From Research to Real-World Impact: AI in Action with Sun Sumei
Fresh out of the studio, Dr. Sun Sumei, Executive Director of the Institute for Infocomm Research (I2R) at Singapore’s ASTAR shares her perspectives on AI’s fast-paced evolution and its broader impact on the future. With over a decade dedicated to AI research, Dr. Sumei reflects on I2R’s journey from big data analytics to the era of Generative AI, emphasizing large language models designed for Southeast Asia. Advocating for responsible AI, Dr. Sumei prioritizes societal benefit and sustainability over sheer technical ambition, urging a balanced, systematic approach to distinguish valuable applications from mere hype. Last but not least she shares her vision of AI as a force to enhance human creativity and address real-world challenges through sustainable innovation.
"The efficiency on top of the efficacy because it is critical that we will achieve this balance of cost and performance. And in addition to that, I think One more aspect we are emphasizing is actually how in the new era of AI, all-pervasive intelligence and pervasive connectivity, how do we bring human AI and machine into a cohesive coexistence and also cohesive collaboration so that we can have the AI machine to support human, empower, human, and at the same time. We look at actually the design of the workflows so that we are able to extract the value at the same time, also elevate the human to do more creative work. So this is one aspect. I think we have been hearing a lot, but how this is going to move forward is some efforts we have started and certainly, we look forward to gathering more ecosystem partners to embark on this journey together." - Sun Sumei
Profile
Sun Sumei, Executive Director, Institute of Infocomm Research (I2R), Agency of Science, Technology & Research (ASTAR) [LinkedIn]
Here is the edited transcript of our conversation:
Bernard Leong: Welcome to Analyse Asia, the premier podcast dedicated to dissecting the pulse of business technology and media in Asia I'm Bernard Leong and the advances made in the AI field are moving at an exponential rate. How should we think about the future of AI research? With me today, Sun Su Mei, Executive Director for the Institute for Infocomm Research (I2R), Agency of Science Technology, and Research, Singapore (ASTAR). Sumei, it was great to be a fellow panellist with you at the Green Tech Festival Singapore 2024 on the topic AI Game Changer or Costing the Earth where we had a very interesting conversation on the topic. |Sumei, Welcome to the show.
Sun Sumei: Thank you for having me. It's very exciting to have this session with you and discuss some of the burning issues in AI.
Bernard Leong: The best part of doing a podcast is that we can do a deep dive. But before that, as for all my guests on the show, I want to know your origin story. How did you start your career?
Sun Sumei: Absolutely, I’d be happy to share. As you may guess from my name, I originally come from China. After completing my undergraduate degree at Peking University, or Beida, I moved to Singapore to pursue my master's studies at Nanyang University [or NTU]. Soon after, I joined ASTAR and began my career with the Institute for Infocomm Research, also known as I2R. I’ve stayed in Singapore ever since, completing my PhD while working at the National University of Singapore. It’s been an incredibly exciting journey filled with opportunities. I’ve witnessed and participated in multiple waves of innovation and technological shifts, and it has been both thrilling and rewarding to navigate these changes alongside our research teams and ecosystem partners.
Bernard Leong: Can you tell us about your role as the Executive Director of the Institute for Infocomm Research and the Institute's mission under ASTAR?
Sun Sumei: Certainly. As you mentioned, I am currently the Executive Director of I2R, a leading digital technology institute. Our work spans various areas of digital technology, including artificial intelligence, communications, and cybersecurity. Our mission is to develop trustworthy, secure, and efficient AI and digital solutions to support Singapore’s ecosystem.
We aspire to be the driving force of digital innovation in Singapore. For me, this means leading our team and collaborating closely with ecosystem partners to realize our mission. It’s an exciting journey, and certainly a challenging one, but incredibly fulfilling.
Bernard Leong: I know that I2R has been at the forefront of applied research, collaborating with companies like Baidu on natural language processing years ago, and working with SD Engineering on autonomous driving. And if I’m not mistaken, one of your former executive directors, Geok Leng Tan, was the CEO of AIDA Technologies, which was recently acquired by a major insurance company. I2R has skillfully blended research with practical, commercial applications.
Given your extensive background in academia and in leading science and technology initiatives, what valuable career lessons would you like to share with our audience?
Sun Sumei: As I mentioned earlier, our mission is to be the innovation engine for digital technology within Singapore’s ecosystem. This role is incredibly fulfilling but also quite challenging. It’s rewarding because we can take our research and innovations and transfer them into industry applications that create real value for both our ecosystem partners and for Singapore as a whole.
However, this also means that our work goes beyond theoretical or academic research. We need to translate research discoveries into engineering solutions that can seamlessly integrate with industry systems while balancing performance, cost, and economic impact. This makes our work both exciting and challenging.
It’s essential for us to deeply understand both the research end and the needs of the industry, along with the challenges they face. Staying open-minded, actively listening, and continuously rethinking and adapting our approaches are all crucial in bridging these two worlds. This is an experience and lesson that I believe will always be relevant as we innovate alongside our partners.
Bernard Leong: It comes down to understanding the pain points and balancing research with practical implementation—translating insights into technology that genuinely helps businesses solve their challenges. So, I invited you here today to delve into your perspectives on AI, especially as it's advancing at an unprecedented rate. Personally, as a lecturer at the National University of Singapore, I find myself constantly updating my lecture notes just to keep up! With your unique vantage point in observing AI’s growth and potential, what’s one aspect of AI’s current trajectory or potential that few others might be aware of?
Sun Sumei: Thank you, Bernard. As you pointed out, we at I2R, along with our ecosystem partners, have been on this journey for over a decade. We began with big data analytics, transitioned into machine learning, and later into deep learning. Now, we’re in the era of Generative AI. Looking at this trajectory, it's crucial to strike a balance between benefits and costs so that we create, realize, and sustain value for both our ecosystem partners and the industry.
Our focus isn't solely on foundational model development but also on making these models efficient and high-performing. For example, we’re exploring ways to enable "learning with less"—less data and minimal computational resources. I’m pleased to share one recent accomplishment in this area. Working with Singapore Airlines, we developed a suite of AI models—around ten, to be exact—that either have been deployed or are undergoing trials for operational readiness. These models enhance maintenance and fleet optimization, driving efficiency and peak performance across their operations.
But our focus wasn’t only on model efficacy; a key aspect was integrating our “learning with less” philosophy. We designed a research program dedicated to minimizing data and resource needs, which we applied directly in this collaboration. This approach has been incredibly fulfilling and rewarding, and we’re committed to continuing along this path, finding balanced solutions with thoughtful trade-offs.
Bernard Leong: During my time leading AI and machine learning at Amazon Web Services between 2019 and 2021, I worked closely with Singapore Airlines on their ML operations. They are among the most advanced teams I’ve encountered in the region. This collaboration with I2R on smaller, scalable models is an excellent example of practical AI deployment in the airline industry.
Could you tell us about the current focus areas in AI research at I2R and the potential applications? I remember you discussed some intriguing aspects at the recent conference, and it would be great to revisit those here.
Sun Sumei: Absolutely. I’m glad to provide more insight into I2R’s work. One of our latest initiatives is the National Large Language Model Development program, which focuses on creating language models for South Asian languages as well as Singapore’s national languages. At I2R, our specific focus within this program is on multimodality—particularly integrating speech into the large language model. Speech allows us to capture not just text but also elements of empathy and emotion, which enriches the information derived from the model.
Our goal is to ensure these nuanced capabilities can be applied across a wide range of downstream applications, where this enhanced understanding can be both valuable and necessary.
Another exciting initiative we’re part of is the newly launched Sectoral Center of Excellence for AI in Manufacturing, introduced by Deputy Prime Minister Gan Kim Yong. This initiative, which we call AIM (AI for Manufacturing), sees I2R as a key research contributor. Our vision is to collaborate with ecosystem partners to develop highly efficient AI models tailored to the complex challenges within the manufacturing sector. By embedding AI effectively, we aim to unlock the next wave of innovation and drive substantial value for Singapore’s manufacturing industries.
Bernard: Regarding the large language model, I’d like to delve a bit deeper. Are you collaborating with the SeedLion team from AI Singapore? I believe they’ve developed one of the language models as well, or is your work more of an independent research line?
Sun Sumei: Great question, and it really shows your understanding of the ecosystem. Indeed, SEALion is AI Singapore’s model focused on text-based large language applications. At I2R, we’re focusing on multimodality, as I mentioned earlier. Our model, which we’re calling Merlion, integrates not just text but also speech, with a particular emphasis on Singapore’s national languages. This national aspect is a core part of the model’s identity.
Please stay tuned for our official launch. We’ll be sharing more details on MerLion’s capabilities, as well as our engagement with ecosystem partners, in the coming months.
Bernard Leong: I look forward to hearing more about that. Shifting to a broader perspective, could you share any lesser-known insights or developments that might reshape how we think about AI and its societal role?
Sun Sumei: Certainly. In our panel session, I touched briefly on the importance of balancing efficiency and efficacy. It’s critical to strike this balance to optimize both cost and performance. Beyond this, we’re increasingly focusing on fostering a seamless coexistence and collaboration between humans, AI, and machines in this era of pervasive intelligence and connectivity.
Our aim is to design workflows that not only maximize value but also elevate human roles, allowing people to focus on more creative work. This human-centred approach is foundational to our efforts, and we’re actively building partnerships to push this vision forward. We’re excited to bring more ecosystem partners into this journey and explore the possibilities together.
Bernard Leong: As AI becomes a core part of many digital tools we use—whether it’s Netflix recommendation engines, chatbots, or running queries on tools like ChatGPT—how do you think about balancing essential applications of AI with those that may be less necessary?
Sun Sumei: It’s certainly an exciting area with immense potential, especially considering the broad range of tasks AI can handle that go beyond individual human capabilities. We also need to look at the knowledge we tap into when building AI models. Developing a robust knowledge graph from AI models allows us to refine and evolve the model without constantly revisiting raw data or large foundational models. This approach not only increases efficiency but also helps us address sustainability by reducing energy consumption as AI models evolve.
Bernard Leong: Are there areas where you feel AI is being over-applied, perhaps for trendiness rather than actual need?
Sun Sumei: Yes, absolutely. With all the initial excitement around AI, there’s naturally a lot of experimentation. For us as researchers, it's essential to take a systematic view, bringing awareness to guide both research and application. Working with industry and consumers, we aim to identify what constitutes responsible and balanced AI use, rather than simply using tools like ChatGPT just because they’re accessible and trendy. Right now, people are in the curiosity phase, but as that settles, we’ll see a shift toward more thoughtful and responsible applications of AI.
Bernard Leong: I should mention to our audience that generating 1,000 cat images with AI has an energy impact—equivalent to driving a petrol car off-road for 6.4 kilometres!
Sun Sumei Leong: Exactly. As you mentioned during our panel, creating awareness is essential. AI advocates like yourself, along with the research community, have a crucial role in raising public understanding of AI's impact.
Bernard Leong: Yes, and as we discussed, responsible AI use is a critical topic. In your experience, how can industries and consumers distinguish between AI applications that genuinely add value and those that are simply overhyped? How do we ensure responsible AI becomes the norm?
Sun Sumei: It’s a great point. Responsible AI requires thoughtful design and awareness. We need to help both industry and the general public understand the value and limitations of AI. By advocating for a balanced approach, we can guide responsible AI development and usage.
Bernard Leong: Are there examples of impactful, responsible AI usage that stand out to you—examples that can inspire others who are navigating this space?
Sun Sumei: Certainly. For instance, in our work with AI models, we see the value of large models, but we also emphasize efficiency. By techniques such as compressing and quantizing models, we make them leaner, which is particularly valuable in specialized enterprise applications. Sometimes, a tiny model can achieve the goal just as effectively as a large one. As researchers, we’re committed to finding this balance and maximizing AI’s potential responsibly. This journey will continue to unfold, offering more insights as we refine and adapt our approaches.
Bernard Leong: How should we measure success in AI? Is it about achieving more technical breakthroughs—like OpenAI releasing advanced models or tools capable of code generation? Or should we instead focus on the impact AI has on humanity and sustainability as the primary measure of success?
Sun Sumei: That’s a crucial question. I believe the idea of "AI for good" essentially translates to AI for humanity and sustainability. Success in AI doesn’t necessarily mean pushing for ever-larger models. Rather, it’s about finding the right model for the right use case, minimizing unnecessary complexity, and aiming for applications that contribute positively to society and the environment.
There are also tremendous untapped opportunities in areas like scientific discovery. For example, AI can assist in uncovering sustainable materials, discovering new drugs, and advancing other critical research areas. In Singapore, we’ve already initiated research programs focused on these applications. These efforts represent another way we can harness AI for good, aligning with the broader goals of benefiting humanity and supporting sustainable progress.
Bernard Leong: What’s one question you wish people would ask you more about AI?
Sun Sumei: Ah, yes! I wish more people would ask, “What have you done to contribute to AI for good?”
Bernard Leong: Great question! Well, let me ask it now: what have you contributed?
Sun Sumei: Certainly! As I mentioned earlier, I2R has been on a 10-year journey with AI, filled with exciting and fulfilling successes alongside our ecosystem partners. Over the years, we’ve achieved meaningful advancements, but we’re also constantly facing new challenges. It’s an incredible role, being at the frontier of AI and finding ways to use it for the benefit of society and industry. This is something I’d love to share further with your audience and the wider ecosystem.
I also hope for greater connections and partnerships as we continue along this journey, building on each other’s efforts to push AI forward in ways that truly make a difference.
Bernard Leong: I know you’ve contributed greatly to AI through research and publications—just a bit of humour here! This brings me to my final question. AI is increasingly seen as key to solving many of the world’s pressing problems, from climate change to drug discovery, manufacturing, and healthcare.
Bernard Leong: In your view, what would “great” look like when it comes to AI addressing these global challenges? What does success really mean in this context?
Sun Sumei: True success would mean moving the needle in ways that traditional methods haven’t been able to achieve—tackling these issues more quickly and efficiently. For us as researchers, contributing to this kind of meaningful progress would be incredibly fulfilling. However achieving it is a long-term goal that requires collaboration and sustained effort.
Bernard Leong: Sumei, thank you for coming on the show and sharing your insights so comprehensively—much more than the time allowed us during the panel! To close, I have two quick questions. First, are there any recommendations or perspectives that have inspired you recently?
Sun Sumei: Thank you, Bernard. This is an interesting question. I’ve been reflecting on both the excitement around AI and its accompanying challenges. I believe a key recommendation for our community—and the public—is to adopt a more systematic approach to AI, one that moves us forward collectively.
It’s an open question, but an important one. Striking a balance and applying a systematic approach might even require us to rethink workflows and common practices in significant ways.
Bernard Leong: Before we wrap up, I’d like to give a shout-out to our friends at GreenTech Festival. Their next event is set for November 14-15, 2024, in Los Angeles. If you’re tuning in from the U.S., be sure to check it out! As for this podcast, you can find us on YouTube, Spotify, and now on LinkedIn, where we also post our transcripts.
Thank you so much for joining us today, Sumei. I’m excited to see the launch of the multimodal large language model!
Sun Sumei: Thank you, Bernard. It’s been a pleasure.
Podcast Information: Bernard Leong (@bernardleong, Linkedin) hosts and produces the show. Proper credits for the intro and end music: "Energetic Sports Drive" and the episode is mixed & edited in both video and audio format by G. Thomas Craig (@gthomascraig, LinkedIn). Here are the links to watch or listen to our podcast.
