Leena Mathur ’17
Current PhD Student, Carnegie Mellon University, School of Computer Science
Recent research intern, Robust. AI, early-stage startup building collaborative mobile robots that work with and alongside humans in warehouses, manufacturing and other settings
BS Computer Science, BA Cognitive Science, BA Linguistics, University of Southern California
Robots currently struggle to interpret the nuances of human communication. Human communication is multimodal, with an interplay of facial expressions, gestures and vocal cues that influence the meaning being created and perceived during interactions. Over thousands of years of evolution, humans have developed social intelligence, which we use daily. Machines currently lack this level of intelligence. My research advances computational foundations for multimodal social intelligence, to enable robots to better understand and respond to nuances in human communication, behavior and emotion.
Making AI more socially intelligent has been understudied. Current AI systems have shown great promise in mathematical reasoning, code generation and other tasks that are not explicitly grounded in social interactions. For example, these models currently do not perform well on the type of social reasoning required to analyze metaphorical language. To build a future with robots that assist humans in homes, hospitals and other spaces, it will be important for these machines to have social intelligence.
I recently interned at a startup that is pioneering collaborative mobile robots in warehouses. Robust.AI is an early-stage startup that builds robots that work with and alongside people in warehouses to enhance their productivity and well-being. During my summer research internship, I worked on perception (helping robots understand the world around them) and interaction (helping robots interact more seamlessly with humans).
Socially intelligent AI systems can improve human well-being. There is great potential to augment (not replace) human capabilities with robots. For example, in college, I worked on research to develop socially assistive robotic tutors for children with autism spectrum disorder. We found that children who typically avoided eye contact with humans would interact more willingly with the robots we placed in their homes. Over a month of these interactions, we found that children developed stronger social skills and began to engage more with the people around them.
It will require community effort to build and deploy AI systems that safely add value to people’s lives. For example, in a future scenario with companion robots helping the elderly, we will need to watch out for unintended consequences, such as increasing their isolation and emotional dependence on robots. In addition, there are ethical concerns related to collecting and using personal data to train these robots. There is a considerable gap between the science of building social intelligence in AI and responsibly deploying robots that add value to human lives. I am motivated to help bridge that gap and see socially intelligent machines broadly deployed to assist humans during my lifetime.