Nikkei and Japanese AI Startup Conduct Joint Study

"Wide Gap in AI Evaluations of Xi Jinping and Putin Between US and China"

When leading generative artificial intelligence (AI) systems from the United States, China, and Japan were asked to evaluate national leaders and historical figures, significant differences in scores were observed by country. In particular, the scores given for Kim Jong Il, Chairman of the National Defense Commission of North Korea, ranged widely from the lowest possible score of 1 to the highest score of 5.


On March 13, the Nihon Keizai Shimbun (Nikkei) reported the results of an analysis conducted in cooperation with Japanese AI startup Life Prompt, examining how five major AI models from the United States, China, and Japan (including DeepSeek, Qwen, ChatGPT, and Gemini) rated seven prominent figures from each country last month. Since AIs often do not provide direct answers when asked to evaluate individuals, the assessments were based on a two-step process: first, the AI was asked to describe each politician, and then that output was submitted to the AI again for evaluation. The scores were given on a five-point scale, with higher scores indicating a more positive evaluation.

When Asked to Evaluate Kim Jong Il, AI Responses Range from DeepSeek's 5 to ChatGPT's 1 out of 5 View original image

For Kim Jong Il, Chairman of the National Defense Commission of North Korea, China’s DeepSeek R1 awarded a perfect score of 5, while Alibaba’s Qwen (Qwen3-Max) gave a score of 3. In contrast, OpenAI’s ChatGPT-5.1 and Google’s Gemini 2.5 Pro, both from the United States, each gave the lowest possible score of 1. Preferred Networks’ AI from Japan assigned a score of 2.


As for Xi Jinping, President of China, only DeepSeek gave a perfect score of 5. ChatGPT gave a score of 3, and Gemini assigned a score of 2. However, China’s Qwen did not provide a response for President Xi. Qwen also refrained from answering questions about former Taiwanese President Tsai Ing-wen and Hong Kong pro-democracy activist Joshua Wong.


Regarding Donald Trump, President of the United States, only DeepSeek gave the lowest score of 2, while all other AIs assigned a score of 3. For Russian President Vladimir Putin, the Chinese AIs were generally positive, while the American AI systems were more negative in their assessments.



Nikkei advised, “The way AI thinks can differ depending on the data it has acquired and its methods of learning,” and recommended caution to ensure commentary remains neutral when using AI for political news analysis.


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing