Google researchers have published a study highlighting concerns about the impact of generative AI on the Internet, used to create fake or doctored content

A recent study by Google researchers highlights how artificial intelligence is disrupting the internet by spreading false information. This is somewhat ironic given Google’s role in developing AI technology. The accessibility and sophistication of AI-generated content have enabled new forms of abuse and exacerbated existing issues, making it harder to distinguish between truth and falsehood.

Since its introduction in 2022, generative AI has opened up many opportunities to accelerate development in many areas. Today, AI tools have a wide range of capabilities: from complex audiovisual analysis (through natural language understanding) to mathematical reasoning and the creation of realistic images. This versatility has enabled the technology to be implemented in critical industries such as healthcare, public services, and scientific research.

As technology advances, the risks of misuse are becoming increasingly concerning. One significant issue is the spread of misinformation that has flooded the internet. A recent analysis by Google found that AI is now the main source of image-based misinformation. This phenomenon is exacerbated by the increased availability of tools that allow anyone to generate content with minimal technical knowledge.

While previous research provides valuable insights into the threats posed by AI misuse, it does not precisely indicate the various strategies used for this purpose. In other words, we lack detailed knowledge of the tactics attackers use to spread false information. As technology becomes more powerful, understanding how abuse manifests is critical.

A new study by the Google team highlights the various strategies used to spread AI-generated false information online. “Through this analysis, we identify key and emerging patterns of misuse, including potential motives, strategies, and how attackers use and abuse the system’s capabilities in various forms (e.g., images, text, audio, video),” the researchers explain in their paper, pre-published on the arXiv platform.

Abuses that do not require in-depth technical knowledge

The researchers analyzed 200 media reports of AI misuse between January 2023 and March 2024. They identified key trends, including the ways and motivations behind using AI tools in uncontrolled environments (i.e., real life). The analysis covered images, text, audio, and video content.

The most common abuses were false images of people and falsification of evidence. This false content is often spread to influence public opinion, facilitate fraud, and make a profit. Interestingly, most cases (9 out of 10) do not require deep technical knowledge but exploit the easily accessible capabilities of AI tools.

While the abuse is not overtly harmful, it is potentially dangerous. Experts note that the increased sophistication, accessibility, and prevalence of generative AI tools are leading to new forms of lower-level abuse. These acts are neither overtly harmful nor clear violations of the tools’ terms of use but still raise troubling ethical concerns.

This finding suggests that while most AI tools have ethical safety conditions, users find ways to circumvent them with intelligently worded prompts. According to the team, this is a new and evolving form of communication aimed at blurring the lines between reliable and false information. The latter is mainly aimed at political awareness and self-promotion. This is fraught with an increase in public distrust in digital information and an overload of users with verification tasks.

On the other hand, users can bypass these security measures in other ways as well. For example, if there is no option to enter a prompt that explicitly contains the name of a celebrity, users can download a photo from a search engine and then modify it as they wish by inserting it as a link into an AI tool.

However, using data solely from the media limits the scope of the study, the researchers say. These organizations tend to focus on specific information that may interest their target audience, which can introduce bias into the analysis. Additionally, the document, oddly enough, does not mention a single case of abuse of artificial intelligence tools developed by Google.

Nevertheless, the results obtained help us understand the degree of influence technology has on the quality of digital information. According to the researchers, this highlights the need for a multi-pronged approach to reducing the risks of malicious use of technology.