The world is constantly evolving, and we have seen some profound technological changes and some brilliant scientific concepts. Therefore, it is crucial to evaluate the legal implications of technological breakthroughs. One such innovation is Artificial Intelligence. The history of the word ‘Artificial Intelligence’ can be traced back to the year 1956 when it was first coined at Dartmouth College. The intelligence generated by “machines” is essentially what artificial intelligence (AI) refers to. The term “intellectual property” (IP) refers to any genuine invention of human intelligence, including works of art, literature, technology, and science.
Artificial intelligence may transform into a new species and enter the field of intellectual property. Given how swiftly. It is crucial to recognize the issues with the system and modify some of the existing Intellectual Property norms. The emergence of artificial intelligence (AI) as a multipurpose technology with significant economic and social applications raises basic concerns that are at the core of the current IP systems. This article makes an effort to address the issue of sufficiency of disclosure in AI inventions, which is vital to the core principles of intellectual property rights but is not yet at a certain stage.
DIFFICULTY IN MEETING THE DISCLOSURE REQUIREMENTS
It can be challenging to satisfy the disclosure requirements when trying to patent AI-based ideas. The patent laws have a quid pro quo at their core. An inventor is required to divulge to the public sufficient information about the invention to let one of ordinary skill in the art to put what is claimed into practise in exchange for a limited monopoly via a grant to prevent others from using it. Meeting this condition can be difficult, given the characteristics of some AI innovations. The primary reason behind the creation of patent laws was to provide an incentive for invention and public disclosure of the invention.
AI-assisted inventions provide humans with control over the process and the parameters that determine the product, with varying degrees of success. The capacity of various software/algorithms used in the process could potentially have an impact on how tightly the parameters are managed. For AI-assisted inventions, the disclosure of human involvement and the nature of the software may be of utmost significance.
The neural network design and the process used to train the data may be made public if the technology is used to produce AI-generated creations independently. Due to the vast amounts of data and several iterations of findings, the training process for developing an idea or product may be challenging to divulge.
By contrast, the machine learning model is the “black box” aspect of AI systems. As stated by IBM, the largest patent filer for AI inventions in the U.S., “AI inventions can be difficult to fully disclose because even though the input and output may be known by the inventor, the logic in between is in some respects unknown.” The fact that modern models like deep learning neural networks are so complex is one reason why the model is seen as a “black box.” In general, AI poses a transparency dilemma since more knowledge about the underlying, opaque workings of AI generates benefits but also dangers and costs.
The machine’s black box processes, which frequently result in AI innovations, make it impossible to divulge the innovations in sufficient detail to comply with existing rules. The reversion to secrecy runs against one of the principles of the IP system, where public disclosure is a condition for limited protection. The world might need other forms of protection that do not currently exist.
In some circumstances, disclosing training data in addition to the algorithms might also be necessary. However, the release of training data could come with its own set of obstacles, including:
- Privacy and confidentiality concerns for the data involved; and
- Third-party from where the data is fetched might not be willing to share the data.
SHOULD THE HUMAN EXPERTISE THAT WAS UTILISED TO CHOOSE THE DATA AND TRAIN THE ALGORITHM BE DISCLOSED?
The developer of the AI system could be reluctant to disclose the data out of concern that a rival might use it to train a different AI system in a shorter period. Since it would be sufficient to recreate the invention without using the training data, the variables utilized in choosing the training data may alternatively be disclosed in place of the training data.
In the Indian context also, in a patent application, the applicant only mentioned that he used multimodal deep learning techniques to model the data utilizing a sizable amount of datasets, but he made no mention of the reasoning, the Artificial Neural Network (ANN) employed, to generate the output. This is because humans can’t gather and analyze the information between deep learning layers or understand why the AI’s underlying algorithm evolved the weights associated with such deep learning levels. Even AI’s operations are becoming more non-linear due to the constant ingestion of datasets, increasing the need for humans to be able to explain how one variable relates to another. This lack of discernment and inexplicability on the part of the patent applicant contributes to the failure of the sufficient disclosure requirement.
The court in the case of Vasudevan Software, Inc v MicroStrategy, Inc that a mere disclosure of the “desired result” does not satisfy Section 112 of the 35 U.S.C., containing the written-description requirements because the description requirement of the patent statute requires a description of an invention, not an indication of a result that one might achieve if that invention were to be made. The “Artificial Neural Network” module was described in US Patent No. 6,792,412 as an AI-based tool that provided a rating value for a variety of items by recalculating the weights. However, due to the network’s capacity for self-learning, the underlying method, the pattern incorporated in the data, and the capture of the statistical structure in a set of data representing observable variables were humanly unintelligible. Consequently, because of the black-box-like and non-deterministic characteristics of the model, a disclosure for a model-related claim may be found a mere disclosure of the “desired results,” which leads to the insufficiency of the required disclosure.
Because the particular properties of the end state cannot be defined by the inventive method that produced it, the complexity of AI necessitates strengthening the disclosure requirement. One method is to apply a deposit mechanism, similar to the Budapest Treaty, to inventions assisted or created by AI. The Budapest Treaty is concerned with the international patent process of microorganisms. The treaty eliminated the requirement that microorganisms be deposited in each nation where patent protection is sought. According to the agreement, a microorganism’s deposit with an “international depositary authority” satisfies the deposit requirements of treaty members’ national patent laws. Similarly, adopting a globally recognized standard will help ensure that such a system is implemented successfully because a single worldwide standard could apply to both types of deposits, AI/algorithms, and training data. Greater international convergence and regional collaboration would be beneficial and should eventually increase legal certainty.
The existing Intellectual Property (IP) laws are not competent to address the problems due to the advancement of technology and the increasing use of AI in IPR. Policy and legislative changes are required to deal with these issues. The quid pro quo, in which an inventor discloses his or her creation to the public in exchange for exclusive rights to use it for a set period of time, is a core tenet of most patent regimes. Recent developments in artificial intelligence (AI) have raised questions about whether the public can benefit from AI discoveries to the point where the granting of the exclusive right is justifiable under the current patent disclosure rules. The “black box” aspect of a specific sort of AI, which makes it challenging to comply with the current disclosure laws in several jurisdictions, is the inevitable focal point of this discussion. As a result, it is unclear what, in the context of AI inventions, constitutes appropriate disclosure under the IP Regimes. The adoption of a globally recognized standard for disclosure is the best method in this regard.
B.A. LL.B.(Hons.) student at Dr. Ram Manohar Lohiya National Law University, Lucknow