AI Overviews

Transparency and accountability 🙈

AI Overviews presents several important transparency, openness and accountability limitations:

• Data training. Google does not divulge what data sources AI Overviews was trained on, perhaps out of fear that it will be found to have violated copyright.

• Source attribution. Although AI Overviews provides links to the sources of information, the summaries generated do not always accurately reflect the content of those sources. In addition, users may accept AI-generated responses as authoritative without fully understanding the context or reliability of the cited material.

• Marketing claims. Google says links generated by AI Overviews results in a higher level of click-throughs, but has provided no data to support its claim.

• Environmental impacts. Google does not reveal the environmental impacts of specific models, making it challenging to assess their energy and water consumption and carbon emissions.

Risks and harms 🛑

Concerns have been raised about AI Overviews' generation of inaccurate, misleading, false, nonsensical and dangerous summaries, and the challenges it faces recognising nuance, satire and sarcasm.

These limitations are seen to result in actual and potential harms to individual and public health and safety, the erosion of trust in information in the public sphere, politics and democratic systems, bias and discrimination, and posing reputation potential reputational damage to named people and organisations.

Experts have raised alarms about the AI's reliability in critical situations. For example, when queried about treating a rattlesnake bite, the system recommended treating it with ice or heat - both of which are considered inappropriate.

AI Overviews is seen to pose a serious risk to the business models and profitability of publishers - many of which have seen their content scraped without acknowledgement or consent by Google to feed the AI Overviews system.

Research, advocacy 🧮

• News/Media Alliance. News/Media Alliance Calls on FTC, DOJ to Investigate Google’s Misappropriation of Digital News Publishing, Stop Expansion of Generative AI Overviews Offering

Investigations, assessments, audits 👁️

• Press Gazette. ‘Devastating’ potential impact of Google AI Overviews on publisher visibility revealed

News, commentary, analysis 🗞️

• https://originality.ai/blog/worst-google-ai-responses

• https://www.wired.com/story/google-ai-overviews-broken-how-ai-works/

• https://www.technologyreview.com/2024/05/31/1093019/why-are-googles-ai-overviews-results-so-bad/

• https://www.scientificamerican.com/article/what-do-googles-ai-answers-cost-the-environment/