What are the limits to biomedical research acceleration through general-purpose AI?

Area of Science:

• Biomedical research
• Artificial intelligence
• Scientific discovery

Background:

• General-purpose artificial intelligence (GPAI) is anticipated to accelerate scientific discovery.
• The practical limitations of GPAI within the biomedical field are not well understood.

Purpose of the Study:

• To develop a framework assessing GPAI capabilities across the biomedical research lifecycle.
• To evaluate the potential speed increases and limitations of GPAI in biomedicine.

Main Methods:

• A scoping literature review was conducted to identify current and future GPAI capabilities.
• Expert elicitation with eight senior biomedical researchers was performed to gauge perceived acceleration potential and limitations.

Main Results:

• Current GPAI offers a ~2x speed increase; future GPAI could yield 25x for physical and 100x for cognitive tasks.
• Significant limitations include biological constraints, infrastructure, data access, and human oversight.
• Experts expressed skepticism about accelerating experiment design/execution but found acceleration in manuscript preparation plausible.
• Community assimilation of new tools was identified as a critical bottleneck.

Conclusions:

• Achieving substantial GPAI-driven acceleration in biomedicine requires addressing technological, infrastructural, and systemic factors.
• Targeted investment in shared automation infrastructure and reforms in research/publication practices are essential.
• Overcoming the human factor in tool adoption is crucial for realizing GPAI's full potential.