This is Simon.

Cool.

He actually used your software before.

Really?

The MoE dictionary?

We’ll do this all English talk, so feel free to…

We do have three parts of presentation with you. The first is introduce AICS, a general introduction. It’s about 10, 15 minutes. I’ll talk about our medical progress. I will let Miranda do the talking about 15, 20 minutes. Then, last 20 minutes, let’s do some brainstorming how we can help.

How do you like working in the government?

It’s fun. I’m working with the government.

A coalition with the government.

Yeah. I’m not working for the government.

Purely as a bridge.

How do you like it so far?

Very much so. It’s very interesting. There’s far more innovation within the public sector that we usually don’t see from the private sector.

Indeed.

A lot of my work is to share their risk, make sure that they don’t absorb the undue blame if they innovate and do something wrong. It’s one of my mantras – you can always blame Audrey.

That leads to, hopefully, more innovation.

All right, if we could, let’s get started. Again, three parts. The first are the first introduction about AICS. Second is the introduction about the medical world we’ve been doing. Third of all, brainstorming what kind of help we can offer.

First, AICS is we call ASUS AI 研發中心.

First slide is about myself, this slide. I finished my basic degree from computer science from NTU in ‘91. I went to University of Illinois at Urbana-Champaign for PhD. My first job is at Microsoft Windows. I worked on Windows in XP, Windows 2000 for five years.

Then I take a break to move back to Taiwan. That was my first trip to Taiwan. I took a job in Tsing Hua University as a computer science professor for about six years.

Cool.

Then, I moved back to California. I worked on Android for two years. Then, I moved back to Seattle for about 10 years. In 2018, me and my family decided to move back with the other 50 people from my church to plan a church in Taiwan.

Oh, wow.

Yes. Also, at the time I took a job in ASUS to start a new division called AICS. I will call it Artificial Intelligence Cloud Service Team.

Next one. Why AICS? It’s great. When I tried to find a job in Taiwan, when I decide to move into Taiwan, I was thinking about what kind of job I should take. One thing, for instance, just went back to Taiwan Microsoft. Then I went to school as a professor.

When I talked to Johnny and Ted, which is a vice chairman, they showed me the vision they have with Taiwan industry. First, they see the computer science talents are in high demand of global industry. Another thing, they also believe, which I think it’s no secret, AI will largely reshape industry…

Of course.

…or even national defense. If they are, even US, Russia, and China, they invest a lot of resources in AI…

Even the idea of nations is being reshaped.

I think I can witness that they give me full authority in terms of setting the service structural, hiring the best talent, our full interview process, very open, American-style office culture.

Do you also do office hours or one-on-ones with your…?

We do one-on-ones.

That’s awesome.

Also, a lot of people do work from home, very American-style office culture.

The goal is to create a product, like it’s global, scalable, highly profitable, but be able to fend off global competitors. It’s almost on a mission impossible. When I do a recruiting, I always tell people that you joined a team that is going to be working on mission impossible.

That’s just improbable, right? [laughs]

Yeah, we’re doing our best.

Next slide. I find those top scholars from Taiwan academics to join me and help me to build a team.

That covers pretty much all of those fields.

Yeah, you know all these people.

Well, know of, but yes.

Good. They are also on board with AICS vision.

Next slide. When we first started the team, I think the most important is to build our confidence. We built an AI core team, but we’re also given a mission to compete in a global benchmark. This shows our three benchmarks result.

The top one is the fixed precision face training, our face recognition. We were number one, which in Taiwan, number one at that time.

The bottom-left-right are the error rate. We use the speaker voice to compete with Google andother techs. Our error rate’s even lower than Google.

Next one. We use core AI engines to try different ideas.

We did it with computer vision to analyze the people’s behavior in a shop in an all-gender, age distribution, their attention, people-to-object, people-to-people connection. Safety. We also use computer vision to map out the behavior inside a factory. Right now, more and more factories want to pay top dollars to guarantee the safety in a factory.

Right now, we have one case set up with Bridgestone, from Taiwan to Thailand. They also treat us like a very big partner. It’s for all kinds of ideas inside the factory.

Smart medical, I’ll let Miranda talk about smart medical.

We still have engineering systems…

Oh, yeah. One thing we’re also very proud of, we use an world-class system to deliver our service. Every software deliver, need to go through all the six cycles to review the processm co-management, and PII protection.

Then we’re a 24/7 service stage, but we’re going to show you some results.

Do you like SonarCloud?

Actually, we use SonarCloud to do the co-quality testing. I learned this from Google that we actually track general productivity by the line of check-in. Every time people go check in to a master branch, we track their quality and then their quantity.

Once a month, we publish a co-champion in the team to encourage more Internet-driven culture. This is…

I’m aware of how it works. It’s like peer management. [laughs]

This is per team. This is per AICS unit and per team.

Next slide. This is for every public service, we monitor their availability and we build a very comprehensive system. If anything below our threshold, we will alert and call people, and wake up the people, to do the service management.

I learned all this from Microsoft.

Yeah, it’s a pretty good culture.

Next one.

I’m going to introduce the smart medical.

Yes.

This started from last year. Since July, we have the opportunity to work with one hospital. We are AI team, and the hospital has data.

The first problem that we encountered is that within the hospital, a lot of unstructured data. We must apply AI to transform the unstructured data to something which computer can compute.

We need to use AI to transform those data.

Also, our vision, we want to transform the data within the hospital, the medical data, into collaborative knowledge. For doctors, they can provide better treatment plans to patients with data-driven decision support.

Yeah, in a cross-disciplinary way.

We helped hospital to build a heterogeneous medical big data platform, then we can have a lot AI applications happening. The first project we’re taking on is disease classification.

When you visit doctor, you tell him what kind of uncomfortable feelings you have, and what symptoms you have, doctor would record what you said and maybe do some lab testing. We will combine all this data, like from the medical records and the blood test features, and we’ll help doctors to code what kind of disease you have.

For the hospital operation, doctors need to submit a code. It’s called International Classification of diseases.

Do you work with the raw interaction data in terms of voice and video, or are you working only from the written part.

We only work with the written part and also the blood tests. Something like biomarkers, we combine those data to make a decision.

Ah, OK, right.

We have a demo video, so we can take a look at it.

There are three key parts. This is a real-world example. The first part is that we use AI to look at the medical data. We could help the doctor to make better decision for choosing ICD-10.

When they added K70.6, does that train your model?

Oh, yeah.

When the doctor perform a manual input, they look at the part of the treatment that led to that idea in their mind. That is, of course, a training opportunity.

We take the current input to become the future training data. The more that they use it, the more human input…I always use the example like Google search. The more searches you type, the more credit…You teach Google how to do better search.

These are things that we do the first step to do natural language processing to analyze all the past data. The doctor also admit that, in the past, when they entered the ICD-10 code, they don’t do it very carefully.

With all this process, the more that they use it, the more people enter the correction, the model become more accurate in the future.

Right, so the doctor doesn’t need to tell you why did they choose that additional coding. You probably try to figure it out, like in the systems, so doctor doesn’t need to change their workflow, making your AI an assistive intelligence. That’s really good, and I like the UX. That is pretty good UX.

It’s a closed feedback loop. We would be getting better and better. Also, we use attention model to help the AI to be expandable. This is something we are already work with hospital and being applied in their daily practice.

Right now, we’re also solving one of their pain points from the hospitals. All the hospitals, they want to do AI, but the data, they are not organized or not cleaned.

If they work with us, we will try to organize a data platform for machine learning. We are not just tackle the ICD-10 coding problem. We are actually tackling a bigger problem. I will explain that later.

Go ahead.

For COVID-19, we read through the papers published for what kind of the symptoms could be related to COVID-19.

We have this simple warning system for doctors to see if they have some potential patients they need to take care of. This is something already deployed to one hospital.

Cool.

From the hospital side, they also want to apply the real-world data to solve their problem. When AI look through their system, we are doing auto-tagging for them to organize their unstructured data to something can be searchable and can be computable.

When they are doing research, they could use the hospital data to do cohort study, or also, if they want to take care of one patient for cancer treatment, they can look through the past data and to find out a more precise, more personalized treatment for that patient.

This is something we are tackling right now with the hospital.

We are actually doing, like providing Google search for the hospitals.

Right now, because you see there are medical records, you can see they are, it’s just free text. We use machine learning, use NLP to check the medical terms for the data. Also, we can…

We also value our AI quality. All our model, we will have some benchmark. We make sure we beat the state of the art for ICD-10.

Also, we’re doing risk prediction. Also, we have good performance, and even Google doing the same project. We are on par with their results.

That’s awesome. Thank you.

I’d like to have some open discussion for how AICS can help.

Yeah, I think the first thing that we can offer is to extend my edition of our AI for medical side. We have so much data locked in inside the hospital.

There is only a small amount of data that’s submit to our system. There’s a lot submitted to NHI…

It’s for insurance purposes.

Yeah, it’s for insurance purposes, right. There’s a lot of information that are hidden in the inside structured, the unstructured data in the hospital. When we talk to the hospital, their doctor, they also want to provide, be able to search their records in a much more efficient way and precise way.

That requires very fundamental, it’s to convert all their unstructured data to structured data. That’s why we’re doing this.

Exactly.

Also, there are clinics that like to do application. We think this, we look at projects to all the hospital. In the end, it will be a very powerful tool for the country to be able to do the first line defense when something like COVID-19 happened.

The government can issue a query to say, “Find me all the people who have travel history, who have any symptom in the past three to six months.” That query is impossible in today’s environment.

Only travel history. Basically, only the whereabouts, of course, because the telecoms has that data, but nothing else.

When we look at all the description from the doctor, has all the value information inside those descriptions.

Right now, is that we actually built our subset of the database, but clean, structural, indexable, and the database eventually will become…will provide search interface for the hospital to use it. If we can duplicate this to all the hospitals, within silos still these owned by the hospital, they can use it to do research or any search they want.

That’s one agenda, how to duplicate the success, apparently, with the hospital ally that you are currently having to more hospitals, and the final outlook being a global query.

Right. From the perspective, if this is passed, although this will take years to build, if this is passed and this will be, how valuable it is to a government.

It is valuable for the public health, for sure, but I’m not sure that it’s very valuable for the government per se. The government’s main role is to ensure that the safety and security, and also because we’re a democracy, legitimacy of the public administration.

There are jurisdictions nearby that are very interested in that kind of power, but in Taiwan, usually when people even thinking about this kind of thing, immediately lowers their trust, one example being facial recognition. It is not even an open secret.

It’s open information that there are certain jurisdictions that vastly prefer facial recognition, because it’s a general-purpose technology, but in Taiwan, when you see, for example, the Taiwan railroad using facial recognition to do something that a simple motion sensor could have done, then there is a very large public backlash against a misuse of a general-purpose technology.

The norm in Taiwan is that you deploy facial recognition only when you opt into eGate, and nothing else. I think there is a very different cultural norm going on. Anything that breaks this norm may be acceptable like during the Corona epidemic, because we understand it’s a time of mobilization.

It is always with the idea that it should create a tangible public benefit, and the governance needs to be either owned or closely monitored by the social sector. In Taiwan currently, we don’t have an independent data protection authority.

The usual legitimacy building device that is conveniently specified in the GDPR is not here in Taiwan. Taiwan, the closest we have is the Ethic Board within the MOHWM within each agency within the MOHW, but the legal interpretation is done by the National Council.

None of these are independent, multi-stakeholder legitimacy devices. When they make a decision, even if they publish the entire transcript as the NHI insurance committee actually do, people will say that they are too professional.

There’s no way to understand in the lay person’s terms. There is no meaningful participation from the Cultural Yuan, or the Legislative Yuan, or things like that. It’s not truly multi-stakeholder, is what I’m trying to say.

To build the kind of query system that you used, there’s two problems that need to be solved. First, that we need to ensure that people can form a useful data coalition, or data mutuals, data unions. There’s many words for that idea, where people opt in to join a pool of data that creates value for everybody in the pool, but also provably to the larger society. That’s a technical and also social innovation issue to be solved.

Also, on the regulatory side, we need to have an enabling regulatory framework that enables this kind of independent data protection authority that can look at each query and say whether this is ethical by Taiwanese standards or not.

Without these two infrastructure, the government is understandably less interested in the kind of big query that you just described.

I see.

The main comparison with, say, Microsoft, is that every Microsoft customer probably opt in to the Microsoft machine learning terms, that if I don’t want machine learning in my Office, where I just tell PowerPoint to stop predicting how to layout, everything is by opting in.

In the National Health Insurance, there is no opting. There is no opting out, even. [laughs] You can’t choose not to use the National Health Insurance. Because of that legitimacy requirement, it’s far higher than that in the private sector.

How does government interest in organize this information to become the next industry for Taiwan?

There’s a lot of interest in assisting the medical industry, assisting pharmacies, assisting biological research.

Everybody is doing that, right?

Right. Automating away the chores, like the part that you did on trying to capture more information so that the AICS coding become more useful to the doctors.

These are seen as, of course, very much aligning with the government goals, which is making sure that people who are very smart and get trained to be doctors don’t need to waste their time doing repetitive job of any kind. That is in close alignment.

The other kind of the work that you do, which is about aggregating data and ensuring that aggregated data better reflect the actual trend of like…digital twin is the usual word, that we don’t see a very strong interest from the MOHW, precisely because the MOHW is mainly interested about keeping people health and safe, and less about enabling an industrial, whatever development.

Their legitimacy, maybe inquiry is the best word. Their inquiry into how to build a legitimacy structure that I just described, you need to show that it is actually possible to build something like that. Then they will say, “OK, then maybe we work with the private sector, which are for profit but with purpose.”

If you don’t have that legitimacy structure that is for a purpose but with profit, then there is no other side of the bridge to connect to the private sector.

I see. The next few slides about the brainstorming part, we offer all health on a pure IT perspective, not using any medical stuff. I defined three scenarios, which is first on the control, second is a getting worse scenario. I hope Taiwan will not get into that situation. The third slide is about totally out of control.

People are talking about an advanced deployment, 超前部署.

I know. Really advanced deployment.

Doing it under control, those are the few items I can think of. First, predict the group that has a high chance to get infected using all the medical records they have. The second is that once you get infected, confirmed, how do you provide the best medical treatment analysis? Given with enough data, we should be able to do the data analysis and big data mine.

The third one is, again, related to structured data. When the medicine has been developed, how can we select the patient to test it? Right now, our social structure would be able to provide a better search than those human search to find the ideal patients.

Number four is about cities, how do we make sure, especially this is just analysis, one meter social distance?

That’s right. We’re keeping it right now.

It’s one meter?

We’re keeping 1.5 meters of distance here.

How do you make sure people follow this rule if they don’t follow rules. Then number five is how do we monitor the situation from different countries and open the border when appropriate?

Number six is the hospital will need to allocate resource to deal with the confirmed cases or even some quarantine cases. How do we monitor the hospital health from the different signal that we have? Remember [inaudible 40:59] is a supporting tool for quality monitoring.

How do we make sure that people stay home? Right now, I know we use a lot of human resource to call the people at home to say, “Are you stay home, quarantined?”

Yeah, they’re automating that with a LINE bot, 疫止神通.

OK. Those are something we can think of when things are still under control. Next one is that it is getting worse. Now, I think we might need to apply individual health indicator. When you walk into a building, we check your health indicator to make sure you are still…This is a red, green, yellow alert system.

Yeah, the health code. A nearby jurisdiction demonstrated that.

Yes. Number two is that when, since getting worse, I think our resource will be very limited. How do in efficiently deploy and supply the resource to different households? We need a very efficient supply logic, dispense system. We can also provide IT power to help build a system.

Number three is that we also need a very good free management tool to distribute the resource through our government-owned vehicles. We can also provide those system to monitor the driver, make sure they deliver on-time.

Number four is that, again, it’s that when people get confirmed infected, how do we know which case will develop into serious, major case? What kind of case will remain a minor case? We can also use big data to help to predict.

Number five is that, when more and more people get quarantined at home – and minor case are also quarantined at home – how do you provide a remote medical service for minor cases? We may go through CV or ASR to be able to provide a good prescription for people.

Number six is that hospital, we need better resource management, including equipment, personnel to be able to do low beds at all the hospitals. I don’t think right now we have a good system to do low beds right now, but I think we do need to think about it.

There are teams in the Presidential Hackathon working on that. They are choosing major trauma, because they work with the Kaohsiung gas explosion cases. Many of these apply also in a Kaohsiung gas explosion, where a lot of time is wasted on repeatedly calling people.

People, the ambulance did not have any prediction or prognostic analysis to redirect to the ambulance to the right place, and so a lot of time is lost. There is a team working on that. I do agree that this is going to be critical in the getting-worse scenario. Go ahead.

Last slide is for out of control, that we actually have 1,000 confirmed cases.

Apocalypse.

I think we need a strict, guaranteed law enforcement tool. Not just monitor, but every time there is a violation, we should be able to integrate and bring the law enforcement to enforce the people follow rule.

Sending in drones or something? [laughs]

Something like that, yeah.

Then we might also build a bunch of cabin hospitals, but we don’t have enough manpower to monitor and run the cabin hospital. How do we do CV and AI technology to do surveillance, vital sign collection, monitoring, and alerting system?

I think we offer our IT resource and anything that we can help, let us know.

Cool. Anything from you? OK.

Yeah, because we are working with the medical problem for some time, and we have very good AI engine using our team. We have AI talent. Right now, it’s about there is something that we can offer.

We can offer to help here, but it’s later something that alignment that can work with the governor or with you, your team. We can see how, what kind of opportunity we can have.

A couple of slides ago, I was…This passed through my mind, that we in Taiwan is really privileged, because we can afford to say social distancing monitoring system, while other jurisdictions that did not have a SARS experience are trying to solve the problem that people not even understanding why social distancing is required.

Or that we’re talking about supporting tools for quarantine monitoring, while other jurisdictions are even debating the constitutionality of quarantine monitoring. It shows how privileged we are. [laughs]

Here’s a few issues that we are receiving from our partners. Mostly from the states, but also from other friendly jurisdictions. You can see that they are tangibly different from the problems that we in Taiwan are looking at.

Like people start believing in conspiracy theories. What should we do? We have set up central epidemic to monitor and translate epidemiological knowledge to useful recommendations that people can rely on.

Also, there is also issues about how to communicate what is really, requires to be done to the elderly people. Again, in Taiwan, any elderly probably knows how to use Line, and so we don’t have that problem. Or they know somebody who have access to Line, but they have that problem.

You see things like this, which would never have been a brainstorm topic in Taiwan. Of course, yeah, and Bill Gates, of course, agrees completely with you. [laughs] That’s probably the one…

Microsoft.

Yeah, it’s a Microsoft mindset.

There is also quite a few things about mental health. People are looking at how the home quarantining affects mental health, or even people in abusive relationships, how it makes the problem worse, and things like that.

Again, we see people who have not experienced SARS. They first need a chat bot or a video game, somehow to visualize how bad it could get for them. The problems that you raised sounds really like first world problems [laughs] when the rest of the world are caught unprepared.

I’m not writing any of your brainstorm ideas off, but mostly, we’re seeing nowadays solutions that could solve for the out of control scenarios, but not in Taiwan. An out of control scenario that doesn’t presume anybody in the civil society understand how bad SARS was.

People who don’t have a mental preparation, who don’t have a mobilization capability. The solution set looks very different. For example, one of the tools that one of my friends is working on is this one. I hope this loads.

Like this is super, super, super simple user experience. It assumes that people knows nothing. Then it gradually guides them through a ask the science anything chat bot that eventually gets people into the public health literacy that we take for granted.

They’re still doing 101, but they use all sorts of chat bots and whatever, and in various languages, to do that. It’s multilingual. It actually also use AI, but predicting the questions and things like that.

All the questions, as you can see, are probably not questions for a Taiwanese audience. They are also deploying cutting-edge technologies, but from our point of view, very, very basic needs and things like that.

Then chat bot allows you to, of course, follow up on these questions, where they try to pair you with a scientist and answer in the language that you can understand, and make sure that the answers are accessible to a broad audience, and things like that.

I’m trying to balance two needs here. One is the scenario that Taiwan gets out of control. Even if we spiral out of control, our social configuration will be very unique, unlike any other jurisdiction that’s getting out of control.

Maybe only slightly comparable to South Korea, but nowhere else. That’s one thing. The other thing is that, around the world, all our allies and like-minded countries are getting out of control in very different ways. One unifying thing is that people cannot get access to the right information, and it is not just for the government but just for everyday individuals.

That’s the main call of help we’re getting, but Taiwan doesn’t have that problem. If you solve for that, you don’t actually solve anything in Taiwan. I’m just sharing the dashboard of my daily work.

I see. Since Taiwan is far more advanced than the rest of the whole world…

Yeah, even in terms of mental preparation because we have a way of mobilizing against chaos or in Mandarin [Mandarin] . It’s something that means a lot to us. It means nothing to other countries.

It’s really good to solve a Taiwan problem, but it then also means that it’s unique to Taiwan, the solution. Or we can solve it for the world, but that means that it’s less applicable when Taiwan goes out of control.

There’s going to be plenty of chances of this brainstorming. We are soon – soon as maybe one or two weeks from now – posting all of these calls to help that we receive online and will ask people to rate whether you think it’s an interesting problem to solve.

Then after voting for a bit, you will be prompted to what do you think is an interesting problem to solve, and it will be proceeding bilingually. This is called additional dialog. We’ve done four of these with AIT before and they always seemed very interesting conversations.

This is powered by, again, an AI system called Populous, which asks simple questions based on shared data. This is an actual dialogue we had when we an UberX case. That was the first case in 2015 where people agreed or disagreed whether this was a good idea or not to work with UberX.

Every time we run this kind of crowd agenda setting, we always see this after two or three weeks, in that there are a few very controversial ideas which we do nothing, a few quietly accepted ideas, but a large at the left where it’s transcultural acceptance.

Whichever society norm you are on, you probably agree that in addition to science, technology, engineering and math, arts are also creative and also very important. Who could have disagreed with that?

This is also something that’s not yet done. It is what we call low-hanging fruit, that we can just mobilize to solve, while knowing that it is a universal problem that we’re solving, and not a Taiwan-specific problem that we’re solving.

There’s going to be this brainstorming process probably a couple weeks from now that I cordially invite you all to join once we get here so we can know for sure which are the thoughts that we should be focusing on and what are the scenarios.

We make sure that all of these asks for help come from real frontline workers, no matter which area they’re working. So that we can pair them with you, then, if you decide to solve them on the second phase. That’s the process that we’re going through, maybe, two weeks from now.

How to prevent the resource working on the same problems?

Basically, we require everybody working on it to adopt open source. You need to create, maybe, a public e-lab or a GitHub repository. There will be online kind of jamming sessions. It’s required to be open innovation. People can just carry parallel swarm-like experiments.

Exactly like what you’re doing, internally, and making sure that all the data that we publish are synthetic data. If you find a solution, and so on, because it’s open source, we’re going to run your code on our premise.

That’s actually the only way, because across jurisdictions, the copyright law, the trademarks, patents, and things like that are all very different. It’s very entangled.

What we are trying to do, because we want to engage international civic hackers, is just to agree on a shared, like Creative Commons, open source licenses. If everybody agrees on that, then we don’t have to worry about patent or copyright laws when doing the co-creation.

That makes sense.

Of course, if it’s proven to work, then you can take that as your portfolio, and then go and add more value after the events, but during the event, we’re going to require open source.

That makes sense. Perfect. All right.

That’s all I have. I’ll keep you posted when that comes.

That’s great. Anything that we can help…

Just keep brainstorming, and then maybe think outside Taiwan a little bit.

That’s a good point.

Chances are that we will not force the worst scenario, where the president will need to be protected and so on.

We’re doing actually pretty well.

This is a global pandemic, and it may well mutate. Once it mutates, then all the places where they did not mobilize will start from ground zero, only with much lower medical resources, and that will be even more apocalyptic.

In Taiwan, the social configuration is such that even if the virus mutates tomorrow, we’re doing exactly the same. We’re not particularly affected by mutation. Everywhere else, places where there’s a herd immunity, things like that, they will be affected by mutations very severely and are probably now just praying that it doesn’t happen.

I see. All right, good conversation. Thank you very much.

Thank you.