According to research conducted by Google, generative AI has the potential to distort our collective understanding of socio-political reality and scientific consensus, a phenomenon that is already occurring.
Unlike tech figures like Sam Altman and Elon Musk who have warned about the “existential risks” of artificial intelligence, Google’s study focuses on the real harm that generative AI is already causing and may worsen in the future.
The majority of harmful instances involving generative AI are not overtly malicious and do not necessarily violate clear content policies or terms of service. The issue lies in the systematic use of these technologies to generate content that is indistinguishable from authentic material.
Out of 200 cases of abuse identified, most involved the manipulation of AI capabilities rather than direct attacks on the technology itself. Generative AI is frequently used to fabricate fake personas, spread misinformation, or create unauthorized intimate images, posing significant risks to public trust and information security.
The accessibility and lifelike quality of generative AI outputs give rise to new forms of exploitation that blur the lines between reality and fiction. The widespread production and consumption of such content can have far-reaching consequences, even if not overtly harmful.
The use of generative AI for political propaganda erodes public trust by making it challenging to discern genuine content from manipulated images. The mass proliferation of low-quality, spam-like material fosters skepticism towards digital information and inundates users with verification tasks.
Researchers acknowledge that media coverage of incidents of abuse can introduce bias into the perception of the issue, as highly publicized cases tend to receive more attention. Nonetheless, they believe that the actual scope of abuses is likely much broader.
One troubling example highlighted in the report is the creation of non-consensual intimate images, a problem that remains inadequately addressed due to the taboo nature of discussing pornography. Generative AI has been utilized to generate countless fake images of celebrities, including instances involving Taylor Swift.
The report underscores the necessity of a comprehensive approach to addressing the abuse of generative AI, calling for collaboration among policymakers, researchers, industry leaders, and civil society. It is crucial to recognize that as a major player in generative AI development, Google bears a responsibility to address these challenges.