From Research to Real-World Impact: AI in Action with Sun Sumei
26:25 Share

Do you manage your own IT for distributed teams in Asia? And you know how painful it is. SFL helps your in-house team by taking cumbersome tasks off their hands and giving them the tools to manage IT effectively.

Get help across eight countries in Asia Pacific from on and off boarding, procuring devices to real-time IT support and device management. With our state-of-the-art platform, gain full control of all your IT infrastructure in one place. Our team of IT support pros are keen to help you grow. So check out ESEVEL.com and get a demo today. Use our referral code ASIA for three months free. Terms and conditions apply.

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 putting emphasis is actually how in the new era of AI or pervasive intelligence and pervasive connectivity, how do we bring human, AI and machine into a cohesive coexistence and also cohesive collaboration so that we are able to

We have the AI machine support human, empower human and at the same time, we look at the design of the workflows so that we are able to extract out the value at the same time, also enable the human to do more creative works.

So this is one aspect I think we have been hearing a lot, but how this is going to move forward is some efforts have started and certainly we look forward to gather more ecosystem partners to embark on this journey together. Welcome to Analyze Asia, the premier podcast dedicated to dissecting the pulse of business technology and media in Asia. I'm Bernard Leung, and the advances made in the AI are moving at an exponential rate. How should we think about the future of AI research?

With me today, Soon Soo-Mei, Executive Director for the Institute for Infocomm Research, Agency of Science, Technology and Research, Singapore, ASTAR. Soon-Mei, it was great to be a fellow panelist 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. And also welcome to the show.

Thank you, Bernard. Thank you for having me. It's very exciting to have this session with you and discuss some of the burning issues in AI. The best part of doing a podcast is that we can do a deep dive. But before that, as of all my guests on the show, I want to know your origin story. How did you start your career?

Sure. Yeah, happy to do that. In fact, from my name, you could very well guess that I came from China some time ago. And in fact, I came after completing my first degree in China with Peking University or Beida. And then I did my master's study in NTU before.

before I joined A*STAR and I started my career with the Institute for Infocomm Research or I2R. Since then, I've been staying here and of course I completed my PhD education while I worked with the National University of Singapore.

So it's so far a very exciting journey and there are so many opportunities. We have been going through so many eras of innovation, technology shift and it's exciting and also satisfying to actually to go through that together with our research teams and also with our ecosystem partners.

Can you share about your current role as the Executive Director for the Institute of Infocom Research and also the overall mission of the Institute under ASTAR?

Sure. Yeah. As you have introduced, I am currently the executive director of S2R. S2R is a digital technology institute. We focus on multidisciplinary digital technologies, including AI, of course, and also communications and cybersecurity. So we aim to develop trustworthy, secure technology.

and also efficient AI and digital solutions to support the ecosystem in Singapore. So we aim to be Singapore's innovation engine in digital technologies. Hence, it's important for me to actually to lead the team and also collaborate with our ecosystem partners to fulfill this mission. So it's a very exciting and also challenging journey indeed.

In fact, I also know that I2R has actually started originally the sort of natural language with Baidu many years back, autonomous driving with ST Engineering. And I think one of your former EDs, Giao Leng Tan, who is the CEO of IDA, which recently got sold.

to one of the big insurance firms. So iSquare has been quite, has this blend of actually mixing research also with commercial applications. But one interesting thing, Su-Mei, given your extensive experience in academia and also in managing science and technology, what are the interesting career lessons you can share with my audience?

As I have shared just now, we aim to be the innovation engine in digital technologies for Singapore's ecosystem. Hence, it is actually a very fulfilling role as well as a challenging role. So why it is fulfilling?

because we are able to transfer what we are researching, innovating to industry for deployment so as to create value for the ecosystem partners and also for Singapore. However, this also means that it is not only about theoretical research or about paper research. And we will have to look at how to transfer the research discoveries

tool engineering solution. And this engineering solution will have to be integratable to the industry's system. And also we'll have to push for a trade-off

between performance costs and also the economic benefits. Hence, it is exciting as well as challenging. And so what will be actually important for us, it's important for us to understand actually the research end as well as the industry's needs and the challenges so that we are able to bring these two together.

And open-minded, willing to listen, willing to really think and rethink will be actually very important. Hence, I think this is an experience as well as a lesson that we will always need to bear in mind when we innovate together with our partners.

I guess it's the question of trying to understand what is the pain point and trying to balance between research and actually bringing it, translating it into the actual technology that can help corporates to be able to solve their business problems. So I got you here today. Very well said, Bernadette.

Yeah, I got to hear you here today on the main subject of the day, actually to try to discuss your perspectives on AI. I think we have a very good conversation. So I think the way I'm going to start off is that given that AI is advancing at such an unprecedented rate, I myself, even teaching in the National University of Singapore, I find that I have to keep changing my lecture notes every week. And you have this unique vantage point in observing its growth and potential.

So what is the one thing you know about the AI's current trajectory or potential that very few do?

Yeah, thank you, Bernard. As you have put just now that we have been going through this journey, we mean, as we are together with the ecosystem partners since I think more than 10 years ago. At that time, we started with data analytics or big data, then we came to machine learning and also deep learning. And now it's actually the era of Gen-AI. So if you look at the trajectory, I think

it's important to look at actually how do we have the benefits and also the costs at a balance so that we are able to create value and also realize and sustain the value with the ecosystem partners and also for the industry. So with that, it is not only just about the foundational model development, but also how are we able to make the models efficient

drive the performance efficiency. And in particular with that, right, we have also been looking at how do we enable learning with less, learning with less data, learning with minimum enough computational resources. So with that, I actually am happy to share one of the recent, I will not say success, but accomplishment we have achieved

work together with our industry partner, the Singapore Airlines. I also shared this example at our panel session. So the excitement was actually coming from the aspect that we have developed quite a few models, as many as 10, that are

either deployed already or being a child to make it ready for deployment of all these 10 models, be it in the maintenance or the fleet optimization so that we can really drive

the efficiency and also optimal performance for the overall airline operations. However, one important aspect, as I shared, is not only about the AI model development for its efficacy, for its fitting into the operation models, but also more importantly, we have deployed

the learning with less design philosophy in this. So we have a research program to drive the learning with less and then while developing it, we have also applied it in the model development together with our ecosystem partner so that we have both aspects taken care of. And this is actually something I found very fulfilling and also exciting. And certainly we are going to continue working along this

approach with good trade-offs. When I was the head of AI machine learning for Amazon Web Services, I think between 2019 and 2021, I worked with Singapore Airlines a lot on

on their ML operations sites. They are probably one of the most advanced teams I've seen in the region itself. I think that was a very good example of how I2R has actually developed smaller models for deployment in the airline industry context. What are the areas of research on AI that I2R is currently focusing on and what are the potential applications? I recall during the conference, you also talked about some of the other interesting aspects that I think we should rehash over here as such.

Yeah, certainly. I'm happy to share and bring better awareness about SQL as well with your audience as well. So one of the recent programs we have started with is actually the National Large Language Model Development. And this

This national program is focusing on the South Asian language and also the Singapore national language. And for I2R in this particular program, our focus is actually on the multimodality in the large language model development. The multimodality involves actually speech

And why speech? This is because we are able to capture beyond just the text in terms of empathy, in terms of emotion, so that we are able to have more conversation.

information derived from the language model. And certainly, we will need to be also able to apply it when we go for the downstream kind of applications. And we found many applications that will be interesting and also necessary to have these aspects brought into the model.

Another exciting journey we are embarking on, as you may also have noticed from all the news, is the recent launch by Deputy Prime Minister Kang Kin-yung of this sectoral center of excellence,

of AI for manufacturing. In brief, we call it AIM-FT, AIM for manufacturing sector. And S2R is actually one of the research issues in this AIM operation or in the AIM team. We are envisioning or we

We set ourselves to work with the ecosystem to drive the efficient model development, plug it into this very challenging manufacturing sector so that we can realize the value, fulfill the value of using AI to drive the next stage of the innovation and also the value capture for Singapore's manufacturing industries.

So for the large language model, just to double-click a bit, are you working with the CLion team? Because I think that they are also one of the models that's done by AI Singapore, or possibly it's also in your own independent line of research then? Sure. Yeah, this is a very good question. You really showed that you know the ecosystem so well. Indeed, CLion is actually coming from AISG, focusing on the text large language model.

And X2R has put the focus on multimodality as I have explained just now. And so the multimodal model, we are crafting the name of Merlion. So it's actually an indication of this is actually the national language will be very important part of the model itself.

Please do stay tuned for our official launch and also more publicities and also more sharing of the capability and also our engagement with the ecosystem partners will be coming on the way.

Okay, I will look forward to hearing about that. So one thing is that I think we talk about some of the research that you're currently working on. Can you share any insights or developments so far that may not be widely known, but also the way that will transform how we are thinking about AI and its role to society? Right. Just I think during the panel session, as well as just now, I briefly touched on the efficiency aspect

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 putting emphasis is actually how in the new era of AI or pervasive intelligence and pervasive connectivity, how do we bring human, AI and machine into a cohesive coexistence and also cohesive collaboration so that we are able to have

have the AI machine to support human, empower human and at the same time, we look at the design of the workflows so that we are able to extract out the value at the same time, also enable the human to do more creative works.

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 gather more ecosystem partners to embark on this journey together.

With AI becoming a very integral part to almost a lot of the digital tools we use, I think during the conversation, we talk about Netflix recommendation engines, chatbots, running queries on ChatGPT. How do you think about the balance between the necessary and unnecessary applications of AI? Indeed, we...

Actually, this is a very exciting area. And also looking at the potential, all the complementary values that we are able to tap on AI, I will say it's really the large dimension the AI tools are able to handle beyond our individual human brain. And at the same time, we also touch that the knowledge

the knowledge that we need to tap out when we develop the AI models and at the same time also the knowledge graph we would should aim at deriving from the AI model so that we are able to evolve and iterate without unnecessary coming back

all the time to the raw data, to the raw large models. Hence, we are also able to have the sustainability energy consumption aspect taken care of when we actually tap on the potential from AI and also evolve from the AI aspects. Are there any areas where you think AI is now over-applied or used for trendiness rather than fulfilling actual needs?

All right. I would say it is at the starting point. There are a lot of excitement. Certainly, there are also a lot of trials and experiments. And through this, it is important for us as researchers to look at really the systematic view so that we are able to bring better awareness and also guide our industry, work with our industry and also our common consumer users to look at

What would be a right use or more balanced use of AI instead of, you know, since tech GPT is offer for free, let's just prompt, right? I think now people are still going through the curiosity stage. But once that is true, we will be able to tap on AI much more responsibly and also more

have a much better balanced approach of using AI. I should just help all my audience that generating 1,000 cat images is actually driving your car with a petrol for 6.4 kilometers.

Exactly. Yeah, this is what you have also put during our panel session, right? I think the awareness will be important for us to bring. And this is our role, you as AI advocacy and also for us as the research community.

Yeah, well, I'm advocating that I remember during the panel, you talk about responsible use. So I think that's very important. So in your experience, how should industries and consumers alike make the distinction between good AI use that adds value and AI applications that are merely overhyped? And how do we bring responsible AI into the mix?

I think I mentioned about design thinking, as you have also reminded me, responsible use of AI will be important. And for that, we will need to bring better awareness to the industry and also the general public. Hence, there is a very important role for us to play there.

Is there any examples of, say, responsible, impactful AI usage that you have seen that is interesting for the rest of us who's trying to think about the space? All right. Yeah. So I mentioned about, you know, the model. On one hand, we do see the value of large models, especially

And at the same time, when we go for deployment, how are we able to improve the efficiency, including compressing, quantizing the AI models, the parameters as one of the technical aspects. And also when we go for certain specialized enterprise use cases, whether we will have to tap on a large model or we should actually aim for a

tiny model to realize the goal. I think this will be the part as researcher or research teams is an important journey that we are actually going through. I believe there will be a lot of potential after, you know, after we go through part of the journey so that we are able to achieve this balance. How should we measure, say, success in AI? Is it just about making more technical breakthroughs like, you know,

OpenAI throwing out O1 or strawberry models or recent like CloudTree actually now have AI able to actually do computer functions or even like code generation? Or should it be the impact of say on humanity and sustainability be the primary yardstick of how we measure success in AI?

Right. Just a rephrasing of AI for humanity, AI for sustainability, right? And with that aspect, I will say really the right model for the right use case without unnecessary driving for large model will be another way of interpreting how do we actually drive AI for good, AI for, you know, for humanity and sustainability.

And also a lot of potentials that we have not been able to touch on, for example, typing on AI to handle the complexity in scientific discovery, right? And for that, you know, how are we able to make use of AI to

discover more sustainable materials, discover drugs, etc. We have already started some research programs in Singapore and I would say this will be another aspect that we can put AI for good use as well. So what is the one question you wish people would ask you more about AI?

Right. Yeah. So what have you done in terms of, you know, contributing AI for good? Wow. So I'm going to ask you that question now. Right, right. Yeah. Yeah. So what have you contributed? I'm quite curious.

Sure. Yeah. As I mentioned just now, it's actually more than 10 years of journey for I2R in this AI journey. And certainly we have seen a lot of exciting and also fulfilling successes working with our ecosystem partners. And at the same time, we also see new challenges coming along. And then

then it is also a very exciting role for us to play pushing the AI frontier and also put it for good use to support our society and also to support our industry. And this is actually something I will be more than really eager to share with your audience and also with the whole ecosystem. And moreover, I hope we can get better connected

and then we can partner with each other to actually to push this journey together. Yeah, so I'm sure that you have done a lot of work in the research publication sites, definitely contributed to AI. I'm just more being humorous to that. So that comes to my final question then. I think AI is now being heralded as the key to solving a lot of many of the world's pressing problems, whether it's climate change, discovery of new drugs, to even basic things like manufacturing healthcare.

In your view, what would great look like when it comes to AI addressing these challenges? What would success be for when AI does this? This is what success means in solving these challenges. If let's say we really look at the true success, I will say the needle moving, that using traditional method, we may succeed.

not be able to address this so quickly or so efficiently. And for that, I would say really, we as the researcher, we are able to make AI contributing to that aspect will be very fulfilling. And it is still, I think, a goal that we will need to work together and push it through.

So, Su-Mei, many thanks for coming on the show and sharing your thoughts and perspectives in a much more detailed manner as well. How much time we had during the panel as such. So in closing, I have two quick questions. First question, any recommendations which have inspired you recently? All right. Yeah, it's still very interesting. I have also been looking at, you know, the AI excitement. And at the same time, the challenges, right?

Hence, I will say the kind of recommendation for us as the community and also to the general public is how are we able to have a more systematic view so as for us to drive this forward together. This is an open question, but also I think it's an important question for us to address it.

and system systematic and balanced. And in some aspects, it may involve some drastic redesigning of the workflow and also our common practice.

I think this is an interesting point about the changing of workflows as well, how we actually use AI is going to change a lot. Last question, how do my audience find you? Since you have a lot of innovations coming out soon. Thank you. Right. We have iSquaredR, A-Star websites, and also we have the A-Star as well as iSquaredR's LinkedIn pages, Facebook pages. I myself also have a LinkedIn page myself. All these, we can get connected.

And certainly we hope to work together along this exciting journey. I also just wish to give one more shout out to our friends from Green Tech Festival, who's going to have their next event in Los Angeles on the 14th to 15th of November 2024. And you do check them out if you listen to this show in the US, because I know where you are.

do check it out. And of course, for this podcast, you can find us on YouTube and Spotify and across all, also our LinkedIn now where we post our transcripts there as well. Many thanks for coming on the show and I look forward to hearing the multi-modal, large language model launch. Thank you, Bernard. Thank you very much.