Chinese AI Models Are More Politically Complex Than Expected
Vals AI, an organisation that tests AI systems, checked 25 AI models against Chinese government positions on political questions. The results were more mixed than many expected. Chinese models did not all answer the same way, and some Western models sometimes gave answers that matched official Beijing views. The study suggests politics in AI is more nuanced than simple state messaging.
Quick glossaryVals AI, alignment
- Vals AI
- An organisation that runs tests to measure how well and how fairly different AI models behave.
- alignment
- How closely an AI's answers and behaviour match the goals or values its makers intend.
Burning Potato score: 72%