The AI Revolution is Here

Anchored by one of the world’s top computer science programs, The Grainger College of Engineering has been making AI breakthroughs for decades. Those achievements are reflected in the #5 ranking of UIUC’s artificial intelligence program by U.S. News & World Report in 2023.

Yet Grainger Engineering AI momentum continues to build: effort on AI is accelerating dramatically. Since 2019, Grainger Engineering has been awarded nearly $270 million in projects related to AI and digital transformation. Most recently, that includes support for four major new centers: the Amazon-Illinois Center on AI for Interactive Conversational Experiences (AICE), the AI Institute for Exceptional Education, the INVITE Institute (INclusiVe Intelligent Technologies for Education), and the IMMERSE Center for Immersive Computing. It also includes support for work in a range of other AI application areas to which UIUC researchers have long been making contributions, notably healthcare.

Amazon-Illinois Center on AI for Interactive Conversational Experiences (AICE)

The AICE center, is housed in the Department of Computer Science and the Coordinated Science Lab. AICE’s director, computer science professor Heng Ji, says that its overarching goal is “to advance conversational AI to the next level.” 

It will do so by funding projects proposed by teams of UIUC faculty collaborating with Amazon scientists. It also offers Ph.D. fellowships.

Each project will advance research towards at least one of three key goals.

The first is to make conversational AI far more knowledgeable. “An ideal conversational AI system should be able to learn knowledge online in a never-ending fashion,” says Ji. Such a system can support the user in all kinds of tasks by presenting relevant information gleaned from vast amounts of multimodal data observed online.

The second goal is to make a conversational AI system that users can interact with in a more richly multimodal fashion, not just by typing text or speaking. The system should feel like a “friend,” says Ji. She envisions, for example, a vision component that would allow you to hold up your cat and introduce it to a system that can see it.

Finally, AICE will work to make systems more fair and transparent. For example, a limitation of current systems is that they know only a handful of the most popular languages, and they struggle with dialects and accents. Ji would like AI systems to offer full understanding of the world’s languages and dialects.

AI Institute for Exceptional Education

This $20 million, nine-university consortium, funded by the NSF and led by the University of Buffalo, is also new in 2023. Its mission is to help children who have difficulty with speech and language processing.

Heng Ji is the PI of UIUC’s subaward. She explains that today’s manual screening and intervention are subjective and expensive; further, not enough experts are available. “The majority of kids, especially those from underrepresented groups, who have the early problems are not receiving any help,” she notes.

The goal is to automate both screening and intervention to make an affordable solution widely available.

The screening process will be informed by a comprehensive, AI-based analysis of the literature on speech-language pathology. It will absorb what’s known about application of various methods to various situations, so the tool can identify appropriate intervention methods based on conversations and other input.

The subsequent intervention will use an appealing robot that will “become the kid’s best friend, tutor, and study buddy,” says Ji. She notes that it will be challenging, because “kids usually don’t trust robots or take them seriously.” Therefore, a major research goal will be to figure out how to make the interactions engaging.

Notably, Ji is also the leader of a new $5.8M DARPA-funded project that aims to implement a concept learning framework that will help teach AI agents to interact successfully with people.

INVITE Institute (INclusiVe Intelligent Technologies for Education)

A third major new AI initiative is another $20 million, multi-institution, NSF-funded AI Institute. It is UIUC-led by principal investigator H. Chad Lane, a professor of educational psychology and computer science. Its intent is to improve students’ achievements in science, technology, engineering and math (STEM) fields, but its goals are much bolder than transmission of subject-matter knowledge. Rather, the team aims to train students in the foundational “noncognitive” skills they must have to succeed at learning. 

Co-principal investigator and Research Co-Director of the INVITE institute, ChengXiang Zhai, a Donald Biggar Willett Professor of Engineering in computer science, explains that research has shown that noncognitive skills, “particularly including persistence, resilience, and collaboration,” are essential to learning. However, “traditionally, people have studied how to use AI technologies to teach knowledge and improve cognition,” he says; the use of AI technologies to promote underlying noncognitive skills is novel. 

“The idea is to have an intelligent agent that can interact with learners to monitor their emotional state... and to provide a just-in-time intervention as needed,” says Zhai. 

He notes that the difficulty of modeling a learner’s state holistically, including both cognitive and noncognitive aspects, will be a big technical challenge. 

INVITE isn’t the first national AI Institute to be led by GCOE faculty. The $20 million USDA-funded Artificial Intelligence for Future Agricultural Resilience, Management, and Sustainability Institute (AIFARMS), led by Vikram Adve (Donald B. Gillies Professor in computer science), and the $20 million NSF-funded AI Institute for Molecular Discovery, Synthetic Strategy, and Manufacturing (Molecule Maker Lab), led by Huimin Zhao (Steven L. Miller Chair in chemical & biomolecular engineering), are two of the original seven AI Institutes announced in 2020.

AI for Healthcare

Jim Rehg, the incoming director of Grainger Engineering’s Health Care Engineering Systems Center (HCESC) and a Founder Professor in computer science and industrial & enterprise systems engineering, says that AI is critical to both of HCESC’s most active areas of research: health data analytics for mobile health and clinical data, and the use of extended reality (XR), which includes both virtual and augmented reality, to support physician training and novel methods for diagnosis and treatment.

With respect to health data analytics, “in medicine, often the challenge is that you don’t actually have large amounts of data; you may have a modest amount of data,” says Rehg. One must “be creative about how you train those models” to avoid problems created by the paucity of data. One strategy is to train models initially based on non-health-related data, and then use them to “bootstrap the creation of health-data-specific models.”

“With respect to AI, the exciting opportunities in XR pertain to how it can be used: how the content and the experiences that you have, when you put on and wear an XR headset, can be obtained by analyzing data,” says Rehg. Creation of that content has been a labor-intensive process, but AI offers the ability to create XR content through an automated process. If it’s successful, “that would free up the burden of creating that content manually through a very laborious process,” he says.

Inki Kim, HCESC’s Assistant Director of Research in Medical Simulation, is pursuing multiple AI projects for XR. Notably, FlightPath, an AI-driven tool developed by Kim and his collaborators, can diagnose concussions in as little as two minutes, and is now in clinical trials; a FlightPath app can be installed on Apple devices such as iPhones and communicates with a cloud-based AI model to diagnose the motions of a person responding to the app.

Kim thinks that adaptive agents—which modify their behaviors in response to their environments and other agents, including people—offer tremendous opportunities to “understand human performance.” He says that use of adaptive agents to elicit and reinforce human behavior is emerging as a major focus at HCESC as it turns towards “smarter” forms of simulation.

Many other AI health projects are ongoing in Grainger Engineering. For example, computer science professor Jimeng Sun is applying AI to a wide range of problems related to deep learning for drug discovery, clinical trial optimization, computational phenotyping, clinical predictive modeling, treatment recommendation, and health monitoring. And the group of Ravi Iyer, the George & Ann Fisher Distinguished Professor of Engineering in electrical & computer engineering, has been working for years with clinicians to build predictive AI models that are transforming patient diagnosis and care, including ones that successfully predict epileptic seizures, onset of Alzheimer’s, and drug responses for patients with depression and patients with triple-negative breast cancer.

IMMERSE Center for Immersive Computing

In 2023, the IMMERSE Center—supported by $4 million in campus Investment for Growth funds and helmed by Sarita Adve, the Richard T. Cheng Professor in computer science—was announced. As a core element of its research mission, it will harness advances in AI to rethink the foundations of immersive systems, such as virtual and augmented reality systems.

Immersive computing, whose goal is to provide users with a seamless integration of virtual and physical experiences, is starting to transform a wide range of industries, from education to healthcare to manufacturing and beyond. 

Adve says that the AI revolution is “making it possible to conceive of previously unthinkable functionalities.” For example, Shenlong Wang, assistant professor in computer science, is working on generating and rendering photorealistic immersive content directly from user-spoken input. However, these AI-driven innovations require huge amounts of computational power. IMMERSE is therefore pursuing latency-, bandwidth- and power-aware design of AI models as well as a fundamental reconsideration of immersive system design and evaluation. To achieve its aims, IMMERSE is bringing together researchers from across campus to develop end-to-end testbeds to prototype and evaluate their designs.

Breadth and Excellence

The above initiatives are only a sampling of the newest and biggest AI efforts in GCOE; a vast diversity of other AI work is ongoing. For example, Qing Cao (associate professor of materials science & engineering) recently found a way to reduce the computational costs of AI by achieving integration of ECRAMs—a key type of artificial synaptic accelerator that performs certain operations efficiently—onto silicon transistors. Lav Varshney (associate professor of electrical & computer engineering) and Nishant Garg (assistant professor of civil & environmental engineering) recently used AI modeling to identify a formula for concrete that reduces carbon emissions by 40% while maintaining strength and durability. And AI research in the IBM-Illinois Discovery Accelerator Institute spans a range of activities in the development and use of foundation models for code synthesis, materials discovery, and modeling of geospatial data for climate and sustainability; it also seeks to harness hybrid cloud technologies to advance computing platforms for training of foundation models and model inference. 

What’s the source of UIUC’s success in AI? Nancy Amato, Abel Bliss Professor of Engineering and head of computer science, thinks that UIUC’s tradition of interdisciplinary work has been the foundation of its research leadership in AI. 

“A lot of it is core CS, the theoretical foundations of AI. But to impact beyond CS requires partnering with the domain experts to really understand the problems and then come up with the solutions. And that’s something that’s been part of Illinois forever, and I think that’s why we’re really excelling.”