Can artificial intelligence (AI) provide a different perspective on patent searching?

Artificial intelligence (AI) is a hot topic at the moment - but what does it actually mean? And can and should we apply to patent searching? 

The answer to this last questions is 'yes', as we will show below -and for the very good reason that it can help searchers achieve their objective faster and more comprehensively than standard patent searching.

But firstly we should perhaps consider what exactly 'Artificial Intelligence' means. Definitions vary, but these include:

It is the science and engineering of making intelligent machines, especially intelligent computer programs. It is related to the similar task of using computers to understand human intelligence...... but AI does not have to confine itself to methods that are biologically observable. - Professor John McCarthy - Computer Science Dept Stanford University

Intelligence exhibited by machines. ... "intelligent agents": any device that perceives its environment and takes actions that maximize its chance of success at some goal.[1] ..". Wikipedia

Roughly speaking, it's technology that takes in huge amounts of information from a specific domain (say loan repayment histories) and uses it to make a decision in a specific case (whether to give an individual a loan) in the service of a specified goal (maximising profits for the lender). (New York Times, 2017).

Artificial intelligence is all around us

Whether we realize it or not, AI is all around us. When we type in a query into Google search, it considers both the query itself and as well as such things  as what it knows about us, where we are located,our past history with Google and a range of other factors, to formulate a list of results. If we go into an online e-commerce system like Amazon, AI is used to make predictions of what books we might like to buy. Our credit card transactions are constantly monitored by artificial intelligence looking for fraudulent behaviour.

But do we use AI in patent searching? We might start by considering conventional keyword based patent searching. 



Lets consider the typical conventional patent search. As an example, a searcher might decide say that they are looking for prior art patents to help invalidate US6956467, a patent filed for a Car alarm with automatic dialer. This has a priority date of 22nd March 2002, and claims a car alarm that can call your phone when triggered

In order to invalidate these patents, a searcher might search for patents published prior to 22 March 2002. So they would create a query, that might include a combination of a date filter, a patent class code filter, and a few keywords - maybe something like a search for "car alarm" and "mobile phone" OR "cellphone" OR "cellular" in either the title or the abstracts of the patents in the patent database.

This searcher would then use this query to search for some patents. The role of the query is to act as a filter, i.e. to identify patents that meet the criteria set out in the search query. The patent search database they are using would then find all of the patents that meet this specific criteria, and then return them as a list of patents. Often these patents are returned in perhaps the order of publication date, which can be helpful in some cases, but which may not prioritize in order of relevance to the objective of the searcher.

In this case, I ran the above query in  a conventional commercially available database, and this returned 18 patents in 11 families. But all searchers would recognise that some other queries can easily return hundreds of results, again not necessarily in a helpful order. 

Does conventional patent searching use artificial intelligence?

I would suggest not*. The role of the patent database was to follow the precise instructions given by the searcher, and return all patents that met these conditions. There was nothing artificial about this query (instead it entirely relied on the intelligence of the searcher), and there were no hints or additional information added by the database about which patents were likely to be most relevant.


HOW can we use artificial intelligence in patent searching?

Ambercite Ai has been developed to apply AI to patent searching - and in a very simple manner. All we need is one (or more) starting patents, and it will instantly find you a ranked list of similar patents. 

The query interface is very easy to use - all we need is the patent number being invalidated, and a data filter setting. For example, to look for patents that can invalidate US6956467, we simply need to set up the query below:

This qeury will return, not surprisingly, 100 similar patents, ranked in order of similarity toUS6956467.  


As an example of the sort of patents found by this process, consider US5081667 found in #5th place in this analysis, and disclosing System for integrating a cellular telephone with a vehicle security system.  Not only is it very relevant to US6956467, it has not been been previously cited as prior art, meaning that it could be used as the basis for reexamination requests at patent offices.

The similarity analysis is based on a sophisticated analysis of the forward and backward citations around this patent, and the forward and backward citations to these citations, and so on.  Because each citation link is an expert opinion that two patents are similar, by analysing a whole group of these opinions, and effectively combining these opinions, we end up with essentially a 'super opinion' of the most similar patents to the starting patent.

It is even possible to base this analysis on multiple starting patents, anything up to 200 starting patents in fact.  In the example below, we have added a similar patent to US6956467 to expand the breadth of the query. 


Is this artificial intelligence?

Consider the New York times definition of artificial intelligence given above - and change some words to reflect this current situation, as highlighted:

Roughly speaking, it's technology that takes in huge amounts of information from a specific domain (100+ million patents and 150+ million patent citations in the Ambercite database) and uses it to make a decision in a specific case (what the most similar patents are) in the service of a specified goal (looking for prior art to the starting patent).

Given this definition, Ambercite Ai is artificial intelligence, and the same would apply to the other two definitions given above. And it is entirely artificial - all we need is a patent number and a preferred date range for the results, and Ambercite Ai does the rest. The searcher has not needed to make any other decisions or assumptions regarding the patent search.


How does AI BENEFIT patent searching?

When compared to conventional patent searching, there are a number of benefits from the AI approach used by Ambercite:

  1. It will find you similar patents with different keywords and class codes compared to what you might have been expecting. No matter how diligent a searcher you might be, it is still possible to miss relevant patents that have unexpected key words or class codes. This has led Ambercite AI being described as 'insurance' or a 'safety net' for conventional searching. 
  2. It is fast and simple to use - and to learn to use. For patent invalidation, all you might need is the patent you are trying to invalidate (and any family member will do). A patent licensing analysis may only require the patents you are trying to license. A patentability search may require you find some starting patents, but you do not need to find a long list of starting patents in order to produce a long list of results. And once a query is set up, a list of result will be shown in seconds. Along the same lines, because it is so simple to use, it is also very simple to learn as well.
  3. Results are listed in a ranked order of patents according to predicted similarity. If you are short of time, just start at the top. 
  4. It integrates well with conventional searching.  As useful as it is, AI may not be the answer to every question. However it is very easy to use alongside conventional systems - patent numbers can be easily exported or imported to other systems, and there is even the capability to exclude other results you have found in previous results.

Given these benefits including its ease of use, it is not surprising that many users say:

"No search is complete without it".


Other uses for Ambercite Ai

The above example is for a patent invalidation search. Users also apply Ambercite Ai to patentability searching ( a preliminary keyword based search may be required to find some starting patents) and licensing searching. 


Do you want to add AI to your patent searching?

If so, please contact us at the details below, and we will set up you with a live demonstration and patents of your choosing. Corporate plans are available at a low annual price per user, and which will easily pay back this investment thanks to the value in the additional information found, as well at the time savings.


*What about other search systems?

There are some semantic search systems that do incorporate elements of AI into their processes - Ambercite Ai is compared to these in the link found here