A few years ago, experts would say, AI couldn’t create images from text. It was too complex back then. Similarly, people in the patent fraternity brushed aside the rumors of patent searches using Artificial Intelligence.
But guess what, all have been proven wrong and it’s just beginning. It is true that AI won’t take our jobs but the person who knows how to use AI will. So, there are both positives and negatives but constant skill upgradation is the key.
When Google Patents was introduced, people thought that one day Google might introduce a tool for AI-powered patent and non-patent search. Later, it also started searching for scholarly articles. But still, the use of full-fledged artificial intelligence software in patent searches appeared to be in the distant future.
Fast forward to the present day, we tested freely available AI tools to contest our fears and find out opportunities in them. In this post, we will take up examples and discuss how far have we come. Moreover, we shall see what can happen in near future and what should you do.
Before conducting any kind of patent search, you should know what exactly you are looking for. Therefore, you need to extract the crux of the whole claim into just one or two lines.
So, without further ado, let’s start one of the most interesting pieces of discussion right away.
You all have heard about chatGPT and its extraordinary capabilities but possibly not about PQAI.
You all know that chatGPT is Chat Generative Pre-trained Transformer developed by OpenAI and works great when it comes to language-related tasks.
After knowing, the strength of chatGPT, a patent search professional would certainly think of using it to summarize and simplify the claims to just one or two sentences.
In fact, there have been many posts on the internet asking the chatGPT to explain a patent claim to a 10-year-old or explain it with real-world examples, etc.
However, the work of chatGPT as a language model ends there. Because when asked to find prior art for a claim or invention, it simply shows its inability as it doesn’t have access to patent databases.
Lamba AI-based Google’s Bard may solve this problem as it’d have access to the databases of Google Scholar, Google patents, and Google Books.
Anyway, if chatGPT doesn’t have access to patents then it doesn’t mean it’s useless for patent search. In fact, we can make it simplify the claim and feed it into another AI-based patent search tool called PQAI.
Before we proceed to show our experimental test search for validity and patentability, let’s first see what PQAI can do.
PQAI is patent quality through artificial intelligence, an initiative by AT&T. It appears to be a great tool for inventors and curious researchers who don’t know much about patent searching. It has many features which we discussed in next sections:
You can search by simply describing what you want to search for in a few sentences. For example, if you write about an invention related to self-driving cars as:
“A car switches to auto-pilot mode upon detection that the driver is unfit for driving.”
It gives you a list of relevant documents.
Along with providing a simple text sentence, you can specify the range of publication date of results, type of documents (patent or scholarly articles), search type (novelty or obviousness), etc.
When we clicked on the retrieved results to read about them, it took us to the Google Patents database, so it doesn’t seem to have its own database.
You can also find insights about the retrieved results which include top companies, top CPCs, publication trends, etc.
You can enter a technology description or a list of IPC/ CPC codes. Even a plain English description in a sentence would suffice. Upon clicking the search button, it returns probable, relevant CPCs with different options for viewing them.
For e.g., for an invention related to the intelligent self-driving car, we typed “A car switches to auto-pilot mode upon detection that the driver is unfit for driving” in the search box, and we got a list of CPCs.
It provides the option to see CPCs in split or joined form. Not only this, but it also provides an option to see CPC in short or full.
What is more interesting is that when you choose the short option, it displays the most relevant text portion of the class definition instead of the full definition.
For example, for the above invention of the intelligent self-driving car, the first CPC it gave was:
|G05D1/0061||Systems for controlling or regulating|
non-electric variables; Control of position,
course or altitude of land, water, air, or
space vehicles, e.g. automatic pilot; With
safety arrangements; for transition from
automatic pilot to manual pilot and vice
But when choosing the short option, it simply returns:
|G05D1/0061||For transition from automatic pilot to|
manual pilot and vice versa
This way, it makes your life easy and at no cost.
It derives technical concepts from the plain text query. You simply provide it with a simple English description in a sentence, and it returns keywords for concepts.
For example, for above intelligent self-driving car’s one-sentence description returns the following technical concepts:
driving, detection, mode, driver, car, switches
In short, if you have a technical feature or invention, it will give you technical concepts or keywords to expound upon. This should be pretty helpful in the search.
You can use these concepts for further related keyword suggestions.
Simply provide a target or technical concept to it, and it will return possible related keywords or synonyms related to the concept.
For example, we typed “coffee” in its search box. The related keywords, it suggested for our input were:
brewing, beverages, beans, hot beverages, beverage, maker, flavor, tea, roasting, drink
Similarly, we typed “phone” in it, and in return, we received the following related keywords:
portable device, calling, party, sent, call, voice, portable electronic device, alert, GPS, personal
Observing the above exercise, it is safe to say that as of now, it is not perfect but is intelligent. Since AI is supposed to learn and improve, it may be possible to have a pretty accurate AI-based search tool in the future.
2.5 GAUs (Group Art Units)
You can find USPTO Group Art Units related to the technical matter in a text excerpt. For example, for the input of “A coffee maker detects when the coffee pot is empty and then automatically turns off its heating element.” it returns 3 probable prior art units:
3761: Refrigeration, Vaporization, Ventilation, and Combustion
3742: Refrigeration, Vaporization, Ventilation, AndCombustion
3726: Manufacturing Devices & Processes, Machine Tools & Hand Tools
By now, we have seen all the main ingredients of chatGPT and PQAI required for the patent search experiment. So, now we can start our test case.
3. Validity and Patentability Search Using AI
Validity and patentability searches are similar in the sense that both searches require finding relevant documents. Similarly, there are many situations when you may need to find patent and scholarly articles. Thus, the scope of this blog post is to look for patents and scholarly articles using AI-powered tools.
For this experiment, we will be taking up an invention that goes something like this:
“modified headphone or earphone that can pick up sounds louder than a predetermined decimal limit, such as vehicle horns. In response, the headphone or earphone can alert the user about an approaching vehicle”
Consider this text written above as an abstract or embodiment of an invention. We take this text and ask the chaptGPT to explain it in one sentence as shown below:
Next, we take up the one sentence provided by the chatGPT and input it into the PQAI search box (see image 3.2). You can also observe, that we have selected the document type as “Patents” and the search type as “Novelty.” Because it can also provide document type as “Scholarly articles” and search type as “Obviousness“
Now, comes the interesting part: The list of search results and insights in Image 3.3. The first result we got was US pat. 7,015,812 B1 which appears to be a good document when it comes to relevancy. Not only this, but it has a mapping feature (see box 2) also to map out the invention features to prior art references.
When we click on “show mapping” in box 2 of image 3.3 above, we get the following query mapping to reference text which is quite amazing, given that the AI is still in its infancy phase.
The PQAI presents you with a list of documents based on its algorithm. If you like one of the results then it offers the option to show “more like this” (see box 3 in image 3.4 above). By clicking on it, you can get multiple similar documents. See image 3.5 below showing one of the documents retrieved by using the button “more like this”
From thus retrieved literature, you can shortlist relevant documents, and find good keywords, classes, inventors, assignees, etc.
On the search result page in image 3.3, you can click on “Insights” button (box 1) to see top companies, relevant CPCs, and publication trends.
By now, we have seen how to find relevant documents, their feature mapping, and insights. Now we will use PQAI’s AI-based feature to find CPCs. As an input prompt, we have entered the same description of the invention which we used for the search in image 3.2. You can also enter a list of IPC/CPC codes to find relevant classes.
In the list of classes, we have chosen to see their definitions in short (See box in image 3.7 above). There are other options also to view classes such as split, joined, and full.
If you are wondering what could be the technical concepts or keywords in an invention then this tool has got you covered at least to some extent. To test this feature, again we entered the same invention description of one sentence and clicked submit (see image 3.8)
To the query entered above, we get the following concepts:
headphone, modified, surroundings, allowing, exceeding, specific, alerting, sounds, vehicle horns, user, enhance safety, earphones, detected, aware
As you can see, with the list of concepts that we’ve got above, we have something to start while embarking on a search project.
We can further use PQAI to give synonyms to a concept keyword. For example, let’s pick “exceeding” from the previous list of image 3.8 above and submit it as a target concept as shown in image 3.9 below.
So far, we have seen the use of AI-based PQAI for patent search but we were amazed to see it providing the feature to search for non-patent literature as well. To test it out, we took our invention description as used in image 3.2 and entered it to search for NPLs.
The image below shows one such retrieved NPL result which appears to be very good prior art. Further, you can check out “mapping” and “more like this” features. See boxes 1, 2, 3, & 4 for clarity in image 3.10 below.
4. Future of Patent and Non-Patent Search (LaMDA-Powered Google’s Bard)
Based on what we have discussed in this article, it can be confidently concluded that the time has come when AI can search for complex topics like patents and scholarly articles.
This should worry a lot of patent search professionals. There are always opportunities in such a situation. AI itself can’t do much and its output depends on the quality of the input. Only a professional can provide quality prompts to it. So, let’s see what it contains when it comes to search jobs.
chatGPT is a great tool for language-based problems and the only factor that seems to limit its capability to search patents is that it doesn’t have access to patent databases. This was the only reason, we went to PQAI and used its ability to access patent databases.
Prima-facie, Google seems at an advantage when it comes to having access to required databases. It has the infrastructure already in place.
However, this doesn’t seem the case at least for now. We have experimented extensively with Bard since its release and have reached the conclusion that Bard is no better when it comes to searching prior art. However, these tools are in many ways turning out to be great assistants.
Also, there is constant debate on privacy of confidential data like our new inventions. What happens when we disclose our invention to these tools even before filing to the patent office? There is no “non-disclosure agreement” that would work here.
We have dealt with data privacy in the case of Bard extensively: Data Privacy with Bard: AI, Patents and Invention Confidentiality.
Moreover, we have also looked into the application of Bard in prior art search, infringement search, finding classifications, and writing claims: Bard AI and Patents: Patent Search, Infringement, Classification, Drafting.
It is interesting to note that the developers of PQAI understand confidentiality issues well, that’s why, their website mentions storing no search query at the backend.
In the future, chatGPT or tools based on chatGPT may gain access to patent and non-patent databases. Such a development has the potential to disrupt the patent sector, so it’d be worth waiting.
There are so many applications of AI being explored by patent professionals, for example, drafting patent claims, searching for infringement, preparing evidence of use mapping, etc. Every capable IP firm is rushing to develop AI-based IP tools.
Let’s see where these fast-paced developments are going to lead us. Adaptability is key to the feature. Lazy will get left out. So, be alert, prepared, and proactive. Focus on the advantage of early movers.
That’s it from us in this discussion. Hope, you’ve enjoyed reading. We have rich resources for learning patents. So, do check out our platform HavingIP
- Prior Art Search Free Guide 101: Do it Yourself
- Find Prior Art: 15 Free Patent Databases, AI Tools, and Search Engines
- A Guide to Google Patents Search Engine (Advanced)
- Data Privacy with Bard: AI, Patents and Invention Confidentiality
- Bard AI and Patents: Patent Search, Infringement, Classification, Drafting