We envision that the future AI systems will "live" with humans in a complicated information-rich environment in the sense that close collaboration between AI systems and humans will happen in many ways, through which humans will receive services from AI systems (e.g., recommendation of useful information or support of a user task) and AI systems will learn from human feedback, both explicitly and implicitly. For example, a web search engine can be regarded as an intelligent sytem that serves user with relevant information on the Web while also learning from users' clickthroughs as feedback information to improve the accuracy of the search engine. AI systems can also facilitate human-human collaboration via infrastructures such as a social network, and one could also imagine multiple AI systems (intelligent agents) may also collaborate with each other, forming a multi-agent application system, where the agents may be homogeneous with the same or similar functions or complementary with each agent having a somewhat different skill. All such collaborations have to be facilitated by an infrastructure.

We are interested in designing innovative intelligent infrastructures to facilitate all kinds of collaboration including both AI-human collaboration and AI-facilitated human-human collaboration. An intelligent infrastructure also opens up opportunities for collecting data and developing new intelligent algorithms for optimization of human-AI collaboration and human-human collaboration, so we also work on Leveraging such intelligent infrastructures to deploy and evaluate those intelligent algorithms.

We are currently working on innovative infrastructures:

• Community Digital Libary (CDL): The Community Digital Library (CDL) is a novel and extensible platform for collaborative information seeking which enables any group of users to (1) describe and save webpages relevant to their shared interests, (2) share and search the saved webpages, and (3) discover content via recommendation. See this CSCW'23 Demo Paper for detail.
• Live Data Lab (LiveDataLab): LiveDataLab is a cloud-based open lab infrastructure enabling a self-sustainable ecosystem for integrating education, research and application development in data science all in one unified platform. It brings together multiple stakeholders including learners, educators, researchers, and industry application developers in an education and technology ecosystem where everyone is simultaneously a contributor and beneficiary. Learners use it to obtain project-based training using cutting-edge toolkits and real-world big data sets, while also contributing new annotated data sets and new algorithms; educators use it to deploy realistic data science assignments with leaderboard competition, while also contributing training projects; researchers leverage the new data sets created to conduct novel data science research and advance state of the art technology; industry partners leverage it to train and recruit relevant workforce and acquire state of the art technology developed by researchers and learners for use in their products, while also contributing realistic real-world data sets and tasks. As a result, LiveDataLab would enable data science education at large scale with low cost, minimize the gap between education, research, and applications of data science, and accelerate data science research as well as technology transfer.

  The LiveDataLab system was initially based on the CLaDS infrastructure, published in an ITiCSE'18 paper ). The current version of LiveDataLab is described in an ACM L@S 2019 demo paper. It has been used to support programming assignments for CS410 for several years now. The capability of LiveDataLab in supporting integration of education and application development has been recently demonstrated in a comprehensive search engine assignment of CS410 where the students have collectively built an expert search engine with algorithms designed by themselves via multiple modularized assignments such as data crawling, crowdsourcing relevance judgments, and competing with leaderboard to improve search engine algorithms. This experience has been described in a SIGCSE'20 paper