Google started delivering answers generated by artificial intelligence in the search for those who activate a test option. The text appears in a prominent place, before any link to news portals.
The model, called by experts and entities representing the newspapers heard by the Sheet an anti-competitive practice, it removes visibility from the vehicles that reported the news and adapted for years to the competition for first place in the list of results delivered on the most visited page on the internet. Furthermore, they say, it can contribute to noise in public debate and increase misinformation.
The testing phase runs until May 2024, according to the Google page. The company warns that generative AI is experimental and the quality and availability of responses may vary.
The search giant says it presents the sources of information used, unlike competitor ChatGPT. Despite this, it prominently displays the AI-generated text and presents reference links in smaller font, to the right of the answer.
By summarizing reports, the AI-generated response reduces the incentive for people to access the reference website, according to University of Houston economics professor Haaris Mateen. He is one of the recent authors of a study that estimates Google and Facebook’s debt to American newspapers.
“Giving users the most important points of the story makes the original content less interesting. This further reduces the chances of news outlets making money from the content they produce and publish”, assesses Mateen. Another risk is diverting the reader from more complex discussions.
Google still does not say whether it is responsible for the answers generated by its artificial intelligence model, as newspapers do. In the case of the Bard AI platform, it only says that the answers may be inaccurate.
The director of research in culture, knowledge and data protection at InternetLab, Alice Lana, states that Google already mediates the relationship between people and information by ordering which pages it prioritizes. “With AI, we will see a deepening of this mediation, because it won’t even be the text written by someone, but the AI itself will generate this text.” Thus, the discussion about platform responsibility gains a new topic.
A Sheet tested the feature, which delivers recent news information. Can you answer, for example, what inflation is in Argentina during the month of October. To do this, it used information from O Estado de S. Paulo, protected by a paywall, and UOL.
The report from Sheet which appeared in first place in the common search was not cited as a source.
In a note, Google Brazil states that it “continues to prioritize approaches that send traffic to the websites of a wide number of content creators, including media outlets, to continue promoting a healthy and open web.”
Google has not yet released ways for news sites to protect their data if they do not want to make information available to train the technology company’s intelligence models. It said, however, that it works to develop new machine-readable protocols to give web publishers choice and control over how their web content is used for AI and search.
In the case of ChatGPT, the developer OpenAI published instructions for creators to protect their publications.
Websites can add a file called “robot.txt” to their source code, with instructions for robots from Google and other search engines to ignore content published at a given address. This solution, however, puts the link in a kind of web limbo, as it no longer appears in searches.
Justice, for example, uses this measure to erase historical data that is harmful to someone, in countries where there is the so-called right to be forgotten.
In the US, the feature has been operating since August, and major press outlets have already spoken out saying they do not want the company to appropriate data without authorization. This was the case of the New York Times and Washington Post, the newspapers with the most digital subscriptions in the world. According to the Washington Post, more than 500 newspapers have blocked ChatGPT bots.
Therefore, in the case of Google, there would be another side effect: these sites would no longer appear among the links cited as references by Google’s AI response.
For the president of the employer organization ANJ (National Association of Newspapers), Marcelo Rech, Google’s new feature is extremely worrying. “From the tests so far, the source of the information is hidden in a tab, which needs to be activated by the user. There is an obvious attempt to reduce the visibility of the links, further cutting off traffic to vehicles.”
“This initiative is part of the big techs’ strategy of creating internets in themselves, with the aim of keeping the user within their own ecosystem, thus increasingly engaging with their ads”, says Rech.
“Unfortunately, this is done at the expense of encouraging plurality and the traffic of journalistic vehicles, which only confirms the monopolistic vocation of big tech and the need for regulation that rebalances the forces present in the digital world”, he adds.
A Sheet, the president of Fenaj (National Federation of Journalists), Samira Castro, says that compiling information without taking responsibility, in a certain way, can generate some form of disinformation. “Therefore, we need to think about regulating technology.”
For Castro, the responses generated by AI make the work of the journalist invisible, whose production is used to train language models such as Bard, from Google, and ChatGPT. “It is necessary to protect news production from this usurped use. Journalism is expensive.”
Google says it will prioritize vehicle compensation strategies as it experiments with new AI-powered features in its products.
The news reaches the public amid discussions about the duty of big techs to finance news production. Google’s search service and the social network Facebook owe between US$11.9 billion and US$13.9 billion to US news outlets alone, according to estimates from an institute linked to Columbia University.
There is no similar study in Brazil. The National Congress discusses the possibility of technology companies paying to broadcast news, given that people consume and share reports on these platforms. This excerpt was removed from the Fake News PL and is being processed in a new text.
Canada and Australia approved similar projects, but are experiencing opposite situations. Australia left a gap for big techs to negotiate directly with newspapers, and companies have reached agreements. In Canada, the definition was up to the law and Facebook and Google responded with boycotts — they blocked the circulation of news.
In the meantime, Google continues to implement its AI capabilities. He stated that he should gradually bring the generative AI experience to his application available on Android and iPhone starting this week.