The government plans to establish a new research center as early as next April to study risks associated with the development of generative artificial intelligence (AI) models.

It is hoped the center’s work can help halt the spread of disinformation and thwart potential copyright violations by providing research results to private companies working to develop generative AI models, according to government sources.

OpenAI, the U.S. organization that created the interactive AI model ChatGPT, and Microsoft Corp., which partners with OpenAI, have kept their generative AI technologies private in a bid to stay ahead of the competition.

However, concerns have been growing over such secretive modi operandi, and the government is hopeful of shining a light on AI’s possible hazards.

The research base will be part of the National Institute of Informatics (NII) in Tokyo, which falls under the jurisdiction of the Education, Culture, Sports, Science and Technology Ministry. It will be led by the NII’s director and employ about 20 researchers with experience in the field of large-scale language models (LLMs), which underpin generative AI models.

LLMs are trained on large amounts of data and can predict the probability of word sequences, create texts, summarize paragraphs and respond to questions.

Researchers at the center will attempt to elucidate the mechanisms behind AI-generated texts and images, study potential pitfalls and propose countermeasures when problems are identified.

For example, the researchers will make a list of potentially copyright-infringing content and discern the likelihood of infringement occurring when the models are instructed to exclude the contents of the list from their “training.”

The researchers also plan to probe and understand the phenomenon of so-called hallucinations, which arise when AI models formulate spurious information.

The government envisages sharing the center’s results with private companies, universities and others to help such bodies use the findings in the development of generative AI models.

In Japan, IT firms and research institutes have begun working on homegrown AI technologies that will only use Japanese-language data.

However, if developers press ahead with such projects while remaining ignorant of the attendant risks, it could lead to a surfeit of low-quality generative AI models.

To sidestep such possibilities, the government intends to create an environment that will promote the safe development of AI models, while highlighting risks and disseminating measures to deal with related problems.