Dec 20, 2023 15:00:00

Google restricts election-related AI use, including limiting election-related queries in Bard

Many elections will be held around the world in 2024, including the US presidential election. Google has announced that it will strengthen its efforts to combat AI fraud to prevent its platform from being misused in elections. As a result, Bard will be subject to various restrictions, including limiting election-related queries.

How we're approaching the 2024 US elections

https://blog.google/outreach-initiatives/civics/how-were-approaching-the-2024-us-elections/

Supporting the 2024 United States election - YouTube Blog
https://blog.youtube/inside-youtube/supporting-2024-united-states-election/

Google makes a preemptive move for the 2024 election, will limit election-related queries with generative AI | Android Central
https://www.androidcentral.com/apps-software/google-to-limit-election-related-search-queries-generative-ai

Over the past few years, Google has supported numerous elections around the world, and with each election cycle, we continue to retrain our AI models to better protect against harmful content and create a trustworthy experience.

To protect its platform, Google has established policies that inform how it addresses areas such as 'manipulated media,' 'hate and harassment,' 'incitement to violence,' and 'demonstrably false claims that could undermine the democratic process.' For over a decade, Google has used machine learning classifiers and AI to identify and remove content that violates these policies, but with advances in large language models (LLMs), it is now experimenting with building a faster, more adaptive enforcement system. Regarding this new system, Google wrote, 'Early results show that it can remain agile and take action quickly as new threats emerge.'

Google is also committed to taking a principled and responsible approach to the deployment of generative AI products such as Search Generative Experience (SGE) and Bard, and prioritizes testing for safety risks, from cybersecurity vulnerabilities to misinformation and fairness. In preparation for the 2024 election, Google announced that it will limit the types of election-related queries that SGE and Bard output in answers.

Additionally, Google has implemented the following measures to prevent generative AI from being used to misuse its platform during election campaigns:

・Advertisement disclosure
Google will require election advertisers to prominently disclose if their ads contain realistic synthetic content that has been digitally altered or generated, including by AI tools.

・Content label
In the coming months, YouTube will require creators to disclose when they create realistically altered or synthetic content, and will display labels to inform users that the content they're watching is synthetic.

Additional context

SGE's contextual features and Bard's double-checking capabilities allow us to assess whether there is content on the web that supports Bard's English responses.

The 'About this image in search' feature allows you to assess the credibility and context of images found online.

・Digital watermark
SynthID , a beta tool from Google DeepMind, embeds watermarks directly into AI-generated image and audio data.

During elections, people use Google to search for information about candidates, voter registration deadlines, polling place locations, etc. To make it easier for users to find the information they need, Google highlighted the following measures:

·search
'We're working with partners like Democracy Works to surface trusted information from state and local elections offices at the top of search results when searching for topics like how to vote and where to vote. As with past elections, Google will work with the Associated Press to bring trusted election results to the public on Google.'

·news
Starting in 2022, Google will be rolling out additional news features to ensure reliable election news from each state is featured in Google News.

・YouTube
YouTube is taking appropriate measures to ensure quality election news and information is available to users.

·map
It clearly highlights polling place locations and suggests easy-to-follow directions, and applies enhanced protections to content like government buildings to prevent bad actors from creating election-related spam on Maps.

·advertisement
Google requires advertisers who wish to run election ads to undergo an identity verification process and include in-ad disclosures that clearly state who paid for the ads. The results of this process are also published in the Political Advertising Transparency Report.

Cybersecurity risks are heightened during election season. Google's Advanced Protection Program is available to elected officials, candidates, campaign workers, journalists, election officials, and other individuals at increased cybersecurity risk during election season. Google is also expanding its long-standing partnership with Defending Digital Campaigns (DDD) to provide people with the security tools they need to stay safe online, including tools to quickly configure security features in Google Workspace. Google has already distributed 100,000 free Titan Security Keys to high-risk users through partners like DDD in 2023. Google plans to distribute another 100,000 Titan Security Keys in 2024.

Google's Threat Analysis Group (TAG) and Mandiant Intelligence help identify, monitor, and respond to emerging threats. For example, TAG tracks more than 270 targeted or government-sponsored attack groups in over 50 countries every day. They also publish their findings to help the public and private sectors stay vigilant and informed. Mandiant Intelligence also provides tools to help organizations build comprehensive election security programs and strengthen their security with comprehensive tools, from proactive breach assessment services to threat intelligence tracking of information operations.