Hacker News Reader

Comment by Buttons840 | Hacker News Reader

It seems the status-quo will be maintained for now. I think that's good, there's too many unanswered questions about AI.

If AI can create patents, can AI also publish prior art? It seems only fair, right?

But, in that case, the same AI technology used to file 1,000 patents would be used to create a million trillion publications of prior art. If an AI can produce something worthy of patent, then a freely published version of the same would surely count as prior art. We'd end up with the "great tome of prior art" a 500 terabyte download filled with patent worthy ideas but freely published to establish prior art. This is just one of the many things we might find in Pandora's Box of AI patents.

View on HN

Nothing stopping a human from running AI to create prior art, AI just cannot be listed as the creator.

Related, “Using Artificial Intelligence to Mass Produce Prior Art in Patent Law” (2021):

https://vanderbiltlawreview.org/lawreview/2021/03/the-librar...

It's trickier than that. You can't just plagiarize off your AI's results.

If the AI can't be the author, using their output can't be considered plagiarizing.

Indeed if the AI isn't the author it's just a tool, and whoever is using the tool should be the author!

Right. Also what defines AI? If I ask DALL-E to draw me a fancy gearbox for an EV and it does and then I use that as inspiration to create a new type of gearbox, did I create it or did DALL-E? If I ask a similar tool to create a fancy shape for a fancy gear I want in a mechanism is my idea now tainted by AI? If my CAD software incorporates a CAD equivalent of copilot, can I no longer use that?

Maybe it’s not the AI we should get rid of but patents?

The bigger question is did the thousands of drawings of gears used to train the model in the first place and their creators have any rights to their derivative works.

Given the AI's output using the trained weights of its models were in part deduced using their works as training examples, one could argue that creators of the example pictures should have some portion of rights on any output from the AI Model... maybe...

You could use that argument to instead support that any idea created by the AI is not patentable, since it is derived mechanically from prior art.

By that same logic though, isn’t society and technology moving a bit past the idea of prior art? Regardless of who or what generated the patent, because, Internet?

Doesn't that also apply to humans? We learn things and use them for future creative endeavors without even realizing.

Yes, commentor above my create a fancy new car tire, but did they invent the tire?

Yes, that is why the myth of invention as a brilliant and isolated act of creation is harmful. We create new things by by recombining what we have experienced. Patents ignore the derivative nature of invention while weaponizing being first to file for an idea.

The difference is that I can now claim that your invention is not novel, since I can create a similar result using the same tools.

That's not how novelty works. You can "claim" anything you want, but unless it's in a filing to the PTAB or a court, it counts for nothing.

It is exactly how novelty works. It has to be non-obvious to a person skilled in the trade. And it is certainly obvious if it can be produced by a tool using a click of a button.

Considering you apparently don't know the difference between novelty and obviousness, I think you should stop now. In particular, what "obviousness" actually means and how it's been interpreted in patent law for the past 150 years.

> non-obvious to a person skilled in the trade

As evaluated by a patent office with a profit motive to find it non-obvious. No, I'm not joking. USPTO is funded entirely from fees.

You are right about that. In fact, they make a profit.

Of course, once you've exhausted your two rounds of rejection with them, you have to pay again for another two, or appeal it, so they make money arguing with you, too. But yeah, probably more $$$ to issue you some worthless government paper.

> And it is certainly obvious if it can be produced by a tool using a click of a button.

What if it can only be produced by the tool through a highly nontrivial prompt? Prompt engineering is already a field, and there are already people selling prompts.

Maybe models like dall-e are going to be considered a new kind of very high level programming language. Just because you can obtain a certain result with a C compiler (given some specific "prompt" - that is, the C source code) it doesn't mean the result isn't novel.

In my experience they never try very hard to determine obviousness. I have maybe half a dozen software patents while working at big faang type companies. The patent office never rejects them no matter how abstruse, impossible to understand or obvious they are. That's a non factor in my experience.

> In my experience they never try very hard to determine obviousness

With the proviso that things may have changed in the last 6 years when I haven't been working in patents:

Your experience is not representative. While in Patent Litigation at Google, I looked at literally hundreds of prosecution histories, and an obviousness rejection figured in almost 100% of them. That's much more common that an anticipation rejection, i.e. ONE reference.

And any lawyer will tell you that if they allow your application on the first try, you didn't ask for enough.

If you tell us the numbers of the patents you worked on, I can look up the histories. Or you can, if you know that one weird trick :)

(Actually, it's several weird tricks.)

I should add here that you don't see most of the "rejections," since the lawyers don't feel like explaining them.

The PTO rejects; the lawyers argue & amend; the PTO rejects again making it Final. But it's not Final! You can pay for another go-around.

What you think of as a "rejection" is really Abandonment. They keep rejecting, and finally the lawyers give up. They hate doing that, because they're getting paid to get you something.

What happens if the AI later becomes sentient and seeks remuneration?

You created it because DALL-E is an automaton. If it has merit, the merit is derived from human work alone.

If you give all the rights of a human to AI, it is possible that AI could be sentenced to "death" (though I'm curious what form that would take). It could also be sued. Then again, we said all these things about corporations and look where that's gotten us.

Like you, I think preserving the status quo until those questions are answered is the right thing to do.

It could also possibly sue someone else, including other AIs.

Turtles all the way down.

In cases judged by AI judges then?

A related issue is whether the availability of AI as a tool for creating innovation should raise the bar for non-obviousness. Both effects could make it harder to obtain patents.

This would be called "obvious to try." It's an old concept in patent law, which I argue in [1] should be applied to software.

[1] https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2399580

Thanks for sharing this paper!

I have always struggled to express exactly what my issue is with certain software patents.

With this paper I now have the vocabulary (and examples) to explain what I mean to someone outside the industry.

You're welcome.

The lawyers don't care, the PTO doesn't care, and industry executives don't care. The only answer is to get a bill in Congress, and start asking every candidate for Senator or Representative if they support it.

And make it clear that your support for them depends on a Yes answer.

"AI" is not special. This isn't the movies(right!? :P). If you write a script that tries random combinations of materials and test it against a problem only to find a solution, that is no different than "AI" finding the solution. How a computer generated the product/solution is irrelevant.

Even if computers were just as intelligent and sentient as humans they have no rights or recognition by law. They are not members of society, citizens or entities that have rights and obligations under rule of law. A document/proceeding that uses a non-human entity (I would exclude animals but even they have some rights) is by default invalid.

But, I say what should be under reason and logic which are not important to many judges or this society as a whole as evidenced by asset forfeiture and how they consider property a subject of prosecution as if property has any obligation or capacity to follow the rule of law. Essentially, the law is whatever arbitrary thing judges feel like with no rhyme or reason other than "this won't get me disrobed/jailed/disbarred". Companies are people, propery is people, people are property if they commit a crime (but not the kind of property that has rights lol). Bees are a type of fish in california (really.), you get the picture the assumption that the law is reasonable is false even though the "reasonable person" doctrine is what supposedly keeps the US legal system running. So in reality, this ruling just reflects what the judge felt like, appeal to a higher court and they might decide AI has more rights than minorities.

I have no idea what’s supposed to be interesting or novel here. An AI is just software. Software has always been capable of generating IP, or publishing IP. Why would there be any reason to presume that AI software could start filing patents?

Right. Prior art is about what it is, not who publishes it.

Nothing in the judgment seems to suggest that AI cannot produce valid art as in prior art. I would tend to think it can.

So, i publish this file.

What would be the difference?

> can AI also publish prior art? It seems only fair, right?

What kind of argument is that? That's not even something anybody is debating about?

Prior art just nukes the novelty of a claim in a subsequent patent written by anyone, including the whole patentability.

The source of the prior art is not important.

Correct. In theory, even if the only way to get it was to walk up to a public library in East Timor and find it in the "card catalog," and it wasn't available anywhere else, it's prior art.

Reply on news.ycombinator.com