CNBC Interview with Dr Min Wanli, Chief Machine Intelligence Scientist, Alibaba Cloud

Below is the transcript of a CNBC interview with Dr Min Wanli, Chief Machine Intelligence Scientist, Alibaba Cloud and CNBC's Arjun Kharpal. The interview took place at CNBC's inaugural tech conference, East Tech West, in Nansha, Guangzhou.

AK: Excellent. Dr Min, I want to just very quickly kick off with something Alibaba's CEO told me very recently, Daniel Zhang, he said cloud business will be the main business at Alibaba in the future. What's going to drive that, at Alibaba? What are some of the things that is helping the cloud computing division at Alibaba grow so fast?

MW: Okay, let me put it this way. So, Alibaba has a unique economy, that has its differentiation from the traditional cloud, because, at the end of the day, everybody, when you open (inaudible) on your mobile app, on the first landing page, it seems as if it's customized to yourself, and behind this calculation, it's powered by cloud, as of today. It's bringing, like, customization to every consumer, powered by this cloud. And imagine if we somehow unleashed this power of the cloud to the sectors beyond ecommerce, say, for example, transportation, traffic management, like the City Brain, or manufacturing, or agriculture, any sector with a data streaming intensive. Data streaming intensive, okay, this is a key point. Okay, now, if you look at that criteria, then categorically, you'll see that even the traditional agriculture, the manufacturing, and even healthcare and city management, they are filled with real-time data. Okay. All these scenarios will be perfect testbeds, and also enjoy the benefit of the power of the cloud.

AK: That's really interesting, because when people think about Alibaba, of course, they think ecommerce company-,

MW: Yes. Yes.

AK: Which, of course, is still very much the core business-,

MW: Yes.

AK: But how much learnings have you taken, from-, all the data that you've been able to process, from consumers, using Alibaba's ecommerce platforms, how has that helped you to develop the cloud business at Alibaba?

MW: Yes, very-, very important question. Let me put it this way. Now, imagine if everybody-, you just did the 11/11, November 11th shopping, okay? And more than half a billion – more than half a billion consumers are doing their shopping, on the very same day, and the backend technology, provided by the cloud, and also the real-time big data fusion, this technology, the capability-, it's not the data itself, it's the data capability can readily transfer to another sector, like the city, because in the city, you have sensors everywhere. You've got all the cameras, you've got the mobile phones, mobile carriers, and then you've got the loop detectors, as well, okay? So, it looks as if it's a perfect analogy to the consumers' behavior on the ecommerce platform. So, once you see this analogy, the commonality, and then you just transfer the technology from ecommerce to the smarter city, or to the agriculture, or to the manufacturing. So, pretty much, we transfer the data capability, we are not transferring the data, okay, this is a categorical difference.

AK: And Dr Min, of course, your focus is on machine intelligence-,

MW: Yes.

AK: And AI. When we talk about-, because it's a big buzzword, it's thrown around a lot-,

MW: Yeah.

AK: Perhaps sometimes a lot of people don't really know what that actually means. So, when we talk about AI applications in the cloud, what does that mean for Alibaba?

MW: Okay, let me put it this way. First of all, without cloud, we wouldn't see today's AI. Because today's AIs heavily rely on the heavy training sample. Okay? On the other hand, without AI, we will not see future cloud, so it's really like a twin-, I call this a twin turbo drive engine to the smarter business. So, essentially, you need cloud to provide the computing power, and you need AI algorithms and the theory, to help you to find the methodology to unleash the value of the data. So, essentially, they are both entwined, okay, and a twin turbo drive.

AK: And you mentioned, of course, the cloud computing can-, it is being adopted by many, many different industries, this data can be used by many, many different industries, as well. Where are you seeing the most enthusiastic uptake for cloud, and your AI applications? Which kind of industries, at the moment?

MW: Okay, let me put it this way, so, three years ago, we started our journey of ET Brains, okay, essentially it's a brain. The ET means Extreme Technology. Extreme Technology. Also the (inaudible) of the movie from Spielberg, okay, the ET-, okay, alright. So, we first look at the sectors which by nature is already data intensive, and it turns out to be the smarter city, and a few decades ago, we spent billions of dollars on the infrastructure level, and we deployed so many sensors, traffic cameras, okay, the data has been sitting there for many years, but if you look at the contrasting, embarrassing fact, we've got more and more traffic cameras on the sky, okay, but on the other hand, o the street, the travel speed is getting slower and slower. So, you see the disconnection. Disconnection of the traffic management versus the data collection reflects absence of intelligence. So, that's where we started with. Okay. We first do our City Brain, so, essentially, today, in the past three years, over seven domestic Chinese cities, and two overseas cities, are already empowered by City Brain. So, essentially, we do a real-time coordination of the traffic signal control. We are the first one to deliver the system which can really cut down the travel time of an ambulance by half. More than half, actually. So, even including Kuala Lumpur, the capital of Malaysia, we managed to reduce the travel time of an ambulance by 48.9%. This is in a field test. Now, after the city management, we realized that there are some more areas in the traditional, more traditional sectors, because the manufacturing, in the production line, by nature it is full of sensor data, but unfortunately, nobody really touched this data until today. We see that, since the IoT sensor data has been there, and we have the computation power, why not do a deep dive in to the data, and then look at the hidden value. That's how we started Industrial Brain.

AK: So, in manufacturing, in this example, what are some of the problems now that you think that you can solve with your technology?

MW: Okay, let me share some of the reference cases already. So, especially in the processing manufacturing, by nature, it's a coordination from different-, different workers, okay? It's really like a 4 x 100 relay. Okay. Alright. So, you've got the data passing from one step to another step, but how do you make sure that, if somebody made a mistake in that particular procedure, in the subsequent procedures, whoever has the best chance to override this error in such a way that you can still stabilize the output of the product. So, many years ago, we got the approach of Six Sigma, from GE, okay, Six Sigma, it's an SOP, okay, Six Sigma. But today, we want to do this 100 Sigma, or 1000 Sigma. Okay? So, essentially, we help-, we use the Industrial Brain to help the workers to do a real-time coordination, in such a way that, first, they can reduce the energy consumption in the production, improve the yield of the production line, and stabilize the volatility of the final yield, as well. So, it turns out to be a tangible, measurable percentage improvement of the yield and reduction of the energy consumptions.

AK: A lot of what you've described to me is about data collection practices-,

MW: Yes.

AK: And you're saying that many, perhaps cities, or factories, have this data there, but they're just not using it intelligently enough.

MW: Yes.

AK: But there are cases, perhaps in other industries, like agriculture, maybe in some older factories, where they may not have that data, you will have to install new sensors, new ways to collect that data, which is cost-intensive, it may not be-, the benefits by the company may not be immediately able to be seen, so is that a challenge, for you guys, as you go out to businesses, and their reluctance, perhaps, to adopt some of this new technology?

MW: Yes, you are absolutely right, if you look at the upfront cost, in terms of deploying the additional collection sensors, yes, it is cost expensive to a certain extent. But however, if you look at the potential percentage return, in terms of the value, then the cost is just minimal, okay, it's nominal. So, especially-, let me give you one example, in Xi An, one of the capital cities of Shaanxi province, we managed to deploy the sensors to help the watermelon-, or the melon-, okay, the honeydew, okay, the melon of the farmers, to improve-, to do a real-time monitoring, tracking of this growth curve of the melon, and also look at the humidity, and look at the environmental variables, and decide, 'Okay, what's the best timing to do the precision irrigation, precision fertilizer, and the precision of the pesticides control?' Okay. So, with that approach, we managed to improve the grade of the final product, in this case it's the honeydew, and also reduce the cost of the fertilizer utilization. So, basically, they get a better product with less consumption. So, and then, with this measurable improvement, with the benefit tangible, then they don't care about the one-time cost of the hardware, the IoT sensors.

AK: Would your aim-, with the Industrial Brain, for example, when you talk about manufacturing and factories, to be able to create, essentially, what are fully automated factories, would that be the end goal?

MW: No, definitely not. So, let me put it this way, I clearly say no. Our goal is not autonomous manufacturing, or workerless manufacturing. Our goal is to make these workers smart-, as smart as a PC, powered machines, and then let the manufacturing line think as smart as the people. As the very experienced, senior workers. At the end of the day, we have to make sure that the final product is in good shape, and stabilize the outputs, the qualities. So, it doesn't matter whether it's from autonomous, or whether from the workers, as long as all these workers have one commander in chief, and in this case, the commander in chief is ET Industrial Brain. Okay.

AK: But there does seem like there will be, along the road, some job losses, as things become more efficient, parts of a factory become more automated, so again, it's a similar question I posed to Henry just there, how do you develop such powerful technology responsibly, to minimize disruption, but also make sure, you know, these businesses are seeing the benefits?

MW: In fact, Arjun, I wouldn't say it's a job lost, it's actually-, it's a job transformation. Because, at the end of the day, some of the workers, because they've already got the domain knowledge, they have the domain know-how, and then they can become a trainer, a coach, to train the program. For example, in our Industrial Brain, we need a lot of training samples to be annotated, annotated by experts, who understand, this is a good product, this is a bad product, due to defect type one, type two, type three, for instance. But who are going to do this annotation? It's going to be the workers who've got the domain knowledge. So, basically, in this case, those workers, they don't have to push the buttons and operate. Rather, they just use the mouse, and they click on the screen and annotate, 'This is a sample A, sample B,' something like that, so essentially, they become the next level of the trainer, the coach of this AI program.

AK: We've talked a lot about the development of AI, and where that goes in the future, as well. The processes are getting smarter, they're becoming more efficient, you have more data to train-,

MW: Yes.

AK: A lot of these AI systems on. So, do you envisage AI getting to a point where we talk about general AI-,

MW: Yes.

AK: The ability to have cognitive ability, to perhaps replicate the tasks of what a human can do? Is that a reality, or just sci-fi still?

MW: I think (inaudible) I'd say it's still sci-fi-, let me put it this way, guys. So, if you look at all, all of the-, today's most advanced, state-of-the-art of AI, it doesn't matter if it's a video camera, for image recognition, or speech recognition, or even a chatbot, so they only show a single dimensionality, a single dimension capability, either your vision, or either your speech, or either hearing, something like that, but now imagine, as everybody, you or me, every moment, our brain was actually processing multiple channels, heterogeneous signal input, either from our vision, or either from our hearing, or either from our sensing, okay, and our brain is smart enough to decide, 'Okay, which is my current focus? And which signal has the highest quality, with the least areas of uncertainty?' in such a way that they can smartly decide, 'Okay, how do I respond to the multiple channel input of different quality, of different formats?' But, as of today, nobody-, I mean, nobody in the AI program, is capable of doing that, even to the entry-level approach-, extent. So, to that approach, I will say there is a long way to go. We have to learn how to coordinate multiple dimensionality, the capability, in such a way that as smart as our-, like a (inaudible). Okay.

AK: But we have more and more sensors, cameras on the road, those are our eyes, perhaps sensors on the road, those are like our skin. So, more and more sensors, more and more data, it seems to me that that could be not too distant. So, how far would you say, perhaps, this idea of general AI is?

MW: Okay, let me put it this way. First of all, all those sensors just give you a starting point, which, essentially, you've got all the raw material, which is the data, and then the next is how fast you embrace cloud computing, again, because with the material available, you need the right machinery, you need these tools, okay, the powerful weapons, ammunition, to deep dive in to the data, and today, as we already see that, in the ecommerce world, okay, we use the cloud computing, we serve more than half a billion consumers, but in the future, if you have sensors everywhere, you digitize the real world, in real time. In real time. And then you couple it with the real-time cloud computing capability, and then it is possible that you can really do a futuristic, futuristic reaction, or even preemptive action, to what will be going to happen, okay? But, again, it depends on how much you embrace the cloud computing.

AK: Wonderful, I hope you guys learned as much as I did, during those 15 minutes, that was a fascinating insight in to artificial intelligence and its future. Dr Min Wanli, Chief Machine Intelligence Scientist at Alibaba Cloud, thank you so much.

MW: Okay, thank you.

ENDS

About CNBC:

CNBC is the leading global broadcaster of live business and financial news and information, reporting directly from the major financial markets around the globe with regional headquarters Singapore, Abu Dhabi, London and New York. The TV channel is available in more than 410 million homes worldwide.

CNBC.com is the preeminent financial news source on the web, featuring an unprecedented amount of video, real-time market analysis, web-exclusive live video and analytical financial tools.

CNBC is a division of NBCUniversal. For more information, visit www.cnbc.com.