OpenAI has reportedly suspended AI startup Delphi for making a ChatGPT-powered bot that violated the AI giant's usage policies by impersonating presidential candidate Dean Philips. A bot made to specifically support Philips' presidential bid.
First reported by the Washington Post, Delphi created a bot named Dean.Bot behind a super PAC that supports the presidential candidate, the bot developer was banned as the use of OpenAI's technologies in political campaigns is prohibited under its guidelines. This ban is reportedly the first time that ChatGPT's creator has prohibited the use of AI in political campaigns.
Dean.Bot reportedly stemmed from its creators, Silicon Valley businessmen Matt Krisiloff and Jed Somers, who were also the ones that established the super PAC in Phillips' favor.
Following the suspension of the account, Delphi took ChatGPT out of the bot and continued to operate it using other open-source technologies but was removed altogether after OpenAI got involved.
Users will reportedly still see the disclaimer if they visit the website right now, but an apology will be displayed in place of the chatbot as it is now unavailable due to "technical difficulties."
Dean.Bot's Potential Harm
Dean.Bot was an early use of a new technology that experts have warned might seriously affect elections. The bot was reportedly able to hold real-time conversations with voters via a website.
The bot asked voters to approve of its usage and provided a disclaimer stating that it was an AI tool and not the actual Dean Phillips. However, even with warnings in place, experts reportedly warned that these technologies may fool users into accepting a risky tool.
The goal of Dean.Bot was reportedly to interact with possible supporters and disseminate the candidate's message. Supporters, such as the organization We Deserve Better, contend that when voters are given an engaging means to learn more about a candidate, bots may educate them when utilized properly.
Experts have warned however, that if disclaimers are not included, the technology may allow for large-scale robocalls to voters who mistakenly believe they are speaking with politicians or supporters. AI systems are also capable of mass-producing false material, such as phony websites or advertisements.
OpenAI on the 2024 Elections
This ban follows OpenAI's recently published guidelines on how the tech giant will act amidst the 2024 worldwide elections. The tech giant is reportedly anticipating and stopping relevant misuse, including deceptive "deepfakes," scalable influence operations, or chatbots posing as candidates.
According to the blog post, OpenAI red-teams new systems before releasing them, solicits feedback from users and outside partners, and incorporates safety mitigations to lower the risk of damage.
Additionally, OpenAI said that it has been refining its tools to increase factual accuracy, lessen prejudice, and reject some requests. For example, it has guardrails in place for its DALL·E support that prevent requests for the creation of images of actual people, including candidates.
