Introduction

Artificial intelligence (AI) is a technology that enables computers and machines to simulate human learning, comprehension, problem solving, decision making, creativity and autonomy. As an example, the computing applications and computing devices which incorporate AI respond to humans in their languages. Currently, AI technology is being transformed by Generative Artificial Intelligence (Gen AI) which has the capability to create original text, images, video and other forms of content. The Gen AI uses deep learning models to create original content[1]. Artificial Intelligence (AI) is a rapidly growing industry and readily revolutionizing every sector of the economy. With its ability to drive technological advancements in almost every field, including healthcare, finance, and transportation, AI is becoming a cornerstone of future innovation. As the technology evolves, protecting inventions originating due to advancement in the AI technology becomes increasingly crucial to protect, particularly through patents. It is already known that patents safeguard intellectual property of AI developers and also promote further research and development. However, protecting the AI inventions needs careful assessment and planning, avoiding the pitfalls[2].

Patenting Artificial Intelligence inventions

This section presents some of the structured ways of protecting the Artificial Intelligence inventions. Every artificial intelligence system has at least a machine learning model and a lot of data which is used to train the machine learning model, which finally configures the AI model to be used for practical applications. The protection of mere data or type of data which is being used to train and derive results in an AI system would be difficult to patent. Likewise, the traditional models, which are being trained and used to input data along with the AI system, are likely to face hurdles in patenting. One of the ways to check patentability of an invention is to carefully understand the contribution of the invention and the associated uniqueness being contributed in the AI system. The AI system uses a prediction function with model parameters where the model parameters are adjusted based on training data set provided to the predicting function. This is known as the training phase. After sufficient accuracy is achieved, the prediction function may be used to provide real time analysis to new data and this is known as inference phase.

While drafting a patent application, aspects such as machine learning models using model parameters, prediction function, training phase, and inference phase are considered.

i) Machine Learning Models

The machine learning models involve use of features that are customized to achieve a desired prediction accuracy. Examples of such features can include arrangements of layers or nodes, activation functions, loss functions, training frameworks, data cleansing techniques, methods for defining feature vectors and/or methods of using hardware to execute an AI model more efficiently. One needs to spend considerable time and effort developing these advancements to produce a well-performing model. Even small changes or improvements to an AI technique may be patentable. For example, while large language models have recently become more popular and well known, still the models may be improved upon using tokenization techniques, self-attention mechanisms or transformer structures. No matter how incremental the contributions may be, it is worthwhile to consider protecting those efforts. One of the impediments would be that the machine learning models are algorithmic in nature and may be difficult to protect based on subject matter eligibility restrictions. However, incremental changes in the machine learning models may be worth protecting.

ii) Training Phase

In machine learning, training data is used to train the machine learning model for obtaining real world output in future. During the training phase, the system identifies patterns of the training data. The training data needs to be prepared and accordingly the data is pre-processed. The data is further transformed into a usable format for the machine learning model to comprehend. The transformed training data is fed to the machine learning model. The other operations involved require training the machine learning model based on the training data, testing the trained machine learning model using test data to identify whether a desired output is being obtained. Each of the available machine learning models may utilize different training processes. Exemplary training processes may include supervised training based on data which is already labelled. Another training process may be unsupervised training which uses training data which is not labelled. The machine learning model accordingly needs to be trained using the unlabelled data which involves more computations. The patent protection of the training phase may involve novel and inventive methods of preparing data for training which includes collecting relevant data, including relevant samples, data labelling, and putting data in standardized form and method of final training data generation. Secondly, the machine learning model runs on a neural network architecture, therefore there can be a novel neural network architecture which can be patented. Another aspect may be the optimization of the neural network architecture and the methods thereof. Moreover, the method involving efficient training of data as compared to the existing methods may be patented. The choice of the patentable subject matter needs to be carefully fine tuned based on the existing arts.

iii) Inference phase

Machine learning models can be implemented using software executing on a general-purpose computer, or using hardware, such as an application-specific integrated circuit (ASIC) or field programmable gate array (FPGA). Therefore, for inference phase use of the machine learning model, the claims may be directed around both method and apparatus. The inference phase of an AI system includes applying the trained machine learning models to make predictions, inferences, classifications, etc. This phase covers the real-world implementations and the insights obtained using the machine learning model based on user input. The inference phase can be claimed separately by defining the machine learning model used and the training of the data, based on where the novel and inventive features lie. The claiming of the inference phase helps in detection of infringement since the real application or output of the machine learning model on applying the data set can be viewed.

Patent Specification Considerations

The specification should apply the claimed technology to a real-world use case. To overcome potential objections related to the written description, the specification should disclose the technical details focussed towards the technology and the technical improvements. The specification must describe the computations of various parameters and variables and how the computation of the parameters and variables is performed. The technical features should be accompanied by the hardware elements used to define different variables and parameters involved in computation thereof. The specification must include the interaction between the hardware elements and the corresponding features being implemented in various computations. The specification should provide the technical effect using the technical features by using illustrative examples and comparing them with the existing art. This would be helpful in establishing the inventive step.

Subject Matter Eligibility Issues

The machine learning models involved in the AI inventions are algorithms in themselves and therefore are subjected to subject matter eligibility issues. If the invention is claimed as simply input-output approach without involving technical details, it has little chances of success since it is likely to be considered non-technical in nature. As discussed above, the AI inventions must have sufficient technical attributes directed towards novel and inventive features. In other words, the technical nature of the invention should be established.[3] To overcome the subject matter eligibility issues, the invention and accordingly the specification must establish the technical contribution by providing technical solution to associated technical problem.

Conclusion

AI is advancing at a rapid pace and there are millions of applications directly and indirectly attributed to AI inventions. As we gain experience in dealing with the technology, there would be new challenges based on the interpretation of technology and we look to see new guidelines to fine tune the patenting aspects of the AI inventions. The field would be dynamic to work upon and one needs to get updated on the latest developments in patent law and nuances of the technology. However, understanding the development that AI has brought to human being and also considering the irreversible impact suggest that the technology is here to stay and it would be better to adapt our laws and practices based on development of the technology.