The aim by year end is to work out an approach for AI that
will likely be closer to the US attitude than the stringent rules championed by
the EU, said the official, who declined to be identified as they were not
authorised to talk with media.
A softer Japanese approach could dull EU efforts to
establish its rules as a global benchmark, with requirements such as companies
disclosing copyrighted material used to train AI systems that generate content
like text and graphics.
EU industry chief Thierry Breton is visiting Tokyo this week
to promote the bloc's approach to AI rule-making as well as to deepen
cooperation in semiconductors.
The government official did not elaborate on areas where
Japan's rules were likely to differ from those of the EU.
The chair of the government's AI strategy council, The
University of Tokyo's Prof. Yutaka Matsuo, called the EU's rules a "little
too strict," saying it is "almost impossible" to specify
copyrighted material used for deep learning.
"With the EU, the issue is less about how to promote
innovation and more about making already large companies take
responsibility," said Matsuo, who also chairs the Japan Deep Learning
Association and is an independent director on the board of Masayoshi Son's
SoftBank Group.
Advances in generative AI by firms such as startup OpenAI,
backed by Microsoft, are stimulating both excitement and concern due to the
potential to transform business and society in general.
Such is its potential that AI is among technologies
including advanced semiconductors and quantum computers that the U.S. and
allied industrial democracies are in a race with China to develop.
"There are things that really are a concern and I think
these things probably should be a concern for any democracy," Breton said.
"With likeminded partners and friends like Japan or the
US, I think it's important to explain what we did," Breton said of the
EU's regulatory approach.
For Japan, AI could help cope with the population decline
that is causing a labour shortage.
It could also stimulate demand for advanced chips that
government-backed venture Rapidus plans to manufacture as part of an industrial
policy aimed at regaining Japan's lost lead in technology, the source said.
Japan's computing power, defined as the availability of
graphics processing units (GPUs) used to train AI, is far behind that of the
US, experts said.
"If you increased the GPUs in Japan by 10 times, it
would probably still be less than what OpenAI has available," said Prof.
Matsuo. © Reuters