A new Science paper proposes using existing intellectual property law to create and enforce AI regulation, an approach much faster than traditional legislation.
“Leveraging IP for AI governance” by Cason Schmit, Meg Doerr and Jennifer K. Wagner:
Our model leverages two radically different approaches to manage intellectual property (IP) rights. The first is copyleft licensing, which is traditionally used to enable widespread sharing of created content, including open-source software. The second is the “patent troll” model, which is often derided for suppressing technological development.
Copyleft licensing is designed to spread virally. Examples include the GNU General Public License (GPL) and the Creative Commons ShareAlike license.
This idea has been discussed in the forum before. See A Viral License for AI Safety: The GPL as a Model for Cooperation by Ivan Vendrov and Nat Kozak.
Another related project is Responsible AI Licenses (RAIL) and their paper “Behavioral Use Licensing for Responsible AI.”
Deepmind seems to have ~400 patents (link), while OpenAI doesn’t seem to file patents. This is a recent Forbes article on OpenAI being potentially open to a lawsuit from google regarding the use of transformers (link).
A really brief and naive search for patent litigations for terms like “neural networks” or “machine learning” here didn’t find any substantial increasing trend, but I don’t understand this database at all..
If anybody is new to the concept of patent trolling, this is a fun podcast episode: https://www.thisamericanlife.org/441/when-patents-attack
Have read about the basic idea about ‘troll for good’.
Feel skepical about how it will work in reality: while giant tech are invincible, trolls might only kill upcoming startups and innovations.