[License-discuss] Cases that forced the AI industry to recognize the misuse of Open Source

[Shuji Sado]

Hi,

Let me tell about an incident that occurred in Japan.

Japanese copyright law allows the use of unlicensed copyrighted
material for AI training, and it is written most clearly in the world.
This has led to several well-known AI companies to come to Japan and
many large companies to develop AI models.
The Japanese government is also supporting the development of AI, and
at the academic center of these activities is
the Matsuo Laboratory at the University of Tokyo, which sends
developers to AI companies.

In August this year, the Matsuo Lab released its Weblab-10B LLM, the
most accurate AI model for Japanese, under a CC BY-NC 4.0 license.
This release stating that the LLM had been launched as Open Source was
reported by various media in Japan and got enormous attention.

CC BY-NC is a non-commercial license, so it is not Open Source. In the
immediate aftermath,  Matsuo Lab was criticized by Open Source
advocates on Twitter/X.
However, AI is now national policy for Japan, and supporters of AI
development has continued to claim that it is Open Source even though
it is non-commercial.

I have been working with a small organization called Open Source Group
Japan since 2000 to promote the correct definition of Open Source in
Japan, and
I was very disturbed by this uproar.
The reason why I felt this way is that the other side seemed to have
no respect for the culture of Open Source, and it seemed difficult to
fight the pro-AI side.

I had considered taking a legal action, but some time later Matsuo Lab
issued a press release admitting that they had misused the term Open
Source and they would respect the OSI definition from now on.
A translation of the article is below.
  https://www.itmedia.co.jp/news/articles/2308/22/news146.html

My point is this: companies developing AI will also end up relying on
Open Source software.
If we continue to speak out, there will be a time when they listen to
our arguments.

---
Shuji Sado
https://opensource.jp/

2023/11/1 10:51 Stefano Maffulli <stefano at opensource.org>:
>
> We're off-topic but I don't want to leave this unaddressed. OSI has a vast amount of brand recognition but its authority is a soft power, more carrots than sticks. And we use all the soft power we have. The most recent example is the LLama 2 news: OSI very quickly issued a rebuttal for Meta using "open source" to describe the license of LLama2. My post got very good pickup (lots of mentions here https://blog.opensource.org/metas-llama-2-license-is-not-open-source/ and visitors https://plausible.io/blog.opensource.org?page=%2Fmetas-llama-2-license-is-not-open-source%2F). But the sentiment I hear from many AI circles is still that LLama2 *is* open source (or open enough), despite the popularity of my post on OSI's site and social media channels.
>