Cover Story
Inclusive AI: Making Intelligence Pervasive

First, let's go back half a year to an article in which I mentioned that it was evident that the current structure of the AI industry in China has huge space for improvement and enterprise adoption of AI will become the norm in the future. At that time, a media colleague argued with me that the AI for the business market could not support the need for the utilities. His support was that the Chinese Internet had been around for a decade, but that the business market never became the mainstream support for the industry.

Over the past year, industry-specific AI adoptions have become popular, and Internet companies have turned their focus to the industry market. It seems that common sense is not always reliable.

I am not sharing this story to prove I was right, but rather to nudge our thinking. What if we had listened to naysayers and didn't build the bridge for enterprise to pass on the road to AI adoption?

We had to be determined to bring AI from the lab into actual application. We had to move from theory into the application for the busy workshop, the bustling intersection, and the growing enterprise.

There is a story that can be used as a reference to depict a part of our journey. At that time, AI was still a general-purpose technology that yet to awaken.

In May 2018, HUAWEI CLOUD announced its Inclusive AI strategy that promised to make AI affordable, easy to use, and highly adoptable. This simple but unique proposition would undergo a multitude of parallel spurts over the coming year.

Let's take a look at the first year of development after Inclusive AI was introduced and loop back to how AI moved from the lab into industry space and from online simulation to actual operation in the economy.

Thousands of engineers, designers, and developers work in the AI that runs on HUAWEI CLOUD, each contributing an ingredient, a utility, an improvement to what culminates into Inclusive AI.

How did these people make it possible for AI to move into the physical space and integrate with thousands of industry-specific applications? I propose that there are four main factors that contributed to the thrust behind Inclusive AI.

First, development of use cases to demonstrate the value of AI in the enterprise market.

Second, enablement for developers to build the ecosystem.

Third, concerting technical capabilities to achieve a comprehensive, intelligence-enabled service stack for the enterprise market.

Fourth, realizing the core value of AI through integration with the industry, and establishing a long-term development strategy for the industry.

Let's look more in depth at each factor.

Zheng Yelai, Vice President of Huawei and President of Huawei Cloud BU announcing the Inclusive AI strategy

From theory to real-world application: AI enters the first round of industrial reality

A year ago, Beijing commuters probably couldn't have even imagined that their daily passage through certain parts of the city would benefit from AI. Even today, most are still unaware that AI has quietly entered their life, though many have perceived that certain streets are much less congested.

In April 2018, HUAWEI CLOUD collaborated with the Beijing Municipal Administration of Traffic Management to carry out the pilot application for signal timing optimization using AI technology in Haidian District of the city. The pilot involved installation of AI recognition and decision-making devices into the traffic light systems.

After the HUAWEI CLOUD traffic intelligence solution was deployed, the average delay on one of the most congested roads in the city was reduced by 15.2% while average vehicle speed increased by 15%, according to the assessment from Cennavi Technology Co., Ltd., China's leading traffic information service provider. 

The real value, however, can be felt by those commuters who now arrive home 10 minutes earlier than before. 

Speaking to the first factor, it was and is important for HUAWEI CLOUD to apply its AI suite to real-world applications in order to demonstrate viability for others. Such cases show the value of combining AI with specific industries and allows others to rest assured that their adoption will be effective. Huawei AI is more than just a bunch of theory on paper, it is becoming a core value proposition for organizations taking on the adoption.

The AI suite in HUAWEI CLOUD has been proven effective in case after case. The AI services have been tried and tested in more than 300 projects in 10 industries. From June to October 2018, three Intelligent Twin applications in transportation, industry, and city management were released. The first round of projects in these sectors has yielded great results.

One example includes Beijing Sanlian Hope Shin-Gosen Technical Service Co. Ltd., China-based provider of nylon polymerization and spinning overall engineering and technology solutions. With the intelligent analysis capabilities of the Industrial Intelligent Twins, production data is intelligently analyzed, and flexibility of production line is improved to meet dynamic needs so the enterprise can better respond to the personalized requirements of the downstream. In addition, the overall data analysis solution implements cloud-based training, edge deployment, and real-time product classification. According to preliminary tests, the match rate is improved by 28.5% for downstream requirements.

In the medical field, the AI team from HUAWEI CLOUD working specifically on applications for healthcare industry cooperated with the KingMed Diagnostics in 2018 and made a breakthrough in the field of cervical cancer pathology. The sensitivity (true positive rate) exceeds 99%.

The results from the 300 plus cases substantiate the feasibility of AI adoption to the enterprise market. Major industries, such as transportation, industry, Internet, and medical care, can look at the successes from these cases and know that their needs can be met in the context of their organization.

Enable the Developer, the Best Engine for AI Growth

I have interviewed an AI developer who wants to develop a dedicated navigation device for the sight-impaired to help them identify traffic and traffic lights, cross street bridges, navigate subway stations, and so on. The solution is clearly something the sight-impaired need, but he told me that there was a great deal of difficulty in actual development because the training for machine vision models requires copious volumes of data, high computational costs and a long training period. All these complications made it difficult for the single developer to overcome the hurdles and get the application to those in need.

From the perspective of underlying technology, AI is not a one-to-many unidirectional output of technology, but rather a many-to-many bazaar of tech. This makes collaboration all the more important in the world of AI. 

Developers are still faced with towering barriers to bringing their ideas to fruition. They are faced with issues in computing resource, data governance, development platform compatibility, and other deployment issues. However, with the development of AI and its specific application in various industries, getting their ideas to market is now not such a painful proposition.

Enabling developers in turn spurs development in AI, so it is the best approach to broadening adoption and adding vitality. To enable more developers to integrate broad-spectrum utilities into their own ecosystems and allowing them to find opportunities in AI for Industry is a sure-footed approach to spurring on growth.

Another major task of Inclusive AI strategy from HUAWEI CLOUD in its first year was to open up the ecosystem to collaboration and provide technological enablement for developers. At HUAWEI CONNECT 2018, Huawei released the full-stack, all-scenario solution powered on the new Ascend series of AI chips. This release gave developers new answers in solving issues in the underlying logic for their applications. The AI HUAWEI CLOUD AI is leading the industry with release of a series of platform tools for developers. The launches include ModelArts — the faster AI development platform, and HiLens — a visual AI application development platform that focuses on actual development requirements. These toolings help developers lower the threshold to entering the AI arena. Huawei also released its AI Developer Enablement Program and Ecosystem Partner Program.

In March 2019, HUAWEI CLOUD launched an AI marketplace for developers. The marketplace includes the first platform in China that allows developers to release and subscribe to AI model services. It aims to provide a secure, open sharing and transaction environment for scientific research institutes, AI application developers, solution integrators, enterprises, and individual developers. The platform effectively connects participants in the AI development ecosystem to accelerate rollout of AI-related products.

On March 20, 2019, Stanford University released the latest DAWNBench ranking. ModelArts was ranked first in image recognition training and inference.  

Launch of the Huawei AI Developer Enablement Program

Stanford University DAWNBench Training Time List (March 2019)

Stanford University DAWNBench Inference Performance Rankings (March 2019)

Concerting Efforts So Industry Customers Can Get More Out of AI

The third focus for the Inclusive AI strategy is streamlined internal and external implementation. As of the end of Q1 2019, the AI services in HUAWEI CLOUD numbered 59 with a total of 159 functions. We have also participated in many competitions and tests to show our strong position. We have developed partner maps to provide customers with better technical choices.

The release of multiple Intelligent Twins offerings extended the capabilities of the AI portfolio in HUAWEI CLOUD. In the face of complex physical world problems, enterprise customers require easy-to-use technologies and a set of flexible solutions armed with smart brain, intelligent edge, and device-side perception systems. 

The combination of technology, product, and service is a symbol of the maturity of the AI services in HUAWEI CLOUD and the beginning of the in-depth integration of the Industry +AI model.

Realizing the Core Value of AI and Industry Integration

The most important factor in realizing the core value of Inclusive AI is in what it can actually do after being brought into the organizational and industrial profile and how it can be applied to the verticals then down to the specific company, organization, and factory.

Before, those that understood AI usually did not understand the dynamics of the particular enterprise while those that had a good understanding on the enterprise did not understand the dynamics of AI. As a result, both sides had to grope about in the dark trying to get the AI apparatus to benefit the organization. A distant light is a welcomed sight when fumbling around in the dark trying to find one's way through the thick fog and heavy night. The distant light gives the direction, which is what the Inclusive AI strategy did for us — it gave us the start point with the benefits to the user in mind.

HUAWEI CLOUD has experienced success through trials. The experience we have gained, the adjustments we have made, the tuning into customer needs, and the hand-in-hand journeys with our early adapters have given others a well-lite path to follow in their AI adoption. Our practices have been recognized and acclaimed as highly viable approaches in the various industries we have already gained success in.

AI is yet again transforming IT technologies from being just support systems into a high-value production asset. AI is creating impressive value in a wide range of production scenarios across many industries. With the proven success we have gained at organizations of all types in regions throughout the world, HUAWEI CLOUD is continuing to up its game in AI tech and promoting its adoption in the industry space. We are committed to making AI affordable, easy-to-use, and highly applicable to the industry and organizational profile — which is what we mean when we say 'Inclusive'. We are not only talking about a bunch of theory or putting a bunch of information on a PPT, we are converting all the theory into practice, placing the value of data into production, and gaining more expertise with each success!

That is not to say that we have overcome all the pain points in joining the industry intelligence with the AI applications. In fact, we still have issues to wade through, but we are doing so with one foot firmly in front of the other as we stay our course. Jia Yongli, General Manager of the HUAWEI CLOUD EI Service Product Dept, said: "A good question is better than ten algorithm engineers." We have taken this as a guiding principle. What this means is that we have to ask the questions upfront, get involved and understand the dynamics while avoiding the outsider-looking-in mentality. We have to dive in, dig in, and then widen out. AI should not be looked at as general-purpose. It is an accelerator and lubricant. It is connected to real people, technologies, and production lines. Are industries ready to take the leap? This is the question we need to ask. 

Industries need to adopt big data utilities, but the problem is scarcity of computing resource in the existing mixes. Machine learning uses intelligence-enabled apparatuses in the learning models then provides a solution in the reverse direction to improve productivity. In essence, the intelligence needs to be activated! In this process, the industry-specific data learned by the intelligent entities in the machines becomes the foundation for everything. Today, however, the actual situation is that various industries are producing a large amount of data, but the collection, storage, and learning of the data is complicated and such processes are compute intensive while resources run scarce. AI for Industry needs to first consider how to achieve high quality while ensuring affordability. These contradictions need to be addressed front the start to gain success in the intricacies involved in implementation.

We are accumulating valuable experience and learning from the data as we dissect AI and apply it vertically, removing the mystery and finding our way out of the heavy fog that we were in a year ago. Things are becoming clearer as we progress on the journey. HUAWEI CLOUD adopts the down-to-earth approach of learning by doing. We started with the specifics of the particular industry instead of making the industry fit into the technology package. We start with the problems that need immediate resolution while keeping the long-term in clear view.

Jia Yongli, General Manager of the HUAWEI CLOUD EI Service Product Dept

The Greatest Measure

Summarizing the first year of Inclusive AI, the first thing we did was to prove the value of adoption to the enterprise market and open up a point of entry for developers to join our effort. We demonstrated the viability of the products and formed an initial strategy for the long-term that could fit the logic of the market.  

It was just the beginning, but we remained utilitarian in each stage. Everything we did over that stage was a build up to the release of the Ascend chips, which provided a powerful connector for joining the ends of our efforts. The release provided the infrastructure - the powerhouse - for industry to better utilize our AI ecosystem. We leverage the wisdom locked inside of industry practice to provide all-scenario solutions. This is the remarkable story of how we made AI affordable, adaptable, and ease to use for organizations.

Everything is difficult at the start, but the mystery slowly reveals itself if we keep digging. This is the natural order of things. The greatest measure is what we do when faced with difficulties. The courage to take on new possibilities yields rewards if we stick to our commitments and let the market guide our tech. 

AI adoption, especially in the Industry +AI context, still has lots of issues that need to be worked out, but we are moving ahead one step firmly in front of the other. We are growing AI and enabling Digital China by remaining down to earth in our approaches and strategies.

When AI is fully launched, we will bring in more tech from the cutting edge while we also continue to seek wisdom from reality and continue to combine tech in practical ways that benefit the industrial economies. It takes courage to remain realistic, and that is precisely what Inclusive AI from HUAWEI CLOUD promises to do. Find out how our AI portfolio can benefit your organization now and well into the future. HUAWEI CLOUD — helping you navigate through the fog!