Item — HN Reader

It's pretty easy to predict what would happen in practice if this policy proposal were adopted.

An entire category of bullshit jobs would be created to write "source code" that met the regulatory requirements but had little connection to the actual code deployed in production. This code would largely be derived from the claims, and, like much patent language, be crafted to be as vague and general as possible. Most of this work could be automated by ChatGPT, as no insight is required. Another class of bullshit jobs would be created to interpret and analyze the code, as it's outside the skill set of lawyers and examiners.

Any code published as part of a patent would immediately be anti-open source, as people actually creating things would be strongly discouraged from even looking at it, as doing so would support "willful infringement." The average level of code quality would be such that nobody would want to, anyway.

There's lots that could be done to make the patent system more beneficial to society (including, in my personal opinion, dropping software patents entirely), but I don't think a code requirement would be particularly helpful.

2 comments

You are wrong on this.

A patent makes claims, patents should be shown to work, code is a proof of that. They are computational existence proofs. We should hold our systems to higher standards of rigor.

You are shooting something down for how it could possibly be subverted, rather than do that, how about we think of ways that it could be made less corruptible? Is this not a pro-fatalistic stance?

Not only should the patent require the code, the product should require the code to be held in escrow. And for safety critical systems, the formal checked models should be presented in a way the proves that the bits on the system were generated from the formally checked code.

Would it fix all problems? Hell no! Would it prevent some entities from innovating in the market, hell yes!

As it currently stands, you don't even have to have a working system. Corporations are effectively patenting problems, not solutions. Patents themselves are a worthless way of transferring knowledge, this is what needs to change.

I may be wrong, and overall I'm very sympathetic to changing the patent system to incentivize actually making things. That said, I think changes to policy must be interpreted in an adversarial context, assuming that people will game the system as much as possible. With this policy, patent filers will clearly be incentivized to give away as little of the store as possible. Also, determining whether the code really works, and really represents the system being protected, is just as difficult a problem (if not more so) than interpreting claims today.

I don't disagree, we should always have a critical stance and understand how the systems we build can and will be subverted against the intended design.

But that isn't a reason to not build them, and in this case, the effort is to repair an already broken system that has been subverted. If creatively deconstructing it isn't an option, we have to add more "features" and this case the added constraint of functioning code.

We can disagree all day what functioning code means, but the goal is worthy and should be explored.

How we apply policy that meets the intended goals and isn't gamed is the problem for law makers. Good thing we have spirit as a backstop and not just letter. A certain hedge fund and a billion dollars in AI and civilization could collapse.

Right now, the inmates are running the patent system, if we value everyone's ability to participate in innovation, we need a way to reduce BS patents and make them intrinsically more valuable as a form of knowledge transfer.

Why is it that software patents should have a higher standard than any other type of patent?

IMO they shouldn't. Personally I'd like for patenting a new physical invention to require a physical object to accompany it that embodies what is being patented.

But that's never been the case, why should it be now? If you forced every inventor to have to make their invention, it'd be a pro-industry burden that diminished the capacities of any smaller organizations or inventors.

> But that's never been the case, why should it be now?

For the same reasons TFA is making the argument "No Source Code == No Patent."

> If you forced every inventor to have to make their invention, it'd be a pro-industry burden that diminished the capacities of any smaller organizations or inventors.

If an inventor never made their invention, are they an inventor? IMO no, they're just an "idea guy" at that point.

You don't even have to be an "idea guy", many patents are patenting the problem domain.

In 1790, when patents were first brought into law in the United States, it was a requirement that the Patent Office be provided with a working copy of the invention for their archives.

In fact, in some cases, you DO have to provide a working copy; admittedly now that's when your patent seems impossible (like perpetual motion). Also, for plant patents (that's something only there so they can put it on the Patent Bar exam /s), I believe you have to provide a plant.

"make their invention" is the law; it's the RP in ARP and CRP. You can't claim a new antibiotic without giving the formula. 112 rejections are common in patent prosecution.

Maybe. Or, maybe such CodeGPT would become subject to obvious "does it actually work?" tests. Joe Average in the jury box might feel pretty confident saying "a Patent for an apple-peeler ain't valid if the so-called peeler completely fails to peel apples". Similar for software that fails to run, or to do anything resembling the claims.

At that rate, is the codegpt output any different than outputting from /dev/urandom?