.png){kind=icon}
.png){kind=icon}
Return
Collaborative Intelligence in Sequential Experiments: A Humanin-the-Loop Framework for Drug Discovery
.png){kind=icon}
PRE
AI
DOI：10.1287/isre.2024.1154.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Drug discovery is a complex process that involves sequentially screening and examining a vast array of molecules to identify those with the target properties. This process faces challenges because of the vast search space, the rarity of target molecules, and constraints imposed by limited data and experimental budgets. To overcome these challenges, we propose a human-centered human-algorithm collaboration framework. Notably, both the algorithm and humans have substantial knowledge gaps as humans possess only partial domain understanding and no prelabeled data set exists to train the algorithm. The proposed algorithm continuously learns from experimental data to recommend molecules, and human experts maintain the final decision authority using their (private) domain expertise to override algorithmic suggestions. Our design artifact leverages dualprocess theory for attention management, surfaces meta-knowledge as a shared state for coordination, and optimizes a joint team objective. Through a comprehensive evaluation on real-world drug discovery tasks, we show that our proposed method consistently outperforms all baselines, including human-only and algorithm-only methods. This demonstrates complementary performance, in which team performance is higher than either the human or AI alone. Our findings offer insights into the role of (private) domain knowledge, meta-knowledge, and human agency, highlighting the potential of such a framework to accelerate vaccine and drug development by combining the best of human and artificial intelligence.
Keywords:
drug discovery
sequential experiments
human-in-the-loop
Bayesian neural networks
human-centered design
human-AI collaboration
Journal
IF:
5.1
Papers:
1.4K
Citations:
1.4W
Organization
View more organizations