Google has created a new suite of AI tools—tools that have reportedly not yet been released to the public, but that are being provided to news outlets as part of the Google News Initiative. According to Adweek, small news outlets are being offered five-figure contracts in which Google pays them to use the tools for up to a year, so long as they publish a minimum amount of AI-generated content.
Nothing in the contract requires outlets to label the articles created with these tools to let readers know how they were produced.
As bad as that sounds, the details are worse. Because what Google appears to have created is a sophisticated “aggregation” engine, one that isn’t just trained on existing text but also incorporates “language taken almost verbatim from the source material,” according to Adweek. The tool makes it simple for outlets to create a “new” article by simply lifting content from outlets that put in the reporting legwork.
Google's efforts to create AI tools have been in the news lately mostly because their Gemini image creator was prone to cranking out "racially diverse Nazis" and other images that were either offensive or historically inaccurate. Or both.
Embarrassment over these highly visible mistakes caused Google to "pause" the tool, but it's planning a relaunch in the next few weeks. But that level of screwup is exactly what might be expected from the current level of tools when it comes to either text or images. Just ask former Trump attorney Michael Cohen how it went when he used AI to prepare legal documents.
AI tools are limited by failures in their ability to correctly interpret prompts, but incorrectly linking material based on names or phrases that may be coincidental, and by the limits of their training dataset. Both text and imaging tools have been prone to “hallucinations” where the results extend beyond the limits of the data based on some kind of perceived model or association. These hallucinations have led to AI-generated articles full of nonsense. Or, more disturbingly, “facts” that seemed reasonable but were not.
But ready or not, Google is pushing this new set of tools as a means of producing news articles. And now it’s luring in small news sites using cash payments to convince them to beta-test the tool. And from Adweek’s description of the tool, how it makes its articles is particularly disturbing.
Almost all journalism makes use of content from other outlets. Except for rare occasions in which everyone is working directly from court transcripts or a political speech, or the far rarer instance of genuinely original reporting, articles are not written in a vacuum.
Articles are created in response to other articles and information, from a variety of sources. I’m doing that right now. This article would not exist had Adweek, CBS, The Washington Post, and others not reported on this topic.
Journalists not using exclusively primary sources are expected to bring their own experience, knowledge, skill, and resources to deliver a perspective that’s not simply a mirror (or slight rewording) of an original article or articles. They should add to the conversation, not just repeat it. The text of source articles should be both suspect and sacred—not necessarily taken at face value, but never quoted without making that attribution clear.
But look at how Adweek describes the output of the Google tools.
What this tool is marking as “most accurate” seems to be exactly what would have earned most students an F on a sixth-grade essay, and gotten journalists fired across the ages. If Adweek’s description is accurate, then Google’s tool is a “plagiarism first, original writing last” system. Considering Google’s recent experience, maybe this is meant to cut down on mistakes generated by their AI model.
The nonprofit Poynter Institute reports that 2023 was the worst year for the journalism business since the COVID-19 pandemic, with 20,324 media jobs lost. The first few months of 2024 certainly don’t seem to be getting any better.
What Google appears to have created is an AI vampire, one that leeches content from other sites and synthesizes "articles" from the work done by its victims. It's certainly not going to save journalism and seems tailor-made to accelerate the loss of news jobs.
As Digital Content Next CEO Jason Kint told Adweek, “It’s hard to argue that stealing people’s work supports the mission of the news.”
Of course, vampires can exist only so long as there are at least a few humans remaining to feed on. But somehow, that’s not very comforting.
Campaign Action
Nothing in the contract requires outlets to label the articles created with these tools to let readers know how they were produced.
As bad as that sounds, the details are worse. Because what Google appears to have created is a sophisticated “aggregation” engine, one that isn’t just trained on existing text but also incorporates “language taken almost verbatim from the source material,” according to Adweek. The tool makes it simple for outlets to create a “new” article by simply lifting content from outlets that put in the reporting legwork.
Google's efforts to create AI tools have been in the news lately mostly because their Gemini image creator was prone to cranking out "racially diverse Nazis" and other images that were either offensive or historically inaccurate. Or both.
Embarrassment over these highly visible mistakes caused Google to "pause" the tool, but it's planning a relaunch in the next few weeks. But that level of screwup is exactly what might be expected from the current level of tools when it comes to either text or images. Just ask former Trump attorney Michael Cohen how it went when he used AI to prepare legal documents.
AI tools are limited by failures in their ability to correctly interpret prompts, but incorrectly linking material based on names or phrases that may be coincidental, and by the limits of their training dataset. Both text and imaging tools have been prone to “hallucinations” where the results extend beyond the limits of the data based on some kind of perceived model or association. These hallucinations have led to AI-generated articles full of nonsense. Or, more disturbingly, “facts” that seemed reasonable but were not.
But ready or not, Google is pushing this new set of tools as a means of producing news articles. And now it’s luring in small news sites using cash payments to convince them to beta-test the tool. And from Adweek’s description of the tool, how it makes its articles is particularly disturbing.
Almost all journalism makes use of content from other outlets. Except for rare occasions in which everyone is working directly from court transcripts or a political speech, or the far rarer instance of genuinely original reporting, articles are not written in a vacuum.
Articles are created in response to other articles and information, from a variety of sources. I’m doing that right now. This article would not exist had Adweek, CBS, The Washington Post, and others not reported on this topic.
Journalists not using exclusively primary sources are expected to bring their own experience, knowledge, skill, and resources to deliver a perspective that’s not simply a mirror (or slight rewording) of an original article or articles. They should add to the conversation, not just repeat it. The text of source articles should be both suspect and sacred—not necessarily taken at face value, but never quoted without making that attribution clear.
But look at how Adweek describes the output of the Google tools.
What this tool is marking as “most accurate” seems to be exactly what would have earned most students an F on a sixth-grade essay, and gotten journalists fired across the ages. If Adweek’s description is accurate, then Google’s tool is a “plagiarism first, original writing last” system. Considering Google’s recent experience, maybe this is meant to cut down on mistakes generated by their AI model.
The nonprofit Poynter Institute reports that 2023 was the worst year for the journalism business since the COVID-19 pandemic, with 20,324 media jobs lost. The first few months of 2024 certainly don’t seem to be getting any better.
What Google appears to have created is an AI vampire, one that leeches content from other sites and synthesizes "articles" from the work done by its victims. It's certainly not going to save journalism and seems tailor-made to accelerate the loss of news jobs.
As Digital Content Next CEO Jason Kint told Adweek, “It’s hard to argue that stealing people’s work supports the mission of the news.”
Of course, vampires can exist only so long as there are at least a few humans remaining to feed on. But somehow, that’s not very comforting.
Campaign Action