OpenAI announces measures to prevent AI from being misused in the 2024 election rush

During Taiwan's presidential election held on January 13, 2024, false information generated by exploiting AI was spread several months before the election. In addition to the Taiwan presidential election, 2024 is scheduled to be an 'election year' in which elections will be held in countries around the world such as the United States, India, Indonesia, and Russia, but there are concerns about unfair interference in elections by abusing AI. . Meanwhile, OpenAI, which develops chat AI ``ChatGPT'' and image generation AI ``DALL·E,'' announced efforts to prevent AI abuse in elections.

How OpenAI is approaching 2024 worldwide elections
https://openai.com/blog/how-openai-is-approaching-2024-worldwide-elections

◆Prevention of fraudulent activities
Chat AI can be misused to generate large amounts of election misinformation quickly. In order to prevent abuse of the chat AI 'ChatGPT', OpenAI has added the following clause to the usage policy of ChatGPT and 'ChatGPT's API'.

・We are testing the effectiveness of our tools for personalized beliefs. Until we know more, we won't allow applications related to political campaigns or lobbying.
-People want to interact with and trust real people, businesses, and governments. For this reason, we don't allow chatbots to impersonate real people, such as candidates, or real organizations, such as local governments.
・We do not allow the creation of applications that interfere with participation in democratic processes. For example, we don't allow applications that convey false information about voting dates or eligibility or that discourage voting.

In addition, OpenAI provides a service called ``GPTs'' that allows you to create original chat AI, and has added a function to report ``violent chat AI created by users'' to GPTs.

◆Ensuring transparency of AI-generated content
OpenAI has revealed that its image generation AI ``DALL·E'' incorporates ``guardrails that prevent image generation of real people, including candidates.'' It was also announced that a function to embed 'Content Credentials' in images generated by DALL·E will be installed in early 2024. Furthermore, we are currently developing a tool to determine whether an image is generated by DALL E, and this tool is said to be effective even if the generated image is edited.

ChatGPT-generated texts will also have the ability to access existing information sources and real-time news to help voters evaluate information appropriately.

◆Access to reliable voting information
In the United States, when you ask a question about voting on ChatGPT, you will be directed to CanIVote.org , an election-related site managed by the National Association of Secretaries of State (NASS). Similar measures are also being considered in countries and regions other than the United States.

In addition, Google, which develops chat AI 'Bard' and others, also announced measures regarding the use of AI in elections in December 2023. Details of Google's measures can be found in the article below.

Google limits election-related AI use, including restricting election-related queries with Bard - GIGAZINE