Last year, I wrote about some thoughts on how generative models can be used to analyse patents. In a typical patent analysis project, most projects have less than 10,000 raw data points. Depending on the technology field, there are projects with 10,000 raw data or more, but if a large company dominates the market, the number of patents is often much higher than 10,000. If you think about it from a company's perspective, it's natural to want to see more patents. In the past, the desire to see more patents has been hampered by time and cost constraints, but with the rapid development of generative AI that can assist patent attorneys and researchers, it is now a reality.
The power of choice for businesses
The easiest option for businesses is to leverage existing commercial AI solutions in-house. You can directly utilise various AI services such as ChatGPT. You can enter patent contents in the interactive interface window or, if you have access to development personnel, you can plan and perform large-scale patent analysis tasks directly using API services. Since there is a high degree of freedom and you can enter or analyse patent documents in the form you want, it seems to be a possible scenario if you have internal experts who can manage and direct the direction of patent analysis. In terms of cost, if you use internal personnel, the cost of using AI services or API charges is negligible compared to using external experts in the form of consulting.
However, in general, most companies do not have experts who have both the ability to design prompts and the ability to analyse patents, which is necessary to establish the direction of patent analysis of AI solutions and validate them multiple times in the early stages. It is expected that only a few large organisations will be able to directly build or implement it. Even if the patent content is analysed, when the patent analysis results are finally obtained, it is necessary to review thousands to tens of thousands of AI analysis contents at a minimum level due to illusions or laziness, which can be a huge burden on the company.
Since the advent of LLMs, we have seen an unprecedented number of IT solution providers entering the patent industry and offering AI-based solutions specialising in patents and trademarks. In a previous article, I shared that the patent industry has been a neglected market, at least in terms of IT, but as LLMs have been developed to a commercial level, various services have emerged that use them to analyse companies based on patents, generate patent literature, expand patent searches, search for trademarks, and recommend products. Perhaps because patent and trademark literature managed by patent offices is inherently valuable as structured information, and because it can be easily provided as bulk data such as xml or structured information via APIs, various AI companies are looking at various opportunities. This is certainly something that patent attorneys should welcome, and it is clear that if more companies can develop various solutions in the patent field and provide sustainable services, the patent industry will be able to develop further.
AI-based solutions for patent analysis are also emerging, providing various functions such as finding relevant patents by entering information or keywords, expanding the search content, or broadening the scope of searched patent documents based on similarities between patent documents. However, it seems that it is difficult to standardise the service flow because companies have different purposes and directions for patent analysis. In order to analyse various information contained in a single patent in the desired direction and extract meaningful conclusions, it is necessary to have flexibility and scalability to accommodate the requirements of the company, and it is difficult to incorporate this into the user experience. Generally, these solutions are targeted at patent attorneys or patent firms, who are not the direct beneficiaries of the service, but rather the service providers who utilise the service to provide secondary services. This makes it important for patent firms to have the flexibility to utilise the service in a way that accommodates the varying needs of their clients.
However, the emphasis on flexibility makes it difficult to define the service flow, and it is difficult to capture the requirements of demanding customers and the demanding patent industry if the service is provided within a certain framework. Providing a solution that can only be serviced in a limited area as a vertical service could be an option, but there is a problem that the scale of IT in the patent industry is not yet large enough for vertical services to survive.
This makes it difficult for companies with patent analysis needs to choose a direct AI-based solution for patent analysis. As companies are already accustomed to receiving fully personalised patent analysis consulting outputs from patent attorneys, it is likely that the use of AI-based solutions for patent analysis will be limited to general patent trend analysis, as they are inevitably implemented with a certain degree of structuring.
AI's greatest strength is its universality and personalisation.
There are many advantages to using AI solutions, but in my opinion, the biggest strength of AI, especially generative AI, is personalisation. Of course, the outputs from general and short prompts may not be very different for each individual, but these general outputs can actually be considered as a level of information that can be obtained to some extent by Googling. This is one of the great strengths of generative AI models: the more granular the prompts, the more specific the individual's requirements, the more personalised the results. This is also the reason why general users are sometimes disappointed with ChatGPT, which is trained to be general-purpose, because when they type in a query without being specific, ChatGPT tends to give a largely abstract answer. Eventually, users want to be able to type shorter and get a more specific answer, and because ChatGPT is a general-purpose model, it tends to answer vague questions in a vague way because it is focused on being general-purpose, which means that it can answer anything that a general audience can ask it. So for customers who want sharper and sharper answers, more specialised services are still being tried, and if there are services that are better at certain domains or certain forms of answers, they are building their own niche.
The same is true for patent analysis using AI-based solutions for patent analysis. After all, every company's needs and circumstances are different, and the purpose of patent analysis is different, so it is unlikely that a general-purpose AI-based solution for patent analysis can properly analyse the nuances. It's a process of personalisation that needs to be tailored to your requirements, and in the end, a custom project will be more appropriate than a standardised solution. To put it in general IT industry language, AI-based solutions for patent analysis are likely to be more of an SI-type on-premise solution than a general-purpose, lightweight SaaS cloud solution. Looking back at BLT's many and varied patent analysis projects, it is difficult to standardise the needs of different clients, and even if the methodology of the project can be standardised, the target and direction of the project are different. Even if the patent analysis core is moved from the brain of a patent attorney to an LLM, it seems that a customised design is essential to clearly present a fine-grained analysis perspective from the scope of the patents to be analysed.
Security issues can also be a consideration. The accumulation of patent analysis perspectives and results from multiple companies in a single SaaS solution can be overwhelming for companies that are sensitive to information exposure. This may explain why large enterprises and public organisations remain conservative in their adoption of the cloud, or compromise with private clouds. Patent firms and patent attorneys are relatively immune to security concerns, not only because they are legally bound by confidentiality, but also because only a limited number of internal personnel have access to the information. Of course, using LLM for analytics means that certain information must be entered into LLM by the person performing the analysis. Of course, large-scale patent analysis cannot be performed on an interactive console, so it is premised on the use of APIs, but from a security perspective, APIs have higher guidelines for information control and security levels than interactive consoles, so I think it is desirable to use APIs if possible.
The best conditions for an LLM
In my personal opinion, there are many advantages of using LLM in patent analysis, but I think that the most effective analysis technique is to explore the organic relationship between components. As you know, a patented invention is composed of multiple components, and the core idea of a patent is the effect of the organic relationship between the components. Different patent attorneys have different depths and use different language to define the components. Therefore, it is necessary to specialise in analysing the organic relationship between the components to ensure that the component you are looking for is the one you are looking for. That's why I use a lot of variants in my search keyword combinations, which I call patent search formulas, so that I don't miss the same component with different keywords. In the LLM, even if you don't need to use variant keywords, you can infer the relationship between components based on the context of the text, classify them according to their function or role, and understand in more depth whether they are similar to the components you are looking for. The organic relationship between components and components, as well as the abstraction of the relationship in their respective languages, or the extraction of common elements to find semantic similarities, is what makes LLM stand out.
Of course, finding and analysing the organic combination of the above-mentioned components is a task that patent attorneys, who are experts in the field of patents, are good at, but there is a quantitative limit to looking at every single patent and extracting the combination of components or principles. Therefore, patent attorneys use various techniques to reduce the number of parameters in order to realistically reduce the scope of core or effective patents to focus on and examine, but it is difficult to deny the possibility of various variables or exceptions occurring in this process. The advantage of an LLM is that it can prevent such losses or omissions from occurring in the first place.
If we extract the intersection of the above conditions, the key point is that large-scale patent analysis projects using LLM should be entrusted to a company that has the capability to perform SI projects, but in reality, it is difficult to find such an institution, and there are not many actual cases because there is not much demand. Due to security issues, it is difficult to directly introduce the large-scale patent analysis project that BLT has carried out, but I would like to have the opportunity to talk about what level of analysis is possible with an LLM next time.
By Cheolhyun Yoo
#BLT #patentLawFirm #artificialIntelligence #collaborativeModel #innovation #deepLearning #machineLearning #patentIndustry #generativeArtificialIntelligence #GAI #naturalLanguageProcessing #bigData #patentAnalysis #artificialIntelligenceLimits #intellectualPropertyProfessionals #ethicalResponsibility #socialResponsibility #technologicalInnovation #professionalEthics #promptChaining #patentSpecification #framework #intellectualProperty #workEfficiency #patentSystem #customerExperience #userExperience #UX #experienceDesign #customerSatisfaction #filingDate #publication #patentability #patent #novelty #inventiveStep #riskManagement #IPRisk #portfolio #IP #business #intellectualProperty #patentLaw #corporateValue #listing #KOSPI #NASDAQ #private #startup #finance #investment #valuation
Last year, I wrote about some thoughts on how generative models can be used to analyse patents. In a typical patent analysis project, most projects have less than 10,000 raw data points. Depending on the technology field, there are projects with 10,000 raw data or more, but if a large company dominates the market, the number of patents is often much higher than 10,000. If you think about it from a company's perspective, it's natural to want to see more patents. In the past, the desire to see more patents has been hampered by time and cost constraints, but with the rapid development of generative AI that can assist patent attorneys and researchers, it is now a reality.
The power of choice for businesses
The easiest option for businesses is to leverage existing commercial AI solutions in-house. You can directly utilise various AI services such as ChatGPT. You can enter patent contents in the interactive interface window or, if you have access to development personnel, you can plan and perform large-scale patent analysis tasks directly using API services. Since there is a high degree of freedom and you can enter or analyse patent documents in the form you want, it seems to be a possible scenario if you have internal experts who can manage and direct the direction of patent analysis. In terms of cost, if you use internal personnel, the cost of using AI services or API charges is negligible compared to using external experts in the form of consulting.
However, in general, most companies do not have experts who have both the ability to design prompts and the ability to analyse patents, which is necessary to establish the direction of patent analysis of AI solutions and validate them multiple times in the early stages. It is expected that only a few large organisations will be able to directly build or implement it. Even if the patent content is analysed, when the patent analysis results are finally obtained, it is necessary to review thousands to tens of thousands of AI analysis contents at a minimum level due to illusions or laziness, which can be a huge burden on the company.
Since the advent of LLMs, we have seen an unprecedented number of IT solution providers entering the patent industry and offering AI-based solutions specialising in patents and trademarks. In a previous article, I shared that the patent industry has been a neglected market, at least in terms of IT, but as LLMs have been developed to a commercial level, various services have emerged that use them to analyse companies based on patents, generate patent literature, expand patent searches, search for trademarks, and recommend products. Perhaps because patent and trademark literature managed by patent offices is inherently valuable as structured information, and because it can be easily provided as bulk data such as xml or structured information via APIs, various AI companies are looking at various opportunities. This is certainly something that patent attorneys should welcome, and it is clear that if more companies can develop various solutions in the patent field and provide sustainable services, the patent industry will be able to develop further.
AI-based solutions for patent analysis are also emerging, providing various functions such as finding relevant patents by entering information or keywords, expanding the search content, or broadening the scope of searched patent documents based on similarities between patent documents. However, it seems that it is difficult to standardise the service flow because companies have different purposes and directions for patent analysis. In order to analyse various information contained in a single patent in the desired direction and extract meaningful conclusions, it is necessary to have flexibility and scalability to accommodate the requirements of the company, and it is difficult to incorporate this into the user experience. Generally, these solutions are targeted at patent attorneys or patent firms, who are not the direct beneficiaries of the service, but rather the service providers who utilise the service to provide secondary services. This makes it important for patent firms to have the flexibility to utilise the service in a way that accommodates the varying needs of their clients.
However, the emphasis on flexibility makes it difficult to define the service flow, and it is difficult to capture the requirements of demanding customers and the demanding patent industry if the service is provided within a certain framework. Providing a solution that can only be serviced in a limited area as a vertical service could be an option, but there is a problem that the scale of IT in the patent industry is not yet large enough for vertical services to survive.
This makes it difficult for companies with patent analysis needs to choose a direct AI-based solution for patent analysis. As companies are already accustomed to receiving fully personalised patent analysis consulting outputs from patent attorneys, it is likely that the use of AI-based solutions for patent analysis will be limited to general patent trend analysis, as they are inevitably implemented with a certain degree of structuring.
AI's greatest strength is its universality and personalisation.
There are many advantages to using AI solutions, but in my opinion, the biggest strength of AI, especially generative AI, is personalisation. Of course, the outputs from general and short prompts may not be very different for each individual, but these general outputs can actually be considered as a level of information that can be obtained to some extent by Googling. This is one of the great strengths of generative AI models: the more granular the prompts, the more specific the individual's requirements, the more personalised the results. This is also the reason why general users are sometimes disappointed with ChatGPT, which is trained to be general-purpose, because when they type in a query without being specific, ChatGPT tends to give a largely abstract answer. Eventually, users want to be able to type shorter and get a more specific answer, and because ChatGPT is a general-purpose model, it tends to answer vague questions in a vague way because it is focused on being general-purpose, which means that it can answer anything that a general audience can ask it. So for customers who want sharper and sharper answers, more specialised services are still being tried, and if there are services that are better at certain domains or certain forms of answers, they are building their own niche.
The same is true for patent analysis using AI-based solutions for patent analysis. After all, every company's needs and circumstances are different, and the purpose of patent analysis is different, so it is unlikely that a general-purpose AI-based solution for patent analysis can properly analyse the nuances. It's a process of personalisation that needs to be tailored to your requirements, and in the end, a custom project will be more appropriate than a standardised solution. To put it in general IT industry language, AI-based solutions for patent analysis are likely to be more of an SI-type on-premise solution than a general-purpose, lightweight SaaS cloud solution. Looking back at BLT's many and varied patent analysis projects, it is difficult to standardise the needs of different clients, and even if the methodology of the project can be standardised, the target and direction of the project are different. Even if the patent analysis core is moved from the brain of a patent attorney to an LLM, it seems that a customised design is essential to clearly present a fine-grained analysis perspective from the scope of the patents to be analysed.
Security issues can also be a consideration. The accumulation of patent analysis perspectives and results from multiple companies in a single SaaS solution can be overwhelming for companies that are sensitive to information exposure. This may explain why large enterprises and public organisations remain conservative in their adoption of the cloud, or compromise with private clouds. Patent firms and patent attorneys are relatively immune to security concerns, not only because they are legally bound by confidentiality, but also because only a limited number of internal personnel have access to the information. Of course, using LLM for analytics means that certain information must be entered into LLM by the person performing the analysis. Of course, large-scale patent analysis cannot be performed on an interactive console, so it is premised on the use of APIs, but from a security perspective, APIs have higher guidelines for information control and security levels than interactive consoles, so I think it is desirable to use APIs if possible.
The best conditions for an LLM
In my personal opinion, there are many advantages of using LLM in patent analysis, but I think that the most effective analysis technique is to explore the organic relationship between components. As you know, a patented invention is composed of multiple components, and the core idea of a patent is the effect of the organic relationship between the components. Different patent attorneys have different depths and use different language to define the components. Therefore, it is necessary to specialise in analysing the organic relationship between the components to ensure that the component you are looking for is the one you are looking for. That's why I use a lot of variants in my search keyword combinations, which I call patent search formulas, so that I don't miss the same component with different keywords. In the LLM, even if you don't need to use variant keywords, you can infer the relationship between components based on the context of the text, classify them according to their function or role, and understand in more depth whether they are similar to the components you are looking for. The organic relationship between components and components, as well as the abstraction of the relationship in their respective languages, or the extraction of common elements to find semantic similarities, is what makes LLM stand out.
Of course, finding and analysing the organic combination of the above-mentioned components is a task that patent attorneys, who are experts in the field of patents, are good at, but there is a quantitative limit to looking at every single patent and extracting the combination of components or principles. Therefore, patent attorneys use various techniques to reduce the number of parameters in order to realistically reduce the scope of core or effective patents to focus on and examine, but it is difficult to deny the possibility of various variables or exceptions occurring in this process. The advantage of an LLM is that it can prevent such losses or omissions from occurring in the first place.
If we extract the intersection of the above conditions, the key point is that large-scale patent analysis projects using LLM should be entrusted to a company that has the capability to perform SI projects, but in reality, it is difficult to find such an institution, and there are not many actual cases because there is not much demand. Due to security issues, it is difficult to directly introduce the large-scale patent analysis project that BLT has carried out, but I would like to have the opportunity to talk about what level of analysis is possible with an LLM next time.
By Cheolhyun Yoo
#BLT #patentLawFirm #artificialIntelligence #collaborativeModel #innovation #deepLearning #machineLearning #patentIndustry #generativeArtificialIntelligence #GAI #naturalLanguageProcessing #bigData #patentAnalysis #artificialIntelligenceLimits #intellectualPropertyProfessionals #ethicalResponsibility #socialResponsibility #technologicalInnovation #professionalEthics #promptChaining #patentSpecification #framework #intellectualProperty #workEfficiency #patentSystem #customerExperience #userExperience #UX #experienceDesign #customerSatisfaction #filingDate #publication #patentability #patent #novelty #inventiveStep #riskManagement #IPRisk #portfolio #IP #business #intellectualProperty #patentLaw #corporateValue #listing #KOSPI #NASDAQ #private #startup #finance #investment #valuation