Brainstorm AI Singapore 2024: AI will have a $1 trillion impact in Southeast Asia, says Microsoft Asia president

  • 3 months ago
Arvind JAIN, Co-founder and CEO, Glean Ahmed MAZHARI, President, Microsoft Asia Suthen Thomas PARADATHETH, CTO, Grab Stephanie SY, Founder and CEO, Thinking Machines Moderator: Ayesha KHANNA, Co-Founder and CEO, Addo AI, Co-chair, Fortune Brainstorm AI Singapore
Transcript
00:00Welcome, everyone. It's a pleasure to have you here.
00:04Can we start with you? can you set us up?
00:08Microsoft is doing so much in asia. What's the opportunity that you
00:12See in asia? what are the success use cases
00:16That you think are really going to make a difference in the region?
00:20Thank you for having me and what a great pleasure to be with
00:24Folks who use ai every day. Perhaps some of you might have
00:28Used the grab app. Thank you for using the
00:31Microsoft ai engine here and for the work you're doing with banks.
00:35It's everywhere. It's pervasive.
00:37This is probably a technology that has had the fastest
00:41Diffusion rate in several centuries, i would say, of new
00:45Technology coming in. It became very real.
00:49The one example that i sometimes quote, and this goes back to
00:54January of last year, where we had this app which was in india
01:00And it's called jugal bandi. What it did was you could go
01:06From a vernacular language as a farmer, speak into whatsapp,
01:11Which would then translate it voice to text, go and search for
01:16Subsidies and return it back to the farmer in local language.
01:21This is within six weeks of gpt 3.5 out in the market.
01:25That's the rate of diffusion and that's the excitement.
01:28There are examples in hospitals across, in universities, in
01:36Schools, in agriculture. It is probably the most
01:40Pervasive opportunity. That's why we're excited that
01:44The economic impact that we see across asia by 2030 is expected
01:49To be about $1 trillion.
01:53I think speaking of grab, a number of us have taken grab
01:57Here, including me. What do you see as the nuances
02:03Of tailoring ai at scale across asia? you're in so many different cities.
02:08The population is youthful, there's mobile penetration, the
02:12Cultures are diverse, but there's also variability and
02:15Tech readiness.
02:19I think we can look at it from three perspectives, the
02:22Consumers or the population here, governments and
02:26Businesses, especially how businesses can interact with the first two.
02:29From the perspective of customers, southeast asia, 700
02:33Million people, very young population, very tech literate,
02:37And i think what's interesting is very mobile friendly.
02:41For many people here, the mobile device is their first
02:44Personal computer and often their only personal computer.
02:48Some places have more mobile devices than people.
02:51This is interesting because i think from an ai perspective or
02:55Data perspective, a mobile device means that you have
02:58Pervasive interactions with computers, the touch points are
03:01Really high, which then means that you can really spin the
03:04Data flywheel there. There's so many opportunities
03:08To create value for customers. As you create value, they use
03:11The product more often because it's right with them.
03:14As they use the product more, you collect more data, thus
03:17Enabling you to create more value, spinning the flywheel
03:20Faster and faster. Interestingly enough, in
03:22Southeast asia, especially, we're also seeing with
03:25Governments, there's obviously they always want to make sure
03:28That technology is used for good, but there's an openness
03:32To embrace technology, ai not being an exception here, to
03:38Include it as part of national development agendas.
03:41Beyond that, there's a willingness to partner with
03:45Private actors to deploy this. Let's go to businesses.
03:49For businesses, i think the takeaways we've seen would be
03:53Firstly, put the customer first. This sounds trite, but really
03:57It can be easy to get carried away with ai and use it as a
04:01Habit, solve every problem. Actually, you should solve
04:04Customer problems with whatever tool is relevant.
04:07Secondly, be proactive in engaging governments.
04:10With governments eager about this, work with them on
04:13Regulation and so on. Finally, take the lead in
04:17Self-governance. We have an internal ai risk
04:20Framework, which covers different aspects ranging from
04:24Preventing misinformation to ethical use, to protecting data
04:28Privacy, protecting ip, and also ensuring we uphold
04:31Cybersecurity. In summary, i would say,
04:35Obviously, huge opportunity here, but first use it where
04:39It's appropriate, where it's useful.
04:41Secondly, minimize the harms. If we can find more use cases,
04:45It will be truly transformative for the region.
04:48I think grab is one of the most successful apps that i've seen
04:52And has so many different uses across the region.
04:56I love, emma, your example about the farmer.
04:59I think that's a staggering example of people who stand in
05:03Lines for subsidies, right, and then get their productivity back.
05:07Arvind, i want to talk to you. You built an ai-powered
05:11Assistant, a workplace assistant, similar to increased
05:15Productivity at the workplace. How can companies in asia
05:21Transform their workplace in a responsible way and really
05:25Benefit from ai assistance in the workplace?
05:28Yeah, so i think ai is going to make a really, really big
05:32Impact in how businesses work in the future.
05:35I think it's clear to most business leaders that this is
05:41The time to actually adopt ai, play with it, experiment with
05:45It, even though the technology is early and there are a lot of
05:50Rough edges, a lot of unpredictability around models,
05:53But the time to build a great practice, build the right
05:57Experience, this is now. One of the things, when you
06:02Think about ai, the key, like the power of ai comes from its
06:07Ability to understand language and your data and your
06:10Information. So one of the things that we're
06:14Seeing, businesses that are seeing early success with ai is
06:18They're actually working actively, you know, figuring
06:21Out, you know, all of the internal companies, information
06:24And data and knowledge and bring it all together in one place
06:27And make it really, make it accessible to these ai models
06:31And technologies. So that's one of the things
06:34That we are encouraging people is think about your data, your
06:38Information, curate it, make it better, make it fresh and make
06:42It ready for ai tools to actually do the magic on this
06:46Data and information. And one of the things that
06:50Allows ai to work with all of your company's information and
06:56Data is this ability to sort of search all of your company's
07:00Information and use the right information to sort of like
07:03Automate the different tasks that you have in your workplace.
07:06So that's what we would encourage like every business is
07:10To sort of like be proactive, play with ai, even when, you
07:14Know, sort of you feel like it's not adding value to you right
07:17Now, but, you know, there's always some hard work that is
07:20Needed, but once you bring it, once you actually deploy it,
07:23You know, you can see tremendous efficiencies in your business.
07:26I have to say i'm totally addicted to ai assistance.
07:29I love them. I have seven of them.
07:32I use everybody's ai assistant. But one thing in the u.S., for
07:36Example, there's this whole movement now, increasingly here
07:40Too, which is all of these ai systems are really about the
07:44English-speaking world. And are they really, do they
07:48Have the cultural nuance for asia?
07:51And stephanie, i'd love to come to you about this.
07:54Talk to me. You build ai assistance for
07:57Some very large companies across the region.
07:59How do you make sure that they're culturally sensitive
08:01And nuanced?
08:03Absolutely. So the first thing i want
08:05Everybody to realize is that there's a huge cultural difference.
08:08This is a bit stereotypical, but in the u.S., businesses will
08:11Buy systems and software to solve their problems.
08:14In asia, in quite a lot of regions where i work,
08:17Businesses will hire people to solve their problems.
08:19So that means the inherent nature of knowledge work is
08:22Very, very different in american organizations versus, like, a
08:25Lot of asian organizations, where in asian organizations,
08:28You tend to have a huge pool of knowledge workers, but all who
08:32Have pretty narrow task realms compared to their american
08:35Counterparts. Why is this even important?
08:38It's important because a lot of us justified ai projects and ai
08:42Spend on the harvard bcg research study that came out
08:46Around september 2023, and a lot of us based the idea that we
08:50Would get 30 to 50 percent productivity gains from ai
08:53Assistance from that paper. We forgot, though, that those
08:57Are bcg consultants. Those are the best of the
08:59Brightest. These are kids who went to all
09:01The top universities and are at the very top tier of knowledge
09:05Workers worldwide. So when you roll out the same
09:08Kind of tooling and infrastructure, i think in this
09:11Harvard bcg study, a lot of the consultants just got access to
09:14Gpt4. It's like here, here's a blank
09:16Chat box, talk to it, figure it out.
09:18And they did. But when you take that model
09:20And bring it to southeast asia, to these types of knowledge
09:23Workers, you get a lot of taste and drop.
09:25You get a lot of people really excited on day one.
09:28Oh, my god, i have access to an ai assistant.
09:30I can do so much with it. And then they see, okay, now i
09:33Have to tell it what to do, but i don't even know what to tell
09:35It what to do. Shouldn't it tell me what to
09:37Do? and then they end up not doing
09:39Very much with it beyond, i think, basic e-mail
09:42Summarizations, crafting text, which is a disappointment,
09:45Right? because that is definitely not
09:48Unlocking the full power of these apps.
09:50What i find has been really useful, so i've been working
09:53With a lot of clients with, like, 6,000 insurance agents,
09:563,000 bankers, people who are knowledge workers, who are
09:59Frontline workers, who just don't want to, don't have a
10:02Time to figure out all of this, if you build them custom
10:06Applications, and you think about it not just as
10:08Technology as a silver bullet, but technology on one hand,
10:11Change management, and thinking about people on the other
10:14Hand, you do that together, then you start to get a lot of
10:17Adoption, then you start getting the daily practice of the
10:20Muscle of using it. Let me give you an example.
10:23One of the rollouts we're doing right now, we're designing a
10:27Literal song and dance. We built them a custom application.
10:30We tested it with a hundred of some best, some worst agents,
10:35And as part of our rollout strategy, we are creating a
10:38Song, which is going to be a very catchy song that tells
10:42Them how to talk to this app, and i think that that piece of
10:47Follow-through, following through, not just showing up
10:50With software as a magic bullet, but following through on
10:53Service design, on how to get people to love it, how to get
10:56People to use it, how to get people talking to it in their
10:59Own ways, because speaking to the point on very mobile,
11:02Friendly world, very i want to talk to it kind of world, once
11:07We design that in southeast asia, that's driven a lot of
11:11Adoption. Now, these models aren't
11:13Perfect, and i realize i'm talking a little bit too much,
11:16But i could go forever about this. When you talk to it in
11:20Thai, when you talk to it in bahasa, nuances of the culture
11:24And the status of the person versus the status of the gpt
11:28App, it's not quite caught in translation, so the next layer
11:32We're working on is making these, they work well right now
11:36As is out of the box with the model, beautiful, perfect in
11:40English. They'll work in tagalog, they'll
11:44Work in ilongo, but now we're looking at fine-tuning them
11:48Even further to capture cultural nuance, to capture just like
11:52The subtleties of translation, of feeling like a person that
11:55You're talking to, and i know once you add that next layer on
11:58Top of it, you're going to get even more, you're going to get
12:01People loving it even more, and i think wanting to use it every
12:04Single day is the number one driving factor of scalable
12:06Adoption.
12:08I think, stephanie, that's a great point.
12:10The devil is in the details. There's so much hype, and i
12:13Want to come back to you, emma. Everybody's talking about ai,
12:16But here's what i want to know. Microsoft helps a lot of
12:20Companies, but i've heard that microsoft is also customer zero.
12:24Is that true? and what has microsoft found at
12:28Scale has helped its own employees, its own company
12:33Unlock growth, increase efficiency at scale using ai
12:36Within microsoft?
12:38That's a great question, and thank you, aisha.
12:41I might perhaps start with two customer examples and then
12:45Transition just to make that connection.
12:48We're announcing results in the next couple of days, so i just
12:52Have to be a little bit careful as well.
12:55Let's start with the example of coals, which in australia is a
13:01Massive retailer. They have about 25,000 queues
13:05Across 850 stores, and with our llm production engine that
13:10They're working, we're able to do 1.6 Billion better
13:14Productions at every transaction to their customers.
13:17So the ability for their consumers to then land the
13:20Product at the price and the time has just gone up quite
13:24Significantly. Working with the hospital in
13:29Taiwan, chimi hospital, there we're able to help doctors cut
13:34The time that they use to write up a student, sorry, a patient's
13:39Report from one hour down to between 15 and 30 minutes
13:43Depending on complexity, and then a pharmacist is able to
13:47Treat twice as many patients from about 15 to 30.
13:51So these are real consumer, our own customer examples.
13:55If i go back to customer zero, which is very critical, because
13:58We have to really try it out ourselves, and you know while
14:01Gpt has been in the public domain for probably about two
14:05Years, we've been at it. I mean, github, which i'm not
14:10Sure if any of you are using, i think you're using python as
14:14Well, github has been in between public and private preview for
14:18About three years now, and it's one of our most scaled products.
14:23In our customer service, we have about 40,000 agents globally
14:27That serve our customers. I mean, we serve millions of
14:31Customers globally. By implementing service
14:35Standards using ai, we have achieved two things.
14:4112% better productivity, and 30% lesser handoff, which is a
14:47Pretty significant and painful thing for an average customer.
14:53But equally, in multi millions of dollars of saving on real
14:59Dollars, in our finance function, which is, this is
15:03Again, public information, we have more than doubled our
15:07Revenues over the last four years, and our head count across
15:11The company is flat. So if i look at my personal
15:17Usage, i actually now have a conversation with powerpoint.
15:21I mean, i'm not kidding. I open a powerpoint document,
15:25And i start saying, okay, can you drive a summary for me?
15:29I look at page 14, and i did this morning, that's why i
15:32Remember page 14. I look at this page 14, and i
15:35Don't quite understand what is being said in comparison to
15:38Page 6. So it's as if you can talk to
15:41The creator of the presentation, and therefore, the ability for
15:46Making this ubiquitous from word, powerpoint, from your maps that
15:52You use, from your consumer behavior, from pricing, across
15:56Industries is so pervasive, that's why we believe in the
16:00Opportunity of the impact for the future.
16:03I think that's really the key, is for all of us to hear
16:07Business success stories in real terms.
16:09If i come to you, you're doing ai scale.
16:12You're investing in compute in microsoft and in others, or
16:16You're using other kinds of tools. How do you measure success?
16:20How often do you get your teams to come?
16:22What are the metrics that you use? is it working?
16:25Is it not working?
16:28First, let's talk about how not to measure success.
16:31This has come out a few times. Don't count ai models in
16:34Production. Don't look at the superficial
16:36Artifacts. These can be vanity metrics.
16:38Really, i think the approach we take in terms of measuring
16:41Impact, we're a triple bottom line company.
16:44We obviously look at financial impact, but also the social
16:47Impact we create. Really, like i said earlier,
16:50Start with the customer. Ai is a potential solution.
16:53It's not the only solution. Don't fall into the hammer,
16:56Nail trap with it. In terms of internally what
16:59We're doing, we're investing in internal productivity.
17:02The point i made earlier, we're investing in our people and
17:05Making sure they become familiar with these tools.
17:08We're using this to allow people to take a step back from bau to
17:12Reset what business as usual looks like.
17:14Some of the early gains we're seeing, we've built an
17:17Automated marketing content generation tool that saved time
17:20By about 90, while also kind of having better engagement.
17:24Simultaneously, we're really investing in kind of improving
17:27Our user experience and the touch points in the application
17:30Really by making it more accessible.
17:32Here, we see the natural language of voice potential of
17:36These ai models, allowing us to democratize access of the
17:39Application to people who are visually impaired or maybe more
17:42Senior citizens. Screen readers are already there,
17:45But they're not as intuitive. We think voice is the most
17:48Intuitive interface. For us, actually, we don't
17:51Think okr has changed because of ai or anything like that.
17:54They say the same, but you can stretch a bit further.
17:56You can find new ways to add value.
17:58I love that. I think coming to using ai,
18:02And arvin, i want to talk about the fact that all of these
18:06Different initiatives and, for example, taking glean, how do
18:10You think about employees using it and their privacy?
18:14How important is it to roll it out in a responsible manner?
18:17Could you kind of talk a little bit about that as well?
18:20Yeah. The enterprise world is actually
18:23So different from the internet and the consumer world.
18:26All these ai models that you see today, they've been trained on
18:29All of the world's public knowledge.
18:31In that sense, anybody can go in and ask any question, and the
18:35Model is going to answer those questions for them using the
18:38World's public knowledge. But the world of the enterprise
18:41Is very different. Most of your knowledge
18:44Internally is private in nature. You as an employee can access
18:48Maybe 5% of all of your internal company documents, but not the
18:51Rest of the 95%. So how do you make sure that
18:55When you bring ai technologies inside your business that you're
18:59Not actually creating a problem and risk with ai, and instead
19:03Actually you're following the right governance model, making
19:07Sure that when you as an employee, you know, you want to
19:12Use ai and ask questions, get answers to those questions, you
19:17Have to make sure that your ai technology understands the
19:20Person who's asking the question and responsibly uses only
19:24Information that they are authorized to use within the
19:26Company to answer those questions. This is actually a very important aspect.
19:31That is how we built our product where we fully understand an
19:36Employee and their role and what information they have access to
19:40Internally. Part of building a great ai
19:44Solution is bringing all of that company's knowledge together
19:47In one place, understand these governance and permission models
19:51And make sure that ai is actually being utilized in a
19:55Safe and responsible way. You limit people to information
19:57That they have access to. That's one thing.
19:59These are table stakes. When you think about investing
20:03In ai within your enterprise, you have to invest in an ai
20:07Platform that can actually handle permissions and
20:11Privacy in the right manner. There are actually even more
20:16Challenges in the enterprise when it comes to how enterprise
20:20Information and data is actually stored today in your enterprise.
20:25There's a lot of sensitive information inside your company
20:28Which even though you make best efforts to protect information,
20:32You take sensitive content, you keep it in a safe place so not
20:36Everybody can actually go and look at it.
20:38There are often governance gaps. Ai is actually going to make
20:42These things, expose these holes and gaps.
20:45You have to think about whether your enterprise is really ai
20:49Ready or not. Once you deploy ai solutions,
20:53Even if they are built the right way technologically, even if
20:57They're making sure that they are obeying access controls and
21:02Privacy rules, they're going to expose governance gaps.
21:05You don't want me, for example, going and finding out what your
21:08Salary is. These are some of the things we
21:11Have to solve with ai and deploy it safely in the enterprise.
21:15I think this is a kind of cyber hacking or exposure or ransomware.
21:20This is the kind of information that these hackers take and
21:24Demand to be paid in various kinds of currencies.
21:27It's really, really important to have trusted partners.
21:31So last but not least, stephanie, i'll come to you.
21:35I love the fact that you have worked with the united nations
21:39And other organizations in southeast asia as well.
21:42We're really coming down to social impact.
21:44Can you in our last minute kind of give us an example of this
21:48Kind of public-private partnership and real impact with
21:52Ai for the local economy and social good?
21:55Absolutely. I think that when it comes to
21:58Models, there is a space for open source models, models
22:01Where you have the weights visible, models where you
22:04Understand exactly what's going on, the explainability is there.
22:07There's a real place for them in this world.
22:10They're very much needed for both academic research and for
22:14People who are not in corporations, people who want to
22:17Do civic projects to be able to use free models that are high
22:20Quality models. This is not just in the gen
22:24Ai space. We had a partnership where we
22:27Built a bunch of open source predictive machine learning
22:30Models on geospatial data sets to identify from geospatial
22:34Data sets, poverty and wealth indicators all over the
22:36Philippines. That is a public data source.
22:38Nobody needs to buy it. Now there's a benchmark data
22:41Set for all civic scientists seeking to do this kind of work
22:44And replicate this work. I've heard of people
22:47Replicating it in many different philippine contexts, thailand,
22:51Indonesia. It's wonderful.
22:53I think earlier the sea lion group, there's a group that is
22:56Trying to build an open source set of models for the southeast
23:00Asian languages. I think that's actually really powerful.
23:03The more we have these trusted benchmark data sets and they
23:07Would be both the models and on the training data sets, the
23:10More we can continue innovating in a scientific way and apples
23:13To apples, compare what we're doing to each other, see who's
23:16Doing better. They think there's scientific
23:20Innovation and corporate financial success exists in a cycle.
23:23One should feed into the other. If we don't continue feeding
23:26Into our scientists and research work, we're not going to be able
23:29To get the business gains out of it later.
23:31So continuing that i think is so important.
23:33That is so important. I think ai democratization in
23:37A responsible manner is the key to ai scale.
23:40Thank you so much to my wonderful panelists and really
23:44Excited to have you here.

Recommended