AI-driven molecular design accelerates pharmaceutical innovation but risks reinforcing extractive R&D models and patent monopolies

Indigenous knowledge systems treat molecular design as part of a living ecosystem, where compounds are not isolated entities but interconnected with cultural and ecological contexts. The AI-driven approach risks erasing this wisdom by reducing molecules to data points for optimization, ignoring the sacred and communal dimensions of healing. Systems like the Andean *ayni* (reciprocity) or Māori *whakapapa* (genealogy) frame molecular discovery as relational, not transactional. However, these perspectives are systematically excluded from patentable R&D frameworks.

The commodification of molecular design traces back to 19th-century synthetic dye industries, where German chemists like Bayer patented compounds derived from colonial botanical exploitation. The 1980 Bayh-Dole Act in the U.S. further entrenched patent monopolies by allowing universities to privatize publicly funded research, accelerating the enclosure of molecular knowledge. AI now accelerates this cycle, with tools like AlphaFold2 trained on proprietary datasets, reinforcing path dependencies established during the pharmaceutical boom of the 1950s–80s.

Cross-culturally, the step-by-step design of molecules is not unique to Western chemistry; traditional systems like Ayurveda’s *Sankhya* philosophy or Traditional Chinese Medicine’s *yin-yang* alchemy also employ iterative refinement. However, these systems prioritize harmony with nature over control, contrasting with AI’s goal of optimization. The Japanese concept of *wa* (harmony) in material design offers a framework for balancing efficiency with ecological and social well-being. Yet, such paradigms are sidelined in favor of Western linear progress narratives.

AI’s role in molecular design is grounded in advances in generative models (e.g., diffusion-based retrosynthesis) and reinforcement learning, which can propose reaction pathways with high success rates. However, these tools rely on curated datasets that often exclude non-Western or traditional knowledge, introducing bias. The scientific community lacks standardized protocols for validating AI-generated molecules, raising concerns about reproducibility and safety. Additionally, the focus on step-by-step design overlooks the emergent properties of compounds in biological systems.

Artistic and spiritual traditions frame molecular design as an act of creation, not computation. In alchemical traditions, the *magnum opus* (great work) symbolizes transformation, where the chemist is both artist and mystic. Indigenous beadwork or textile patterns (e.g., Navajo *sash belts*) encode medicinal knowledge through geometric symbolism, a form of non-verbal molecular design. AI, by contrast, reduces this to algorithmic pattern recognition, stripping away the sacred and intuitive dimensions of discovery.

Future scenarios for AI in molecular design range from utopian open-access platforms to dystopian patent wars over life-saving compounds. A collaborative model could leverage federated learning to pool global datasets while respecting indigenous data sovereignty. Conversely, unchecked corporate control risks exacerbating the 10/90 gap in global health R&D, where 90% of funding targets diseases affecting 10% of the population. Scenario planning must account for climate-induced shifts in biodiversity, which could disrupt AI training datasets derived from endangered species.

Marginalized communities—particularly in the Global South—are excluded from the benefits of AI-driven molecular design despite bearing the brunt of its risks (e.g., patented drugs inaccessible to local populations). Women, who comprise 70% of the global health workforce, are underrepresented in AI training datasets and leadership roles in pharmaceutical R&D. Grassroots innovators in Africa and Latin America, such as those in the *African Centre for Biodiversity*, develop low-cost solutions but lack access to patenting systems. Their exclusion reinforces colonial-era knowledge hierarchies.