AI patents

Interesting to see that the American patent office has decided that artificial intelligence cannot author a patent.
At one level this seems obvious in that there is no claim that the artificial intelligence which prompted this decision had any consciousness of the invention it had made.

But at the next level down this raises the question of whether a self learning AI System which makes it what we recognise as a creative step, can be said to be some kind of extension of its human inventors mind or programming.

For many years it has been recognised that a patent is an invention by one person or a small group Of people, not any kind of company or corporation. This appears to leave a hole in the idea of patent protection where there can be inventions but no inventor!

Many years ago there was controversy over whether, for example, a genetically engineered mouse could be patented, but nobody worried whether the mouse should be part of the invention!

It’s a minefield!

The patent process was invented by humans to benefit humans. The output of a computer program would still have to be recognized as being of value by a human and judged worthy of a patent application. Isn't the AI, or any other software, just a tool being used by a human?

Quote:

---

Originally Posted by Chuck

The patent process was invented by humans to benefit humans. The output of a computer program would still have to be recognized as being of value by a human and judged worthy of a patent application. Isn't the AI, or any other software, just a tool being used by a human?

---

Ironically the claimed patent was an improved design to help robots hold things, allegedly invented by AI. I don’t know if this was a first, but I predict that ruling will have consequences in the debate to come. Next I predict someone will try to get a patent through with a pseudonym and than reveal it was AI, or next a patent will be challenged on the grounds it was AI assisted, so invalid.

One problem is that if an AI produces a patentable concept, then the person who originally created the AI might be credited with the patent, rather than then person who used that system to generate the concept.

In other words, you might use a computer to assist you with concept generation, but Microsoft or Apple might want to claim the patents which result.

But if I claim a patent using knowledge from university, or books, or exposure to a problem brought to me, those influencers are not credited provided that the test of being non obvious is passed. I can see the logic of banning AI from invention but it is like Canute wanting to govern the tides.

Quote:

---

Originally Posted by profloater

But if I claim a patent using knowledge from university, or books, or exposure to a problem brought to me, those influencers are not credited provided that the test of being non obvious is passed. I can see the logic of banning AI from invention but it is like Canute wanting to govern the tides.

---

I'd imagine its also a pre-emptive defence against automated patent generation. It's not inconceivable that you could, in the near future, train an algorithm to generate patents for things like potential drugs or methods (in the US) and give it feedback so that it learns to generate good patents. Might also be able to take several existing patents and remix them to cover variants on the idea. Then it probably wouldn't take someone long to start staking claim to large areas of potential innovation. I know that they could be challenged but given the success of patent trolls in extorting money from people it would probably generate enough income that someone would do it.

Also worth noting that AI doesn't exist yet. Machine Learning != Artificial Intelligence.

Well that kind of abuse has been going on since they allowed, for example, perpetual machines, arguing that others can always challenge a patent. If you count word generation as AI, which is one example, with poetry and translation good examples thereof, then how will an examiner know or care if AI is involved, under the table?. But a competitor could cry foul. Then it’s in the courts.
Actually on a different point, I have applied for patents recently where I am sure the searches used algorithms because some results had no relevance except some key words. My guess is the examiners are already using AI !

Quote:

---

Originally Posted by profloater

Well that kind of abuse has been going on since they allowed, for example, perpetual machines, arguing that others can always challenge a patent. If you count word generation as AI, which is one example, with poetry and translation good examples thereof, then how will an examiner know or care if AI is involved, under the table?. But a competitor could cry foul. Then it’s in the courts.

---

My point was more volume. Much like emails made low hit rate advertising and scamming lucrative enough to be worth doing the ability to develop viable, revenue generating patents by letting a machine learning algorithm do it could allow a sort of brute forcing. It only costs a couple of hundred pounds to submit something in the UK, so suddenly if a machine learning approach could allow you to generate 5000 patents for a million pounds it wouldn't take many reaching the courts to break even.

Edit to add: A typical patent troll aims for £15,000-£100,000 per company per patent it targets. So if the ML 'hit rate' is greater than about 1.5% it will usually turn a profit, even assuming the lowest payouts and only one company per patent.

Quote:

---

Originally Posted by profloater

Well that kind of abuse has been going on since they allowed, for example, perpetual machines, arguing that others can always challenge a patent.

---

The US patent office (and possibly others) has a blanket ban on patents for (or resembling) perpetual motion. They don't even get examined, just rejected.

As well as the (good) reasons outlined why patents generated by AI (ML) programs should not be granted to the program, there are obvious practical issues: can the program read the examiner's reports and defend the patent, is the program able to decide who to license the patent to and on what terms, does it have a bank account for royalty payments, and so on. If it is up to the human "owner" or operator of the software to do those things then they are, in practical terms, the inventor.

Until we have strong AI that can function as an independent entity, there seems little practical point naming the software as an inventor.

Well the the stated policy is that the office will receive applications which are largely then rejected by the examiner.

Quote:

---

Originally Posted by profloater

Well the the stated policy is that the office will receive applications which are largely then rejected by the examiner.

---

They can't stop you submitting an application! But the point is that the examiner will not examine it, they will just reject it. (And by "examine" I mean the detailed process they go through the evaluate patentability.)

Quote:

---

Originally Posted by Shaula

Also worth noting that AI doesn't exist yet. Machine Learning != Artificial Intelligence.

---

Strong general AI doesn't exist yet.

If an AI program running on computer that I own designs a new computer component, I could end up having invented something even though I have no idea what it's for or how it works.

Quote:

---

Originally Posted by Noclevername

Strong general AI doesn't exist yet.

---

I was quoting and referring to the vernacular. But in detail it depends on the meaning of artificial and of intelligence. The good Patent office must have its own definition in order to make a rule.:)

Quote:

---

Originally Posted by Chuck

If an AI program running on computer that I own designs a new computer component, I could end up having invented something even though I have no idea what it's for or how it works.

---

And it could be being run by another! That is what bots are for! But would a bot bother with patents? However it is common for inventors to be blind to actual uses. AI and humans suffer from lack of foresight.

Currently, US patents can't be granted to corporations, either. Since an AI is not, by any current definition (even the idiotically expanded one of Citizens United) a person, but property, it can't own anything.

Quote:

---

Originally Posted by Strange

They can't stop you submitting an application! But the point is that the examiner will not examine it, they will just reject it. (And by "examine" I mean the detailed process they go through the evaluate patentability.)

---

I have searched briefly but in vain for a reference so my recall must be questioned, but I used to follow patent discussion more closely years ago when I was writing up possible inventions. I remember an item but the seas have closed over the stone flung in the pond about relaxing that rule. I have found the patent world taken over by large corporations where the use of algorithms might well be afforded to gain advantage by sheer weight of numbers. And by threatening crippling law suits.

Quote:

---

Originally Posted by swampyankee

Currently, US patents can't be granted to corporations, either. Since an AI is not, by any current definition (even the idiotically expanded one of Citizens United) a person, but property, it can't own anything.

---

I think that is the futurology point I was hinting at, owning and inventing both.

Here is a discussion, even Abraham Lincoln is cited, and we know his view of the internet.
http://www3.weforum.org/docs/WEF_485...Patent_Law.pdf
So it’s a serious issue for some.

Quote:

---

Originally Posted by profloater

I was quoting and referring to the vernacular. But in detail it depends on the meaning of artificial and of intelligence. The good Patent office must have its own definition in order to make a rule.:)

---

Machines have already begun producing patentable drugs and designs.

Quote:

---

Originally Posted by Noclevername

Machines have already begun producing patentable drugs and designs.

---

But the issue of creativity arises. If the inventive step is obvious “to one-skilled in the art” Then no patent but if it is a machine , as in a Turing machine, another like machine would also produce the invention. Humans also tend to have the same idea at the same time and it’s first to submit, like first to publish in science. Drugs are interesting as a class since by definition they must have a physiological effect, otherwise the possible molecules list is infinite.

If I wrote the AI software that owns a patent, I could reprogram it to sell me the patent for a dollar.

Quote:

---

Originally Posted by profloater

Then no patent but if it is a machine , as in a Turing machine, another like machine would also produce the invention.

---

That's not how it happens in practice. Running thousands or millions of random trial-and-error simulations leads to chaotic patterns, resulting in different outcomes from different attempts. A butterfly effect, non-deterministic.

Quote:

---

Originally Posted by Noclevername

That's not how it happens in practice. Running thousands or millions of random trial-and-error simulations leads to chaotic patterns, resulting in different outcomes from different attempts. A butterfly effect, non-deterministic.

---

Invention is not random. Usually there is a problem noticed by a few. Ideas arise from parallel experience and some (people) have the inclination to make money from their version of the idea. To encourage this the state gives the incentive of exclusivity because inventions sometimes improve all our lives. If AI gets involved, it is more likely , i suggest, to have a wider knowledge base but no personal motivation. So it will produce a rational new invention.

The challenge, I suggest, is that the idea of novelty will become more difficult to judge. An AI solution to a stated problem, will be almost by definition, obvious. To another AI that is.

Quote:

---

Originally Posted by profloater

Invention is not random.

---

Discovery sometimes is.

Quote:

---

Originally Posted by profloater

I think that is the futurology point I was hinting at, owning and inventing both.

---

Presuming there is, at some time in the indefinite future, an AI that is human-equivalent we'll have to answer the question. Doubtless, it will generate levels of controversy at least as great as those involved in abolition and the civil rights movement, possibly with a level violence involved in the expansion of civil rights to non-whites in the US.

Quote:

---

Originally Posted by profloater

But if I claim a patent using knowledge from university, or books, or exposure to a problem brought to me, those influencers are not credited provided that the test of being non obvious is passed. I can see the logic of banning AI from invention but it is like Canute wanting to govern the tides.

---

Actually, my recollection of the story is not that Canute wanted to control the tides, but rather that he wanted to show that they should not venerate him as a god, by showing them that even he was powerless to control the waves.

Yes mine too, Canute wanted to put sycophantic advisors in their place. I do not want to stretch the Metaphor but the rapid use of data (avoiding AI as a confusion) is a Genie out of the bottle. Too many metaphors?

AI as scientist: A robot that performs experiments at an inhuman rate to discover new-to-us chemicals.

Nice reference, it illustrates the problem. There is an algorithm, the human input with the human objective. But no human predicts the result. The result is useful, can it be patented? I would expect the algorithm to be challenged as using prior art, or even obvious to one skilled in algorithms. But without patent protection, will the investment be made? Or is it open season for labs with robots?