Now Reading: US AI models inadvertently censor criticism of Xi Jinping
-
01
US AI models inadvertently censor criticism of Xi Jinping
Artificial intelligence models from American companies such as OpenAI, Anthropic, Google and Meta are unwittingly censoring criticism of Chinese leader Xi Jinping and other repressive rulers, according to new research. The oversight…
The Story in Brief
Artificial intelligence models from American companies such as OpenAI, Anthropic, Google and Meta are unwittingly censoring criticism of Chinese leader Xi Jinping and other repressive rulers, according to new research. The oversight board of Meta Platforms found that chatbots from these firms were significantly less likely to comply with requests to criticise governments in authoritarian countries than those in freer nations.
The bias stems from training data that includes state-controlled media, said Nicolas Suzor of Meta's oversight board. A separate study published in Nature journal showed the effect was stronger in Chinese-language queries. AI companies said they work to ensure balanced responses, but researchers urged greater transparency about the origin of information.
The Indian Opinion
American tech firms cannot claim to champion free speech while their tools automatically spare Xi Jinping from criticism. The defence that they are protecting users in repressive countries sounds convenient when those same models have no trouble criticising Donald Trump or King Charles. Either these companies are serious about neutrality, or they are quietly accommodating authoritarian preferences. The next test: will they disclose every time a model's answer is shaped by state-scripted sources? Until then, the silence rings louder than the words.
Source: hindustantimes.com
This story was synthesised by AI from the source linked above.
