Concerns about language bias affecting LLMs have led to call for cooperation between Japan and EU.
Japan is looking to cooperate with the EU on AI in an attempt to develop models that are not based on English and Chinese since these languages dominate generative models, a senior government official has told Euronews.
AI development was flagged as an area for Japan-EU cooperation by Motoki Kurita, the deputy director of Japan’s Ministry of the Economy Trade and Industry’s (METI) IT industries division.
“We think that the models for generative AI are skewed towards English and Chinese language models. So we believe we can work together on non-English and non-Chinese AI and we can share insights with the EU on data and spread that, expanding it to other regions that are non-English and non-Chinese speaking countries,” he told Euronews during a briefing in Tokyo.
“Many AI models currently represented by ChatGPT for example are not based on open algorithms, which means that we may be dependent on a model for which the prompt for which we receive an answer is unknown, so language barriers play a part,” Kurita added.
Generative AI models are language neutral in the way they interact with data, but most are developed by English and Chinese language developers, which may instil language-essential bias.
Unlike the EU, with its AI Act, Japan currently has no legal framework dedicated to AI but METI and MIC are working on an AI Guideline for Business.
“We are cooperating with the US, looking at each other’s guidelines for overlaps, and we’ll publicise the results,” Kurita said, adding that AI regulations need to be interoperable because they need to work across borders.
He said that it would also be an option “to spread that initiative to other countries including the EU since we believe there is a lot we can learn from the EU.”
Read the full article here
