Искусственный General Int ...

Почему мы так много слышим о ИИ, доминирующем в Китае, когда (по состоянию на 2019 г.) большинство ведущих исследователей ИИ и исследовательские центры находятся в США, Великобритании или Канаде?

Любит | Нелюбов | Ответы | Вид | 3035


Янус Донгье Кименг интересующийся историей и географией Китая   
@ | Обновлено Right Now


As a researcher in the AI field working in the UK, I would like to share some of the honest opinions about this field.

Firstly I would say the entire global AI research community has yet to make significant breakthroughs. Conventional AI based on mathematical reasoning has just failed to achieve higher accuracy than current machine learning based techniques. Despite so many variants in building different learning networks or optimization techniques, we still can’t escape the curse that everything we’ve built is just statistical fitting concluded from a large chunk of training data.

For example, you can train a human baby to recognize a cat by showing it several cat pictures. But we have to use millions of cat pictures and non-cat pictures to train a neural network to recognize cats in a machine.

Therefore, before our AI research gets any real breakthroughs, the current key to “AI dominance” is still the training data. Thus the one who owns millions of cat pictures wins. In this case, you can’t deny the fact that China holding a huge amount of data that matters. The Chinese government, state companies, and private companies have all their means of collecting data from the 1.4 billion people and build their own knowledge graphs for all different purposes.

The knowledge graph may consist of information about people such as user profiles, preferences, tastes, payments, transactions, and connections. The graph can also have geographical, physical information such as real-time camera footages, regional goods demand and supply such as live taxi orders and live food takeaway orders etc.

Yes, all US companies such as Google, Facebook also collect user information as their own private assets. But what’s different for those Chinese companies is that they have to follow Chinese law so that their data can be accessed and coordinated by the Chinese government. This is similar to having a central command that owns and manages all the data across companies. With this efficient control of data, China has much deeper data usage ***** than the US AI companies.

So enough about the data, how about technology? Does China really dominate the AI research field?

If you look at all the established publishing venues in machine learning such as CVPR, NIPS, AAAI etc, China is still far behind from the US and Europe in terms of citation counts.

Does it mean China is lagging behind the US?

Of course not. This is far more complicated than just thinking in terms of publications.

Let us instead have a look at the authors from the publications in the US. From this link can you identify how many papers in CVPR are actually written by Chinese Ph.D./researchers in US universities and companies?

https://dblp.org/db/conf/cvpr/cv...

If you don’t have time to look at it, then I will tell you that over half of the papers are written by the Chinese researchers (based on their special PinYin names, such as Zhang, Xiang etc). And I am sure most of them are international students studying in the US but not US citizens.

Have you ever thought of where would they end up going after graduation?

The best of them would be staying in US academia, and the second best of them are recruited by Google, Microsoft or Facebook. The rest of them might go to a few small start-up companies in the US. Then the remaining talents would prefer to go back to China for better opportunities.

For those who stay in the US, how many of them are actually staying permanently in the US? Imagine you work hard and climbed the ladder aiming to the top. The US government enforces various restrictions on the Chinese ethnicity in fear of IP theft and spying. Your Indian colleagues get promoted but you still remain in the old rank. You hit the implicit ceiling and you have no choice but to return to China. The Chinese company would give you a competitive pay in China at a similar level to that of silicon valley. Of course, you will bring your experiences and knowledge back to China.

Eventually, most of the Chinese talents would return to China and contribute to the vast ecosystem of the Chinese AI world. This would make China catch up quickly in technologies. I hope this can solve a few myths questioned by many Americans. Now you should be able to understand why the newest technology actually can get migrated to China so quickly. Is this really IP theft?

In reality, the Chinese talents are kicked out of US, bringing knowledge and experiences back to China. And it is the US government’s sore fault. I know they won’t even admit this, but at least please call it a “talent fishing” to be more accurate.

(The above scenarios are only applicable to AI researchers but it might not be the case for other professionals.)

Meanwhile, the Chinese government is undertaking every measure to lure those talents back to China. Firstly, the Chinese AI companies and startups get enough domestic investments so they can create as many positions in China.

Secondly, talents working in the AI field enjoy a 75% reduction in their personal income tax from some regions in China (this is even a huge plus for the insanely high salary). And there are also various benefits and prizes from the government in housing mortgages and insurances. As far as I know, the US does not have such competitive policies or benefits from the government.

Thirdly, those talents would be working on the *real* projects instead of the toy projects they worked on when they were in US and UK research centers. This is because those projects are mostly from government contracts, which is catering directly to the 1.4 billion people and enjoys a greater abundance of training data.

Or if you are working on private projects in startup companies, you will have much more freedom in doing everything because you don’t need to worry about political correctness, work ethics, or lack of data. You will be gaining special privileges from the Chinese government in doing business with others because you are the “foster target” of the Chinese government. The government caters your AI startups just like their own state enterprises.

For example, an AI startup company is given the Chinese government licensed VPN so that all its employees can freely access the resources from Google, Github etc. And the company also enjoys discounts on renting offices and paying taxes etc.

Finally, thanks to those talent programs and policies, China is gradually taking over the lead in AI adoption and applications. It might not be the pioneer in the most cutting-edge of machine learning technologies in the research domain. But China is a devout believer and leading adopter of the existing AI technology.

Have a visit to China, especially Hangzhou and Shenzhen, and you will know what I mean.

| |



Онлайн-курс
«Машинное обучение для людей»
Один из самых трендовых курсов в сфере Цифровой экономики.