98. China’s AI Plan for 2030

12/11/2021

A couple of months ago, Eric Schmidt, former CEO of Google, was talking to the US congress. He indicated that, in terms of AI, China may be only 1 year behind the USA. Schmidt was worried that the USA may lose in the next years to China the leading position in AI that they are having at present.

In this post, we are going to explore how the world has reached this situation. We will review the plan of China on AI for the next years, what are the reasons behind it, what it involves and what is its current state.

China’s AI awakening moment

According to Kaifu Li, ex-director of Google China, it all started in 2016 when a reinforcement learning program created by DeepMind won the game of Go to the world human champion, Lee Sedol. For Chinese people, a computer winning a person at this ancient game was shocking. It was at this moment when China realized the importance of AI and how their future as a country depended on leading it.

Moved by that realization, in 2017 China’s government released their strategic plan to lead the world of AI by 2030. They called it: New Generation AI Development Plan for 2030.

In this strategic document, China’s government states that being a leader in AI technology is critical for their military and economic position in the world.

Their goal, states the document, is to become an AI power by embedding AI in all aspects of life, industry and commerce.

China’s advantages and drawbacks

Advantages

They plan to leverage their advantage to other countries on four aspects:

First, a large amount of data that they are generating both online and offline. As you probably know, China has already a society that functions based on the mobile internet. Chinese can pay for anything and everything with their phone. This feature generates a lot of data from each transaction that Chinese people do on it. And this data is not only related to online purchases and actions, but also offline payments made at the supermarket, the barber or the cinema.
Second, a race of startup gladiators, as Kaifu Lee calls them in his book AI superpowers. Those are all the startup leaders that are used to compete against each other in the most difficult conditions, trained in an environment where copying others is seen as the normal thing to do. This implies that for a Chinese startup to survive, they have to thrive even more, move faster, innovate faster, and take possession of the market faster. Otherwise, another startup will copy them and take them out of the game.
A committed government, which is willing to put the resources to achieve the success of the plan by providing the infrastructures and the supporting conditions for companies to flourish. In order to better understand this point, let me clarify that the government AI plan acts more like a wish list rather than an actual list of commands. The government indicates a list of technologies they would like to see built, and then incentivises local officials to promote in the private sector the development of those technologies, by using subsidies, public contracts, and AI-friendly policies.
A society hungry for success. People in China want to be rich, and they will go wherever there is a chance to become one, and they will put all their effort into it.

Drawbacks

Having indicated the advantages of China to lead the AI of the world, the document also indicates the drawbacks they are facing:

1. A Lack of cutting-edge AI talent: The Government of China identifies that they lack people trained in AI and related fields. So in order to solve this problem, they launched in 2018 the AI Innovation Action Plan for Colleges and Universities with which they plan to train 500 teachers and 5.000 students in 5 years.

as for their 2020 report, they have achieved 103 non-university scientific research institutions in the field of AI
in their 2020 status report, they indicate they have 215 colleges and universities with undergraduate major on AI

However, there is still a lack of AI talent. Actually, there is at present a war for talent in China by the most important companies, trying to get the best AI talent to their companies with incredible salaries.

2. Lack of major original results and knowledge in relevant sectors required for the success of the plan, which include:

high-end semiconductors: AI runs of top of them, so if you want to be leader on AI you must master the making of semiconductors.
AI Technical standards. Launched the white paper about AI standards in 2018.
Software frameworks and platforms: none of the popular ones used by the AI community, like Tensorflow, Spark, PyTorch, Caffe…, have been developed in China

3. Lack of an ecological niche that allows fruitful interactions between research institutions and enterprises, allowing the application of latest results into products. China has some big companies working on AI like Tencent, Baidu, Alibaba, but their AI level is still behind the level of the big ones from the states. China also has at least ten privately owned AI start-ups valued at more than US$1 billion, including facial-recognition firm SenseTime, but again, that doesn’t make a rich ecosystem for the transfer of good AI research ideas into products. For that, a bigger ecosystem of startups is required.

In their 2021 report, they indicate that they have so far 2205 artificial intelligence enterprises distributed in 20 application fields, with an 8.39% of them applied to intelligent robots.

Key AI areas to develop

The AI plan of China identifies 7 key AI areas they must master, and provide specific expected results:

Medical imaging systems: the commercialization of medical imaging diagnosis support systems for the early detection of diseases, lead by a machine
1. Goals:
  1. Keep false negatives below 1%
  2. Detect common diseases with 95% accuracy
Audio intelligence: smart devices with speech recognition abilities
1. Goals
  1. 96% accuracy in speech recognition
Connected vehicles: create smart vehicles that are able to autonomously navigate in complex scenarios
1. Goals
  1. Cover low-level automated driving by 2020
  2. Cover high-level automated driving
Language translation: produce translation solutions that are extremely reliable and accurate in multi-language scenarios.
1. Goals
  1. Achieve 85% translation accuracy
Service robots: deploy robots that are able to replace humans in sectors like education, caregiving and cleaning
1. Goals
  1. Overcome challenges of novel scenarios
Unmanned aerial vehicles: vehicles with completely automated cruise control capable of operating in highly complex environments.
1. Goals
  1. 360-degree omni-directional sensitivity
  2. An accuracy margin of 0.005 degrees
Image recognition: this is one of the most ambitious goals. Starting from improving image recognition, they also are pushing into video understanding and summarization, search of specific images inside a video and human-video integration
1. Goals:
  1. Help build China’s all-encompassing Social Credit System by 2020

What I specially like about the plan for each area is how specific are the goals to achieve and deadlines for them. This is the SMART technique which has been demonstrated to increase the chances of success.

I haven’t found any report that explains the results obtained in each one of those areas. Maybe I wasn’t able to find it, so please let me know in the comments below if you have results for any of those lines of research.
However, I suspect that at this point, the results may not be as good as expected. We can say that we will have 20 new research centers for next year, because for that, the only thing that you need is money, people and somebody that leads the development. If you need to speed up, you put more people, more money and then it speeds up.
But scientific progress doesn’t work like this. To a big degree, research doesn’t depend on the money you pour in it. It depends on the research pace itself. So it is almost impossible to predict when a certain breakthrough will be obtained (that is actually the situation of AI since its infancy (true AI is always 20 years ahead the current time!).

The plan in time frames

China has committed hundreds of billions of dollars to the success of this plan. In order to achieve success, the full plan was divided into three time frames:

First one, by 2020, to put China into the world in each one of those subjects, this means put China up to date in all those fields.
1. We are already 2021. Can we consider that they have achieved this goal?
  1. So far, based on their 2018, 2020 and 2021 reports, China reported the following achievements:
    1. #1 in both total AI research papers and highly cited AI papers worldwide
    2. #1 in AI patents
    3. #1 in AI venture capital investment
    4. #2 in the number of AI companies
    5. #2 in the largest AI talent pool.
  2. In 2021, a Stanford report indicates that Chinese AI researchers are being more cited than any others.
  3. So yes, we can consider achieved this goal.
Second one, by 2025, Achieve major breakthroughs in AI in all the previously indicated subjects
Third one, by 2030, dominate each of those sectors and become the world AI leader

The dangers of this plan

While China is moving forward with their plan, some voices inside the country have started to point about the dangers of working towards achieving it. Some officials state that there is a chance that global competition over AI could start an AI arms race among countries which could lead to a war. The officials are concerned that states may be more prone to attack other countries with AI military systems due to the lack of casualties. On top of that, the automation of decision systems lead by AIs, can cause many misperceptions and quick conflict escalation, leading again to a war.

In order to prevent those situations, 2018 China’s Academy of Information and Communications Technology published the AI Security White Paper, where they advocate for and international cooperation to create norms that govern the application of those systems at the level of the planet. Then a dialogue between the countries has already started to try to prevent it.

Actually, I consider this point one of the real dangers of artificial intelligence: not the creation of intelligent robots that enslave us, or machines that go wild and convert the whole world into paperclips. But instead, the creation of smart enough machines that take decisions by themselves and escalate conflicts between humans due to misunderstandings. That is, those intelligent devices misinterpret information and then we make fatal decisions based on those suggestions of the AI. That is for me the real possibility where an AI can destroy humanity.

Conclusions

We are only 9 years before 2030.
Previous China’s plans have been on track (like for example, remove poverty from their country or the first phase of the 2030 AI plan). Chances are they will achieve it.
Additionally, other China plans are in the middle which are also pushing this one: Made in China 2025 (we’ll talk about it in another video).
What I’m pretty sure is that this plan of China has revolutionized the AI world and is speeding it up, which is good for all of us.

As a final comment, I would like to stress how China’s AI plan puts a lot of emphasis on using the rules of the market to get results, in terms of commercialized products. This means, the results of the AI research must be closely tied to AI products that the market wants. That is why they stress the interconnection between education, research, investment and building enterprises.

Final thoughts

So given that speed that Artificial Intelligence is getting, the whole field is going to need a lot of experts in AI and robotics.

“The future [of AI] is going to be a battle for data and for talent,”

David Wipf, lead researcher at Microsoft Research in Beijing.

So this is clear: there is a lack of AI and robotics talent

If you want to become part of this revolution, you need to start now and get up to speed about artificial intelligence and how to apply it to robotics. For that, go into The Construct’s Robot Ignite Academy and become one of the pioneers of artificial intelligence applied to robotics. The opportunity is now, because there are not many people doing that right now.
So start now learning about Deep learning, Reinforcement learning, machine learning for robotics at The Construct. Take our step by step learning path on AI for robotics, practice with the simulated robots and then connect to our remote robots located in Barcelona, Spain and practice with them from your location.
All the courses are practice based, including robotics theory learning. So you will understand why you are learning a certain AI algorithm or subject.

Closing

If you liked the post and would like to know more, check below the links to all the documents I examined to build this video.
In the next posts, I will be discussing the response of the USA to this plan, and the plan of Europe… if any!
Let me know what you think about this kind of video in the comments below. Do you want me to do more like this?
Thank you and keep pushing your ROS learning.