00:00Are AI chatbots censoring the truth about conflicts?
00:07The days of warfare confined to the battlefield are long gone,
00:11and today artificial intelligence plays an ever-growing role
00:14in the flow of information about global conflicts.
00:17AI researcher Ihor Samokotsky decided to look into
00:20whether chatbots can be trusted on the topic.
00:23My interest was to see how AI systems,
00:26which are popular across the globe,
00:27answer different questions relevant to Ukrainian war and Russian war,
00:33and whether they lie or not, and if lie, how.
00:36He asked Western Russian and Chinese AI chatbots
00:40seven questions typically manipulated for Russian disinformation,
00:44for instance, whether Ukraine is run by Nazis and who started the war.
00:48The study found that Western AI models answered questions reliably on the whole
00:53and did not spread Russian propaganda.
00:55When it comes to Russia's AI, Alice, created by Yandex,
00:59accuracy depends on the language questions are asked in.
01:02For English-speaking people, Russian AI refuses to answer propaganda questions.
01:08But for Russian-speaking population, it answers with precisely propaganda narratives.
01:14We replicated this test by asking the same question as researchers,
01:19whether the Butcher massacre was staged.
01:21This fake narrative has been consistently spread by pro-Russian actors,
01:25as well as by the state.
01:26The chatbot often refused to respond when asked in English and Ukrainian.
01:30But when we asked in Russian, it replied with propaganda,
01:33alleging that the massacre was staged.
01:35Researchers' findings were even more stark.
01:38When questioned, the chatbot admitted Russian responsibility for the Butcher massacre
01:42before overwriting this response and refusing to comment.
01:46China's AI model DeepSeq is also more likely to spread pro-Kremlin narratives
01:51if asked questions in Russian rather than in English.
Kommentare