The disciplinary model of learning starts from deep knowledge in a specific domain and attempts to solve complex problems through interdisciplinary combinations of domain-specific knowledge. This talk proposes that disciplinary learning models can primarily address problems that are proximal to the involved disciplines.  The synthetic (human-environment-technology) and fast-evolving challenges of the 21^st century are high dimensional and not proximal to any one specific domain. They therefore require learning models that leverage the innate human ability to originate learning from the complexity and meaning of embodied real-life experiences (such as models suggested by Dewey, Polanyi, Schon, and others). Over the past 10 years, the Iovine Young Academy (IYA) at the University of Southern California (USC) has been developing and implementing a related model called Challenge-Based Reflective Learning (CBRL).  The primary aim of the CBRL is to cultivate expert techniques in sociotechnical creativity that can be applied to current and future synthetic challenges and the heterogeneous domain-specific knowledge they encompass. The talk will discuss how the IYA partners with industry to apply the CBRL pedagogy to challenges at the intersection of technology, human-centric design, and business (e.g. extended reality, product innovation, transformative AI, design strategy) and to assist students in developing their own innovation identity and related student-driven ventures.  The talk will also discuss the potential of CBRL to diversify learning.

The Minamata Convention on Mercury, a global treaty to which 148 countries (to date) committed to “phase-out and phase-down … mercury use in several products and processes,” inadvertently created a public health challenge. One of the mercury-containing products being phased down and out is dental amalgam, the oldest, most cost-effective, durable, and widely used restorative for dental caries, commonly known as “cavities.” While countries are stepping up caries prevention, these measures cannot eliminate dental caries entirely. As a result, tens of millions of people in the US and around the world are left with no good, affordable alternative to restore teeth affected by dental caries, other than tooth extraction, which leads to several other issues, both functional and social. Other restorative materials currently in use are more expensive, less durable, and more difficult to administer than amalgam. There are currently few efforts to identify a cost-effective amalgam substitute for resource-limited populations. Researchers at UB’s School of Dental Medicine (SDM) have joined with UB’s School of Engineering and Applied Sciences (SEAS) to identify and develop a cost-effective alternative. The initial work will test the feasibility of applying AI and high-throughput screening to dental-materials discovery, an application for which these approaches have not yet been widely employed.

Smarter chemistry is Apple’s approach to identifying chemicals that best serve all our priorities, including safety, performance, and environmental impact. This supports Apple’s circular supply chain efforts by minimizing the recirculation of potentially harmful substances. Apple’s efforts to remove potentially harmful chemicals from their products help create safer and healthier workplaces for their employees and suppliers, and safe products for people and the environment.

The labor market is experiencing one of the most transformative periods in history. Employers are adopting emerging artificial intelligence (AI) technologies in production and labor-management relations. The use of AI algorithms in hiring and evaluating workers is increasing, without a clear understanding of their effectiveness. The incidence of nonstandard work arrangements such as digital platform gig work and independent contract work has increased. The workers’ bargaining power has been eroding due to an increase in employer concentration and the decline of unions. There are complex interaction effects of emerging technologies in light of deteriorating workers’ power.  This talk will discuss several questions that are important to the current labor market.  What can we do to protect employment, workers’ rights, welfare, and job quality so that the advances in technology would be welfare-enhancing? What are the effects of emerging technologies on employment, productivity, and job skills and how do these vary with workers’ education, gender, or race? What are some implications of technological advances when companies spend inadequate amounts on workforce training? When is job training appropriate and effective and what policies have been effective for re-skilling workers?

As semiconductor technology nodes shrink, discovering and optimizing materials for these scales becomes increasingly complex. Data-driven innovation provides a robust toolkit to address these challenges. This presentation explores how machine learning, data analytics, and computational modeling accelerate materials development for advanced nodes. Critical applications include discovering materials with properties tailored to nanoscale dimensions, optimizing the integration of novel materials into existing processes, and maximizing manufacturing precision through data-driven control. In addition, specific challenges will be discussed, such as addressing potential bottlenecks in experimental validation and the need for specialized materials and informatics platforms for advanced node development. This approach has the potential to unlock the next generation of semiconductor materials for devices and processes, enabling unprecedented performance and functionality.

The United States recorded more than 42 trillion vehicle miles traveled in 2022 and, except for a small single-digit percentage of these miles, most were powered by internal combustion engine (ICE) vehicles. The ICE vehicle to electric vehicle (EV) transition is rapidly underway going by the annual growth of EVs compared to ICE vehicles, but this is a long transition as the average age of vehicles is more than12 years and there are about 283 million registered vehicles in the U.S. (mostly light-duty passenger vehicles). This industrial-scale transformation affecting trillions of miles traveled per year and several hundred million vehicles in the U. S. alone hasn’t been witnessed since the advent of the mass-produced horseless carriage in the early 20th century, when “vehicle miles traveled” and the human population was a fraction of what we have today.  At the center of all this is the need for continued materials innovation to drive the changes to energy storage, distribution, propulsion systems, computing, and communication that will be required to support the EV transition.  This is a design challenge to increase the energy density of EV batteries to reduce “range anxiety,” while bringing costs down; a design challenge to drive mineral and materials circularity. It is also the challenge to drive the materials innovation needed to develop more efficient “compute” stacks (from chips to “cloud”), semiconductors, and the quantum computing capabilities needed to support AI “everywhere.” This is the grand call to materials innovation, for an increasingly fragile planet with finite energy, finite resources, and a relentless need for humans or goods to travel trillions of miles a year, year after year.

There has been much discussion and fanfare around ChatGPT and large language models (LLMs) recently including their ability to respond to complex “prompts” without any additional training. These models are “pre-trained” on vast amounts of data spanning multiple domains and modalities.  However, there is also a need to incorporate datasets for specialized domains and tasks into pre-trained LLMs to improve their task-specific performance, specifically where complex reasoning is required. This talk will explore various methods of leveraging small, expert-curated datasets.  We start with effective methods of fine-tuning LLMs on datasets using supervised learning.  The focus of the talk will be on in-context learning (ICL) and new approaches for constructing and learning effective prompts through curated demonstrations.  Several applications of in-context learning will be presented including generating new datasets as well as complex reasoning tasks.  The talk concludes with recent innovations for enabling discovery through LLMs, including specialized decoding models capable of performing complex reasoning.

The phenomenal increase in compute power and storage, and the advent of new computational processing architectures, places us in a special circumstance where we are capable of enormous calculations that can provide insight into the most difficult of scientific and societal challenges. Our methods of addressing these challenges have, to date, been a mixed blessing. AI systems spot cancers that escape highly trained radiologists. The structure of large protein molecules can be predicted, allowing structural biologists to design medical treatments for life-threatening diseases. Yet these same methodologies can be hijacked without warning by poor data. These computational approaches often yield answers without providing scientific insight. And our misplaced trust in algorithms allowing them to operate without human oversight creates new problems for society.  We will discuss computational methodologies that might solve important questions in science and offer thoughts about the responsible use of these capabilities.