When AI Invents

Last month (July 2024) the journal Chemical Engineering Progress published an article What AI Invents by Jason Balich. The article starts as follows,

The rapid advancement of artificial intelligence (AI) has led to unprecedented capabilities in creativity, innovation, and problem-solving.

The process safety management discipline is mature, and there is a definite need for creativity. But whether AI ― or at least Large Language Models (LLMs) ― can be creative is a challenging question. (The models are backward-looking; creativity is forward-looking.)

The article discusses the question of property rights, whether an inventor has to be a natural person, and the use of trade secrets to protect AI-generated inventions.

Whether or not AI can be considered a person, we are still faced with the question of ownership of intellectual property, and of how that property was created. For example, even if a patent proposal has a natural person’s name on it, how do we know that that person did not use a LLM to create the information in the proposal? (Indeed, how do you, dear reader, know that this post was written by a human being?) Furthermore, how do we know that that person did not inadvertently use someone else’s intellectual property while writing the proposal? (The key word here is ‘inadvertent’.)

We are also faced with product liability issues, as discussed in our posts The Real Challenge for AI and PSM and Artificial Intelligence, Process Safety and Litigation.

Behind all these concerns lies the problem of recursivity and data degradation, as we discuss in ChatGPT, Process Safety and Recursive Gibberish.