Patent Disclosure And Black-Box AI Systems: Doctrinal Breakdown And The Limits Of The Enablement Requirement

Full Text Link

Amirtha K, Presidency University, Bangalore

ABSTRACT

The rapid advancement of artificial intelligence (AI), particularly through machine learning systems, has begun to generate significant tensions within patent law, especially regarding disclosure requirements. Many of these systems exhibit opacity, unpredictability, and are heavily reliant on data, complicating their explanation in conventional terms. Central to patent disclosure is the enablement requirement, which mandates that inventors articulate their inventions sufficiently for a person possessing ordinary skill in the art (PHOSITA) to make and utilize them without undue experimentation.

The challenge arises with modern AI systems, particularly the so-called "black-box" models, which do not conform to these expectations. Their internal mechanisms are often difficult to interpret, challenging to reproduce, and sometimes not fully comprehensible even to their creators. This brings forth a crucial inquiry: how can one adequately "disclose" that which cannot be fully elucidated?

This paper examines the legal foundations of the enablement requirement and assesses its efficacy when applied to AI-based inventions. It posits that black-box AI reveals both conceptual and practical deficiencies within the current patent framework, thereby destabilizing the traditional quid pro quo of patent law. By analyzing legal developments, case law trends, and policy concerns, the paper contends that the existing approach to enablement may be insufficient for fostering AI innovation. Ultimately, it advocates for a re- evaluation of the framework, potentially by embracing more flexible disclosure models or alternative standards of reproducibility, to ensure that the patent system continues to promote innovation while preserving its core objectives.

Keywords: Artificial Intelligence (AI), Enablement Requirement, Black- Box Models, Patent Disclosure, Reproducibility in Machine Learning

533. PATENT DISCLOSURE AND BLACK-BOX AI SYSTEMS- DOCTRINAL BREAKDOWN AND THE LIMITS OF THE

Download PATENT DISCLOSURE AND BLACK-BOX AI SYSTEMS- DOCTRINAL BREAKDOWN AND THE LIMITS OF THE • 305KB