Ethics of AI in Scientific Research: Case Study on AI-Accelerated Drug Discovery
Introduction:
The following case study was developed as part of independent research on the ethical use of artificial intelligence in scientific research, supported by the Hackworth Applied Ethics Research Grant through the Markkula Center for Applied Ethics at Santa Clara University. It is intended to prompt critical discussion among scientists, students, and faculty navigating the growing role of AI in research and drug development.
Case Study:
A pharmaceutical company is developing a new drug to serve as a treatment for a neurodegenerative disease. Traditionally, scientists work in research labs for years to identify and develop promising drug candidates (Source). This company, however, decided to use AI models to narrow down the number of potential molecules from thousands to just a handful of “high-probability” candidates. This was completed in only a few short weeks, although typically AI models take around 18 months to discover a drug (Source). The AI model used massive biological datasets from various sources to make its predictions about which molecules were most likely to be successful candidates. The pharmaceutical company claims that using AI for this purpose is beneficial because it speeds up the drug discovery and development process, allowing life-saving treatments to be brought to market faster, and therefore accessible to those in need of treatment. However, the company has also faced recent backlash, as some claim that the company is using AI to reduce research costs while cutting corners on traditional, human-led research. Critics of AI believe these AI models are not reliable enough to choose drug candidates and are skeptical of the data used to train these models. The pharmaceutical company stands firm in its practices and suggests that using AI to reduce research costs helps make the drug price more affordable to patients.
Discussion Questions:
How much validation is needed before moving AI-selected drugs into clinical (human) trials?
Who is responsible if an AI-accelerated drug causes harm?
How should the speed of drug development be weighed against a careful validation of an AI-accelerated drug, especially when human lives are at stake?
What are some important data privacy concerns regarding this case?
How short is “too short” for a drug discovery process accelerated by AI, and how should this timeline be determined?
What safeguards could be put into place to increase the safety of this situation?
What are some important questions we could ask about the biological datasets that the AI model used? (e.g., Where are these biological datasets sourced from? Are the datasets complete and diverse enough to properly inform the AI model?)