This past February in Arlington, VA, Yolanda Gil (University of Southern California Information Sciences Institute) and Haym Hirsh (Rutgers University) co-organized a workshop on discovery informatics, assembling over 50 participants from academia, industry, and government “to investigate the opportunities that scientific discoveries present to information sciences and intelligent systems as a new area of research called discovery informatics.” A report summarizing the key themes that emerged during discussions at that workshop is now available.
From the executive summary:
- To improve computational discovery processes: We must understand how to make processes explicit, so they can be better managed and easily reapplied. Tools are being developed to automate or assist with specific aspects of these processes. We must develop a methodology to design these tools for usability, learn from what has worked and has not worked, and understand what features lead to broader adoption by scientists. We must reduce the effort needed to integrate different sets of tools to process data. Research must be carried out to facilitate the recording of provenance of scientific processes, making them inspectable and reproducible. We must develop user-centered design and visualization techniques that augment human abilities to analyze complex data with complex processes, and enable understanding and insight.
- To strengthen the interplay between models and data. Data is often separated from the models that explain it, hurting our ability to do science effectively. We must increase the expressiveness of model representations and their connections to data. We must map the landscape of different types of models and develop general mechanisms for automated or semi-automated model construction, data collection guided by models, and data analysis. We must design scalable methods to navigate large hypothesis spaces and their validation or disproval that results from data analysis.
- To manage human contributions and opening participation in science problems. We must innovate scientific processes by creating effective human-computer teams, where human creativity can complement brute force computation. We must invent new ways to approach scientific questions by opening up science and exposing the possibility of contributions from the broader public. We must develop a science of design for such systems, so we can understand the incentives, norms of behavior, and effective communication of tasks.
Advances in these areas will advance the practice of discovery Informatics in two ways: 1) improving existing discovery processes that are inefficient and suffer from human cognitive limitations, and 2) developing new discovery processes that increase our ability to understand challenging scientific phenomena. Further, outcomes in these areas are not domain specific, and can be leveraged across different sciences and engineering disciplines, having multiplicative returns, avoiding the inefficient, redundant development of computing innovations that would otherwise be repeated in specific disciplines (e.g., bio-, geo-, eco-informatics).
Check out the workshop website for the complete report, workshop materials, and other helpful information.
To capitalize on the momentum from the spring workshop, Gil, Hirsh, and others will run a symposium at AAAI this fall — to be held Nov. 2-4, 2012, in Arlington, VA — titled Discovery Informatics: The Role of AI Research in Innovating Scientific Processes.
(Contributed by Erwin Gianchandani, CCC Director)