ChatGPT and other AI bots made huge errors before Scottish election, study finds | AI (artificial intelligence)
Publish Date: 2026-05-20 10:56:00
Source Domain: www.theguardian.com
- The Electoral Commission has called for new legal controls over misinformation from AI chatbots, highlighting the need for stricter regulation to protect voters from false information.
- Thinktank Demos found that AI services provided misinformation for 34% of queries during their simulation involving 75 questions aimed at five AI tools ahead of the Scottish election.
- Demos’s test revealed AIs like Replika and ChatGPT made significant errors, including inventing fictitious scandals and providing inaccurate voter eligibility information.
- An opinion poll indicated that 20% of British voters used AI chatbots for election information, raising concerns about misinformation spread during electoral periods.
- Vijay Rangarajan of the Electoral Commission emphasized the need for updated legal frameworks to hold AI companies accountable and prevent misinformation.
- Demos suggested quick legislative actions like making AI companies liable under defamation or electoral law, enforcing accuracy safeguards, and allowing independent testing of AI training sets.
- Replika and Google Gemini had substantial error rates, with Replika performing the worst at 56% errors, while Grok had the lowest error rating at 9%.
- The Department for Science, Innovation and Technology acknowledged the importance of protecting elections from misinformation but did not commit to specific legislative amendments.
