Explainable Artificial Intelligence as a Legal Tool for AI-Assisted Pharmaceutical Patents: Lessening the Black-Box Problem Under 35 U.S.C. § 112

Abstract

The rapid integration of artificial intelligence (AI) into pharmaceutical research has accelerated drug discovery, but the opaque, "black-box" nature of deep learning models creates severe legal hurdles under 35 U.S.C. § 112. Because inventors often cannot explain how complex algorithms reach specific molecular predictions, AI-assisted patent applications risk rejection for failing to satisfy statutory written description and enablement requirements. Inventors are further constrained by the dilemma of either publicly disclosing proprietary source code or foregoing patent protection entirely. This paper examines whether Explainable Artificial Intelligence (XAI)—utilizing methods such as LIME and SHAP—can serve as a legally sufficient tool to mitigate this dilemma. By translating opaque computational decision-making into interpretable feature attributions and transparent reasoning pathways, XAI allows applicants to demonstrate possession and enable a person having ordinary skill in the art (PHOSITA) without compromising trade secrets. Although explainability cannot bypass core requirements like patent eligibility, novelty, or non-obviousness, it provides a practical evidentiary bridge that aligns AI-driven drug development with the public-disclosure bargain foundational to patent law.