4 ms·

Machine learning is coming for the examiners jobs. :P

by nullc 4y ago

pessimizer 4y ago
You don't need machine learning to say yes to everything.
- nullc 4y ago
  now now, the examiners often say no a couple times before saying yes.
thallium205 4y ago
Probably one of the easiest but also higher paying jobs of the federal government. They seriously work maybe one day a week.
- _ggwa 4y ago
  Current USPTO patent examiner here. The job is regarded as difficult by the vast majority of examiners, as it's based on a quota system and examiners don't get enough time to do a quality job. I don't know of a single examiner who works only one day a week. Many work unpaid overtime to meet their quotas, euphemistically called "voluntary overtime", which is rarely voluntary. There's about 50% attrition in the first year for new hires right now. That sound easy to you? I don't know any other job that hires white-collar professionals with such high attrition. I encourage anyone who believes being a patent examiner is an easy job to read this subreddit: https://www.reddit.com/r/patentexaminer/ https://www.reddit.com/r/patentexaminer/ You can find patent examiners who say that being a patent examiner is relatively easy. These folks typically changed from a notoriously difficult job in law or academia. So it's only relatively easier. Beyond these cases, there do seem to be some examiners who have an easy time, but that's quite rare in my experience, and I'd question the quality of their work. As for why I took the job: I came from academia, so the job is easier in some respects. The quota isn't entirely bad: It's hard to meet, yes, but it's also mostly objective. If an examiner's numbers are good, the USPTO is happy with them. That contrasts with my experience in academia, where one's job performance is often subjective. If a researcher's boss decides they don't like the researcher for whatever reason (office politics, the researcher is socially awkward, etc.), there might be nothing the researcher could do to recover from that. The job also does give me a lot of freedom in other ways: I can live mostly anywhere I want to, I'm not expected to work all the time (production numbers beat looking busy by working all the time), there are few IP restrictions (aside from that I can't get a US patent), etc. The main problems for me are that the quota is too high and that I don't care for the technology I'm assigned (though the technology could be a lot worse). (It should go without saying that this comment is my own opinion and not that of the USPTO, US govt., my previous employers, etc.)
pbhjpbhj 4y ago
Some aspects of patent administration are being assisted by machine learning, which might mean that statistically a few jobs have already been lost, maybe not examiners per se. I gather IPCCAT https://ipcpub.wipo.int/?notion=search&version=20220101&symbol=none&menulang=en&lang=en&viewmode=f&fipcpc=no&showdeleted=yes&indexes=no&headings=yes&notes=yes&direction=o2n&initial=A&cwid=none&tree=no&searchmode=ipccat https://ipcpub.wipo.int/?notion=search&version=20220101&symb..., uses ML - and other offices have similar tools. UKIPO use AI for trademark searches and to streamline applications, eg https://ipo.blog.gov.uk/2020/10/29/introducing-the-trade-mark-pre-apply-service/ https://ipo.blog.gov.uk/2020/10/29/introducing-the-trade-mar.... Patent examining seems pretty complex analytically, perhaps some aspects of patent drafting, making a common description from a bunch of 3D renders (like scene description) might be the next low-hanging fruit after classification?
- _ggwa 4y ago
  The USPTO uses a lot of automation, but frequently not where they should. Note: I am a current USPTO patent examiner, and this is my opinion, not that of the USPTO or US govt. The USPTO apparently has two contractors to classify patent documents. I've heard that some sort of AI system is used for classification, in combination with a lot of poorly paid contractors. In my experience, the classification is so frequently wrong that this is clearly not working. It might seem okay to upper management, who never has to actually deal with the classification being inaccurate. But examiners aren't happy with it. Many people are calling for AI search. The new head of the USPTO mentioned it during a recent all-hands meeting. Unfortunately, the people who propose AI search don't seem to realize that 1. the USPTO has at least 5 AI search tools at their disposal (PLUS, More Like This, Dialog's similarity search, IP.com's similarity search, and Google Patents similar documents) and 2. none of these AI search tools work that well. In my experience, most of the time these tools don't return useful documents. (I still try them for every application as there's little downside.) The documents are usually close but it's rare that I'll actually use one of these documents in a prior art rejection. AI search sounds good to people who have never searched for patents and particularly have never used the existing AI search tools. AI search technology probably won't be good for a decade or more. In contrast, tools to analyze patent claims for various problems (basically, linters) have been available for around 30 years and can be quite useful in my opinion. But the USPTO has no such tool available to examiners, and analysis under 112(b), etc., is almost always done manually. I wrote my own tool, which I run on my USPTO computer on a regular basis. There are a huge number of opportunities to streamline USPTO operations with automation. Why are IDS forms not computer-readable? Why is so much information not auto-filled? Why do I have to fill out "bib data sheets" for every application? Why do I have to manually upload my search history when the system could easily automatically grab it for me? Etc. And automation isn't enough. There should be more data validation in the process, as a lot of problems can be automatically caught at the time of filing or when I post an office action. That's when fixing these problems would be easiest.

Machine learning is coming for the examiners jobs. :P

by nullc 4y ago

Machine learning is coming for the examiners jobs. :P

pessimizer 4y ago

You don't need machine learning to say yes to everything.

nullc 4y ago
now now, the examiners often say no a couple times before saying yes.

thallium205 4y ago

Probably one of the easiest but also higher paying jobs of the federal government. They seriously work maybe one day a week.

_ggwa 4y ago
Current USPTO patent examiner here. The job is regarded as difficult by the vast majority of examiners, as it's based on a quota system and examiners don't get enough time to do a quality job. I don't know of a single examiner who works only one day a week. Many work unpaid overtime to meet their quotas, euphemistically called "voluntary overtime", which is rarely voluntary. There's about 50% attrition in the first year for new hires right now. That sound easy to you? I don't know any other job that hires white-collar professionals with such high attrition. I encourage anyone who believes being a patent examiner is an easy job to read this subreddit: https://www.reddit.com/r/patentexaminer/ https://www.reddit.com/r/patentexaminer/ You can find patent examiners who say that being a patent examiner is relatively easy. These folks typically changed from a notoriously difficult job in law or academia. So it's only relatively easier. Beyond these cases, there do seem to be some examiners who have an easy time, but that's quite rare in my experience, and I'd question the quality of their work. As for why I took the job: I came from academia, so the job is easier in some respects. The quota isn't entirely bad: It's hard to meet, yes, but it's also mostly objective. If an examiner's numbers are good, the USPTO is happy with them. That contrasts with my experience in academia, where one's job performance is often subjective. If a researcher's boss decides they don't like the researcher for whatever reason (office politics, the researcher is socially awkward, etc.), there might be nothing the researcher could do to recover from that. The job also does give me a lot of freedom in other ways: I can live mostly anywhere I want to, I'm not expected to work all the time (production numbers beat looking busy by working all the time), there are few IP restrictions (aside from that I can't get a US patent), etc. The main problems for me are that the quota is too high and that I don't care for the technology I'm assigned (though the technology could be a lot worse). (It should go without saying that this comment is my own opinion and not that of the USPTO, US govt., my previous employers, etc.)

pbhjpbhj 4y ago

Some aspects of patent administration are being assisted by machine learning, which might mean that statistically a few jobs have already been lost, maybe not examiners per se. I gather IPCCAT https://ipcpub.wipo.int/?notion=search&version=20220101&symbol=none&menulang=en&lang=en&viewmode=f&fipcpc=no&showdeleted=yes&indexes=no&headings=yes&notes=yes&direction=o2n&initial=A&cwid=none&tree=no&searchmode=ipccat https://ipcpub.wipo.int/?notion=search&version=20220101&symb..., uses ML - and other offices have similar tools. UKIPO use AI for trademark searches and to streamline applications, eg https://ipo.blog.gov.uk/2020/10/29/introducing-the-trade-mark-pre-apply-service/ https://ipo.blog.gov.uk/2020/10/29/introducing-the-trade-mar.... Patent examining seems pretty complex analytically, perhaps some aspects of patent drafting, making a common description from a bunch of 3D renders (like scene description) might be the next low-hanging fruit after classification?

_ggwa 4y ago
The USPTO uses a lot of automation, but frequently not where they should. Note: I am a current USPTO patent examiner, and this is my opinion, not that of the USPTO or US govt. The USPTO apparently has two contractors to classify patent documents. I've heard that some sort of AI system is used for classification, in combination with a lot of poorly paid contractors. In my experience, the classification is so frequently wrong that this is clearly not working. It might seem okay to upper management, who never has to actually deal with the classification being inaccurate. But examiners aren't happy with it. Many people are calling for AI search. The new head of the USPTO mentioned it during a recent all-hands meeting. Unfortunately, the people who propose AI search don't seem to realize that 1. the USPTO has at least 5 AI search tools at their disposal (PLUS, More Like This, Dialog's similarity search, IP.com's similarity search, and Google Patents similar documents) and 2. none of these AI search tools work that well. In my experience, most of the time these tools don't return useful documents. (I still try them for every application as there's little downside.) The documents are usually close but it's rare that I'll actually use one of these documents in a prior art rejection. AI search sounds good to people who have never searched for patents and particularly have never used the existing AI search tools. AI search technology probably won't be good for a decade or more. In contrast, tools to analyze patent claims for various problems (basically, linters) have been available for around 30 years and can be quite useful in my opinion. But the USPTO has no such tool available to examiners, and analysis under 112(b), etc., is almost always done manually. I wrote my own tool, which I run on my USPTO computer on a regular basis. There are a huge number of opportunities to streamline USPTO operations with automation. Why are IDS forms not computer-readable? Why is so much information not auto-filled? Why do I have to fill out "bib data sheets" for every application? Why do I have to manually upload my search history when the system could easily automatically grab it for me? Etc. And automation isn't enough. There should be more data validation in the process, as a lot of problems can be automatically caught at the time of filing or when I post an office action. That's when fixing these problems would be easiest.