Can AI invent?

A client of mine recently worked with ChatGPT on an idea he wanted to develop. To his surprise and pleasure, ChatGPT gave him a really good solution! Not only that, ChatGPT then provided an implementation and a technical spec for it. He was so amazed at what they produced together in such a short time!

He asked me if the AI might be an inventor. I knew that the USPTO’s answer to this question was that there needed to be a human inventor. But, the rules of inventorship require that ALL the inventors be listed. So, if the AI came up with the inventive implementation, then the AI should be an inventor as well, no?

How would that work? If the AI was an inventor, how would it sign the declaration? And how would it sign the assignment to his company? Wait – CAN the AI assign rights to the company? Maybe it has to assign rights to the company that owns the AI (in ChatGPT’s case, OpenAI). So, who owns the rights?

Well, my client asked ChatGPT about this. ChatGPT categorically stated that the OpenAI’s license to its users says that both the questions and the answers belong to the user. So, we were OK on the ownership question.

But, had an employee provided the response that ChatGPT did, my client would have listed the employee as an inventor, as is required, even though the rights belong to his company.

So, should we consider the AI to be an inventor?

As one does nowadays, we asked ChatGPT (full disclosure, my husband, David Goldfarb, who is my resident AI expert, did). ChatGPT 4.5 provided a full discussion around AI inventorship. The crux of its argument is:

“Inventiveness appears fundamentally tied to these key elements:

• Conscious intent or purpose behind the creation.

• Original conceptualization or recognition of novelty.

• Meaningful selection or interpretation from random or external stimuli.

LLMs like ChatGPT lack intentional consciousness or genuine conceptualization; thus, strictly speaking, they are catalysts for human inventiveness rather than being inventors themselves.”

I think a better way to understand this may be – the LLM does not make any value judgements on the responses it gives to your prompts. If you asked it to give you pro-Palestinian propaganda, it would, just as it would give you pro-Israel propaganda, if you asked for it. It doesn’t care about, nor does it have any stake in, the issues.

For inventions, the LLM does not know that its response to your prompt included inventive ideas. It is in YOUR response of WOW that the invention lies. You could have just liked or disliked the response. But, when you saw the amazingness of what the LLM wrote, you defined it as inventive.

This is similar to a recent situation of a 3 year old girl from Bet Shemesh who found an interesting stone on a hike in the area. The 3 year old, like an LLM, didn’t know that she had found something special. It was her parents who realized the importance of the find and brought it to the Israeli Antiquities Ministry. Turns out, it was a VERY interesting stone – a coin from over 3,000 years ago!

The 3 year old girl discovered the stone. But it took her parents to REALIZE its importance. Same for discoveries – they only become inventions when someone, usually the discoverer, realize what they are and how they might be useful to the world.

So, too, are ChatGPT’s responses. They provide facts, concepts, etc. from the huge database of public knowledge. Like discoveries, ChatGPT’s responses only become inventions when its user looks at the responses and realizes how they might be useful to the world. And this makes the user the real, and only, inventor.

Another way to think of this is to consider the conversation two colleagues might have that leads to an invention. One might ask a question, or make a comment, and the second might respond or realize something as a result. Both colleagues are thinking about the problem and are looking for solutions. When they hit on a solution, they usually both know that it is a good idea.

Compare this to the interaction with an LLM. It’s a one-way conversation – the user asks a question and the LLM does its best to provide a good answer. But the LLM doesn’t bring in concepts not related to the question, or come up with interesting metaphors, or imagine other scenarios, all of which a colleague might do, which might lead the conversation towards the solution. The LLM just provides answers and it is up to the user to move the conversation towards the solution.

In fact, when I questioned my client about ChatGPT’s supposed part in inventing, my client said that ChatGPT’s first solutions were not good enough and my client had to keep pushing ChatGPT to bring him a solution which solved his problem.

That is why the user is the only inventor.