Artificial intelligence (AI)-driven personal assistants like Apple’s Siri and Amazon’s Alexa have gained significant popularity due to their user-friendly natural-language interfaces that facilitate interaction with the Internet. These agents, embedded in various devices, assist users with tasks such as retrieving information, controlling home automation, and managing calendars. However, current assistants lack awareness of the physical environment; they cannot connect directly to sensors to interpret a user's physical environment, limiting their ability to respond to contextual queries, such as identifying the manufacturer of a toaster pointed out by a user.
Recent advances in AI, particularly with models like GPT-4, have enhanced capabilities for machine-based reasoning that combine visual, verbal, and gestural inputs. Projects like Google’s Project Astra are exploring these multi-modal assistants, which could revolutionize human-computer interactions. Additionally, the integration of sophisticated sensors in mobile and IoT devices, such as LIDAR technology in smartphones, opens the door for new situated agents that can leverage spatial awareness and multimodal reasoning to deliver contextually relevant information.
Developing these situated agents presents significant challenges, notably in energy consumption and computational complexity. High-power sensors like LIDAR are unsuitable for continuous use, and current AI models are too large to run efficiently on mobile devices. To address these issues, the project proposes innovations in energy-efficient sensing techniques and optimization of deep neural networks (DNNs) to minimize latency and energy use. Through innovations in triggering novel spatial sensors on-demand, processing relevant multi-modal sensor data selectively and by exploiting the cached results of recent reasoning actions, these advancements aim to enable responsive and pervasive situated agents that can interact meaningfully with users. The project also seeks to embed these innovations in a prototype conversational agent that can serve as a real-time tutor, responding intelligently to a learner’s queries and enhancing educational experiences.
In recent years, the rapid advancement of Large Language Models (LLMs) has driven significant progress in Generative Artificial Intelligence (GAI). By emulating human language capabilities, GAI has unlocked a multitude of applications, ranging from chatbots and virtual assistants to translation services and content-generation tools. However, GAI has evolved into a double-edged sword, giving rise to ethical concerns surrounding transparency, privacy protection, misinformation, bias and fairness, job displacement, and environmental impact.
In this project, the team aims to quantify and validate firms' ethical awareness of GAI and employ econometrics models and quasi-experiments to comprehend its determinants and document consequences. Through these efforts, the project makes a distinctive contribution to the promising trajectory of developing responsible and ethical GAI systems in business, ultimately fostering a sustainable society.
(This is additional funding to SMU for the existing research project.) Digital wellbeing has arisen in public, governmental and policy discourse as a key measure of a person’s wellbeing through a healthy use of technology. This project aims to identify and measure digital wellbeing for digital readiness, inclusion and safety.
A trusted digital identity is an essential component for securely and conveniently accessing services and authorizing transactions in cyberspace. With the rapid development of decentralized technologies and applications, such as distributed ledgers, Web3, and decentralized finance, there is an urging demand for decentralized digital identities (DID), also known as self-sovereign identities, which empower end users to create, own and govern their digital identities and assets in an autonomous, reliable, and privacy-preserving manner. The overarching goal of this project is to develop and implement a trusted, versatile, reliable and user centric DID framework covering a complete DID lifecycle. Specifically, the project aims to investigate novel techniques for enabling key components and features that are either missing or inadequately addressed in the existing DID proposals. This research is supported by the National Research Foundation, Singapore and Infocomm Media Development Authority under its Trust Tech Funding Initiative.
This project focuses on “human capital development” that involves, as one of its core pillars, the ability to use AI to improve learning. Through this project, the SMU Research team will work with the MIT team to build up key AI based models for supporting interactive learning by supporting natural multi-modal question answering associated with learning tasks. The SMU team will also develop initial prototypes that embed such interactive learning in online learning platforms.
This project focuses on “immersive workplaces” that involve extensive and interactive co-working by humans and AI-enhanced agents/robots. The research will investigate techniques to combine the latent cognitive state of a human worker, with explicit instructions issued by humans using a natural mix of visual, verbal and gestural cues, to build powerful new capabilities for human-robot co-working in immersive workplaces and industrial settings. The SMU Research team will develop optimized AI models, as well as prototype pervasive systems, that allow a variety of robotic agents to disambiguate, comprehend and respond to commands issued by human workers.
In this project, we aim to develop a generic framework to coordinate human and AI for tackling practical decision-making problems, e.g., supply chain delivery planning and healthcare scheduling, for improving the operational effectiveness and efficiency of activities in different scenarios. We will develop various cutting-edge machine learning methods to build reliable, generalizable, and explainable AI models to assist human decision making in various complex and large-scale contexts.
In this project, we propose to use a computational social science approach to analyse career trajectories using very large resume datasets combining social science with AI methods (i.e., Large Language Model-based methods augmented by knowledge graphs). The objective is to perform fine-grained analysis of career trajectory data determining the factors that contribute to career mobility as well as factors that prevent career mobility. Our new computational social science approach can be reused for future follow-up studies to reveal other detailed career trends and patterns. Hopefully, our new proposed work will also detect early signals on types of career trajectories and skills among specific demographic groups, as well as emerging trends that threaten employment, career progression and wellbeing of the workforce.
Recent advancements in AI make it possible to process large amounts of medical imaging data and replicate clinicians’ decisions with competitive performance. However, the adoption of AI in clinics has been challenging due to several issues, such as clinicians’ inability to understand how AI operates to trust and adopt it in practice. In this project, we aim to develop and evaluate a human-AI collaborative system and practices for improving collaboration between clinicians and AI in the context of head and neck cancer screening. This system learns representations of clinical videos to identify urgent referral cases and generates AI explanations on interactive visualizations to improve clinicians’ understanding of AI and their practices. After implementing the proposed system, we will conduct user studies to evaluate the effectiveness of the system.
The main goal of this project is to develop new technologies to test how well the perception module of an autonomous driving system functions and understand how perception errors impact other parts of the system, like decision-making. The project team aims to create innovative solutions to evaluate the performance of the perception module in autonomous driving. Throughout the project, the team will utilize software testing technologies, machine learning technologies, formal methods, and evolutionary algorithms to explain and develop their methods. The resulting technologies will contribute to improving the safety and security of autonomous vehicles from their development phase to actual use on the road.
