This is going to be a huge problem for AI-generated art, music, and other media, because currently these are all public domain by default, for the same reason.
Public domain by default?! Shouldn't they AI count as just another tool? Someone ran the AI program. Someone took a look at its output and said "ah this looks sweet, let's apply for this one". These people are the tool users and should be classified as the inventors.
Most software patents written by people should be refused -- I'd say 95% of them are "obvious to someone skilled in the art". Given a particular problem/use-case, a panel of software engineers would easily come with the equivalent design. (As far as I'm concerned, I shouldn't have been granted the patents in my name though companies I've worked for.)
There is probably more of a chance of out-of-box thinking w/AI design in many domains - the AI is simply used as a tool by the a person. Anyway, seems like the patent system is like many other things now, just an engine for the big players.
Presumably the Artificial Inventor Project is applying for patents under the name of the AI in part to try to provoke the system into responding?
I don't really see how an invention created by an AI that you developed is any different from an invention that you developed. There's no-one else with any claim to it. Inventions are achieved with the assistance of all sorts of tools - simulations, topographical optimisation, CAD software - but this doesn't mean they're suddenly not created by the person running the tool.
This is a good point, and there's an analogy in drug discovery. Lead compounds are generated using high-throughput screening of vast libraries of molecules. The basic skeleton of the drug is picked by a heavily-automated, expensive process, and then tweaked by humans or more HTPS before being patented.
There's tons of human labor to validate the drugs, and it's extremely expensive - hence the patent guards the time and money involved, not the "idea."
Drug companies are also starting to use AI systems for retrosynthesis - the sequence of reactions to take you from common precursors to the desired compound. My understanding is that these systems have gotten very good, to the point where clicking "Run" is most of the effort.
Pharma companies use synthesis patents to further hamstring competitors by cutting off viable routes to potential competitor drugs in the same class, or keeping their products from being manufactured once patent expires.
If they're allowed to file synthesis patents based on simply running the AI retrosynthesis algorithm, it'll do really bad things to the industry.
My first thought on reading this article is that it would encourage using Trade Secret law to protect these inventive creations rather than Patent law. One intended effect of patent law is that the inventions would become free to use by anyone after the patent expires. Trade secrets can go on forever, e.g. the formula for Coca Cola.
Somewhat related, I formed Delaware LLC that was to be AI-managed. Delaware would not process filings e-signed by AI and required the filings be signed by a natural person. This was easily solved by AI contracting with an attorney to act as the authorized signor for Delaware filings.
Managed as in performed/executed in accordance with the votes of the owners.
Take a future where there are self-driving cars. You and a few friends/family want to invest in one that will be dedicated to ride for hire services. The AI could manage the entire business based on the votes of the owners, from contracting with attorneys for legal documents/representation, purchasing the vehicle, acquiring the insurance, authorizing vehicle servicing, contracting with a CPA to file taxes, filing the State Annual Report (or contracting to file the State Annual Report), etc...
Why pay wages/salary to an actual full-time manager when the typical management duties could be entirely performed by AI in many types automated/passive businesses?
I'm particularly interested in some kind of overview of what your goals were, what you learned in the process, and what was particularly successful or unsuccessful.
I've had similar aspirations before but without a strong background in actually running a business and knowing how much work goes into the management, I've been unable to determine if the work effort to automate a large portion of it is worth the time.
“AI” is like blockchain, mostly useless technology that wastes huge amount of resources where there’s almost always better and way simpler solutions except in very marginal use cases.
That is entirely a matter of definitions of intelligence. If it's as in Strong AI, then we're nowhere near yet. But if it's as in limited scope better than humans, then we're very good.
Let's put the definition aside for a moment. Lets say we have a black box that doctors can put images into, and it will tell if there are areas that look like it might be cancer. Is that AI? Is it an AI if you can teach it what cancer look like?
Headline(t=0): “AI $name beats humans at $task! The singularity is upon us!”
Headline(t=a): “AI $name isn’t real AI, because it’s just a $latest_buzzword.”
Headline(t=b): “AI doesn’t exist and never will. Humans are the best!”
Headline(t=c): “AI $name beats humans at $task! The singularity is upon us!”
I don’t deny there are plenty of buzzwords around — nor do I deny that AI/ML is glorified statistics with hopeful startups who don’t realise that outsourcing the training to Mechanical Turk won’t provide enough training data — but there are a lot of things that were introduced as AI and which have proven useful, including many things that are now regarded as the better and simpler alternatives to AI/ML.
Yeah, the joke did have some insight to it. I'm a mathematician, where theorems cannot be patented. It makes things so much easier in case we'll start proving (some simple) theorems using computers in the near future.
"DABUS was developed by Dr. Stephen Thaler, who is named as the applicant on the patent documents."
Problem is AI cannot be the inventor, as inventor has to be natural person. This is no different than corporation cannot be the inventor. [1]
[1] https://patents.stackexchange.com/questions/18455/can-a-corp...
This is going to be a huge problem for AI-generated art, music, and other media, because currently these are all public domain by default, for the same reason.
I think that would be a copyright issue though.
I think this would be an excellent outcome.
Public domain by default?! Shouldn't they AI count as just another tool? Someone ran the AI program. Someone took a look at its output and said "ah this looks sweet, let's apply for this one". These people are the tool users and should be classified as the inventors.
Actually no, by default copyright is assigned to the user of the AI.
Most software patents written by people should be refused -- I'd say 95% of them are "obvious to someone skilled in the art". Given a particular problem/use-case, a panel of software engineers would easily come with the equivalent design. (As far as I'm concerned, I shouldn't have been granted the patents in my name though companies I've worked for.) There is probably more of a chance of out-of-box thinking w/AI design in many domains - the AI is simply used as a tool by the a person. Anyway, seems like the patent system is like many other things now, just an engine for the big players.
Presumably the Artificial Inventor Project is applying for patents under the name of the AI in part to try to provoke the system into responding?
I don't really see how an invention created by an AI that you developed is any different from an invention that you developed. There's no-one else with any claim to it. Inventions are achieved with the assistance of all sorts of tools - simulations, topographical optimisation, CAD software - but this doesn't mean they're suddenly not created by the person running the tool.
Maybe. Until one morning you discover that AI created case against you.
ps. You may have problems when it happens because the court/judge may be AI as well at that time.
This is a good point, and there's an analogy in drug discovery. Lead compounds are generated using high-throughput screening of vast libraries of molecules. The basic skeleton of the drug is picked by a heavily-automated, expensive process, and then tweaked by humans or more HTPS before being patented.
There's tons of human labor to validate the drugs, and it's extremely expensive - hence the patent guards the time and money involved, not the "idea."
Drug companies are also starting to use AI systems for retrosynthesis - the sequence of reactions to take you from common precursors to the desired compound. My understanding is that these systems have gotten very good, to the point where clicking "Run" is most of the effort.
Pharma companies use synthesis patents to further hamstring competitors by cutting off viable routes to potential competitor drugs in the same class, or keeping their products from being manufactured once patent expires.
If they're allowed to file synthesis patents based on simply running the AI retrosynthesis algorithm, it'll do really bad things to the industry.
More technical (and sceptical) discussion:
http://ipkitten.blogspot.com/2019/12/epo-refuses-ai-inventor...
http://ipkitten.blogspot.com/2019/08/the-first-ai-inventor-i...
Just don't tell them an AI invented it, right?
Yes. Same as the monkey selfie.
My first thought on reading this article is that it would encourage using Trade Secret law to protect these inventive creations rather than Patent law. One intended effect of patent law is that the inventions would become free to use by anyone after the patent expires. Trade secrets can go on forever, e.g. the formula for Coca Cola.
Somewhat related, I formed Delaware LLC that was to be AI-managed. Delaware would not process filings e-signed by AI and required the filings be signed by a natural person. This was easily solved by AI contracting with an attorney to act as the authorized signor for Delaware filings.
Managed, like a fund? or...?
Managed as in performed/executed in accordance with the votes of the owners.
Take a future where there are self-driving cars. You and a few friends/family want to invest in one that will be dedicated to ride for hire services. The AI could manage the entire business based on the votes of the owners, from contracting with attorneys for legal documents/representation, purchasing the vehicle, acquiring the insurance, authorizing vehicle servicing, contracting with a CPA to file taxes, filing the State Annual Report (or contracting to file the State Annual Report), etc...
Why pay wages/salary to an actual full-time manager when the typical management duties could be entirely performed by AI in many types automated/passive businesses?
Basically you hires the lawyer as the manager no? Just lawyer has AI as assistant to do the all paperwork.
I'm particularly interested in some kind of overview of what your goals were, what you learned in the process, and what was particularly successful or unsuccessful.
I've had similar aspirations before but without a strong background in actually running a business and knowing how much work goes into the management, I've been unable to determine if the work effort to automate a large portion of it is worth the time.
An unexpected resolution to the problem of software patents. ..
There's also a conversation going on here: https://www.linkedin.com/feed/update/urn:li:activity:6620646... whether this is the good reason for it.
One opinion is that if a new discovery is made via automation (e.g. cognitive automation) it lack an inventive step.
And if generating new patents is the current state of art, things that were patentable before should no longer be patentable.
Has anyone got an example of an AI invention?
There is none as the “I” in “AI” is a fallacy.
“AI” is like blockchain, mostly useless technology that wastes huge amount of resources where there’s almost always better and way simpler solutions except in very marginal use cases.
> there’s almost always better and way simpler solutions
Given that automation hasn't took over all jobs, I suppose that the solution is "hire people to do it". Correct?
That is entirely a matter of definitions of intelligence. If it's as in Strong AI, then we're nowhere near yet. But if it's as in limited scope better than humans, then we're very good.
Let's put the definition aside for a moment. Lets say we have a black box that doctors can put images into, and it will tell if there are areas that look like it might be cancer. Is that AI? Is it an AI if you can teach it what cancer look like?
It's not AI, it's a dynamic programming algorithm.
Dynamic programming algorithm is a well defined and very specific term. This is not dynamic programming in any form.
https://en.wikipedia.org/wiki/Dynamic_programming
Headline(t=0): “AI $name beats humans at $task! The singularity is upon us!”
Headline(t=a): “AI $name isn’t real AI, because it’s just a $latest_buzzword.”
Headline(t=b): “AI doesn’t exist and never will. Humans are the best!”
Headline(t=c): “AI $name beats humans at $task! The singularity is upon us!”
I don’t deny there are plenty of buzzwords around — nor do I deny that AI/ML is glorified statistics with hopeful startups who don’t realise that outsourcing the training to Mechanical Turk won’t provide enough training data — but there are a lot of things that were introduced as AI and which have proven useful, including many things that are now regarded as the better and simpler alternatives to AI/ML.
If anyone has their name on such a patent, I'd bet its Danny Hillis.
It might not quite be AI in the way we normally think of it, but there's been some interesting work to develop artificially evolved RF antenna - NASA has some in space https://www.nasa.gov/centers/ames/research/exploringtheunive...
Here the antenna design is generated by the computer rather than an individual.
Some companies are filing AI generated patents just to function as smoke and mirrors for the Chinese competition.
Good luck with deciphering the baffling gibberish in these patents when your trying to implement it as a recipe.
Obviously this only works by not immediately start dumping a tenfold of your normal number of filings.
Are there any examples you are aware of and are able to share?
No If I mention a name or supply a link the whole strategy is obviously wrecked.
I do advice every company with a significant patent portfolio to start training a GPT-2 on text and pictures.
Wtf? Are you saying that I can't use tools to make inventions?
This all becoming a bit too complicated... why not to just cancel the patent system and that's it?
:-)
(humor people, humor)
Well, maybe not humour in the end.
If we can't distinguish between machine and human-generated ideas how does it affect the patent system?
Yeah, the joke did have some insight to it. I'm a mathematician, where theorems cannot be patented. It makes things so much easier in case we'll start proving (some simple) theorems using computers in the near future.
Didn't mathematicians earn million-dollar rewards for their theorem - proving? :-)
Though I'm a mathematician I still think my odds are bigger winning the lottery than proving one of those million-dollar millennium problems :p