Large Language Model ChatGPT has received an upgrade in the form of voice and image capabilities. OpenAI confirms that the language model offers a…
Not all software is patent-eligible, so it’s important that you know what to look out for.
In a previous article, I wrote about why your software startup needs a patent. Let’s look at what kinds of software one can patent.
Read more: 4 reasons why your startup needs a patent
As patent-eligibility differs from one country to the next, this article generally covers US and European eligibility requirements.
Taking an existing algorithm or method which is already known and applying it using a computer is not likely to be a patent-eligible invention, even if you are the first person to do so.
The same would be true if the algorithm could be performed by a human, for example using a pen and paper. There must be something more. Generally speaking, the algorithm must provide a “technical solution” to a “technical problem”.
‘An algorithm must provide a technical solution to a technical problem to be eligible for patent’
Here are some examples which should give some meaning to this otherwise nebulous phrase.
1. Filtering internet content
This example goes back a few years and addresses the issue of controlling access to internet content (by employers or parents). Solutions at the time either required software to be installed locally on each client computer or at a local server.
Problems associated with this approach were that the same filtering criteria were applicable to all users connecting through the client computer or local server, as well as time-consuming installation and maintenance.
The invention enables content filtering to be performed at the internet service provider’s (ISP) server and involves associating a network account to a filtering scheme and a set of logical filtering elements. Upon receiving network access requests from a client computer, the ISP server executes the associated filtering scheme using the logical filtering elements.
The ISP then determines whether the filtering scheme authorises the request and, if so, allows the requested content to be provided to the client computer.
Although challenged in court last year, this invention was confirmed to be patent-eligible because it provides a technology-based solution that overcomes the problem of filtering content on the web.
2. Automatic lip synchronisation and facial expression animation
The invention in this example is directed towards a method for automatically animating lip synchronisation and facial expression for three dimensional characters in movies.
The patented method essentially relates to setting “morph weights” (which relate to the facial expressions of an animated speaker as he or she speaks) and “transition parameters” (which relate to the sounds made while speaking) to produce lip synchronisation and facial expression control of an animated character.
The basis for awarding patent-eligibility is that the invention is directed towards an improvement in computer-related technology. The method allows a computer to produce more accurate and realistic lip synchronisation and facial expression in animated characters, which previously could only be done by human animators.
That is, the method enables the automation of specific animation tasks that previously could not be done by a computer. The technological process is improved.
An important take-home from this example is that an improvement in computer-related technology may be in the form of a set of rules that allow a computer to perform a function that it previously could not perform.
3. Training a neural network to retrieve semantically equivalent questions
This example addresses the problem of different users of a Q&A website formulating the same question in different ways. The differing formulations (vocabulary and structure) make searching for similar questions difficult.
The patented method trains a neural network to retrieve semantically equivalent questions. The method determines semantic similarity between a first question and a second question as a function of a first similarity and a second similarity.
The first similarity is computed based on weighted bag of words representations of the first and second questions. The second similarity is computed based on convolutional neural networks-based distributed vector representations of each of the two questions.
Parameters which are attributed to the first similarity and the second similarity in the function are learned by the neural network, which is transformed as the parameters are learned and updated.
This can be repeated for a number of questions to present a ranked list based on the semantic similarity determined between the first question and each of the other questions.
The problem of semantically equivalent questions not matching an input question is one that only exists in the realm of computing technology. The problem is addressed with a specific and clearly defined technical solution which contributes to the invention’s patent-eligibility.
While many computer programs are not patent-eligible, if your software is solving a technology-based problem, patents should be top-of- mind.
The next question: is it novel and inventive?
Stephen Middleton is a patent attorney at Von Seidels.