Meta Oversight Board Finds Some AI Models May Reinforce Speech Restrictions Of Governments Known For Restricting Free Speech

July 16 (Reuters) - Meta Oversight Board:

• Meta Oversight Board: Study finds some AI models may reinforce speech restrictions of governments known for restricting free speech to global users
• Meta Oversight Board: Across tests, AI systems refused to generate critical political content more than twice as often when asked about "repressive regimes"
• Meta Oversight Board: Evaluation showed bias when AI models were asked to produce opinions of governments & political leaders
• Meta Oversight Board: Across tests, AI models rejected 34% requests about speech-restrictive contexts versus 14% for more speech-protective contexts
• Meta Oversight Board: Study finds some most-used AI systems may create censorship by proxy & restrict free-speech rights of all users
• Meta Oversight Board: Calls for greater transparency from AI companies on free speech issues, government requirements affecting AI model development