.

It is believed that these ad-filled websites are likely created to generate ad revenue.

Risks Increasing with Technological Advancements

Often, many large language models are limited by the training data they are based on. For example, a model trained on data up until 2021 would not be able to provide accurate "news" about world events in 2022.

However, this is changing as models can now be fine-tuned to respond to specific sources. In recent months, the use of an artificial intelligence framework called "retrieval-augmented generation" has developed to allow models to use the most recent data.

Through this approach, it becomes possible to create news websites using licensed content from a few news agencies.

While this may be convenient from a business perspective, it is another potential way in which artificial intelligence is pushing humans out of the news creation and dissemination process.

Curated news pages are valuable and carefully considered products. Allowing artificial intelligence to do this work may expose us to various misinformation and biases (especially in the absence of human supervision) or result in a lack of important localized reporting.

We All Lose with Cutting Corners

Australia's "News Media Bargaining Code" aims to maintain a "fair competitive environment" between large tech companies and media companies. Since the code came into effect, the use of generative artificial intelligence has brought about a second change.

Setting aside click rates, there is no comparison between the quality of news that journalists can produce and the quality of news that artificial intelligence can produce.

While generative artificial intelligence can assist journalists in their work, such as helping them sift through large amounts of content, if we start to see it as a substitute, we will lose a lot.