🇨🇳 ChatGPT is fine, but China's long-term AI race is not going away
While ChatGPT has been unleashed on the world, I started to read about European politicians and intellectuals demanding our version and asking why we couldn't invent it. This amounts to discovering a jungle and asking that we plant trees immediately to catch up ('sigh'). And while we are focused on the U.S. (again), we turn our back on what's going on in the East (again).
Ironically, only a few years ago, China was center-stage in discussing what would be the future of AI systems unleashed at large scale in our societies. Social scoring, retail automation, autonomous vehicles ... All were hot topics and, quite often, triggering ones. Note that at the time, I was already flagging microchips as the critical bottleneck of all AI (not forgetting that Europe, or at least the Netherlands, was holding a major ace up its sleeve).
As we are (justifiably so) fascinated by OpenAI and Microsoft, we just forget how larger and more comprehensive China's efforts have been opening as much data as possible deep learning AIs at scale for the last five years. We could argue how good their technology is compared to the U.S. Still; we can't argue about their fearlessness about experimenting on their society and (as importantly) the vastly bigger their data set will ever be.
It's worth remembering that in 2017, the Chinese Central Party announced a sweeping initiative to make China a leader in the AI race by 2030. Since then, things have accelerated without much concern from the West.
And because, by nature, deep learning tools are what they digest, to begin with, it should be fascinating to see how they evolve and morph into products with different minds of their own, depending on where they were grown.
- Sensetime (2014) is probably the most well-known and largest facial detection company involved in different smart city projects all over China.
- Megvii specializes in "AIoT" for consumer, industrial, and governmental markets.
- Cloudwalk (2015) also works on big data for finance and supply-chain.
- Uisee (2016) develops self-driving backend and embarked platforms.
- Meta Sota (2018) is a research and writing tool for Chinese lawyers that vastly automate their workflow and allow traduction for international contracts.
- Fadada (2014) started as an e-signature company for Tencent's and governmental operations; they now provide on-the-fly reviews and risk evaluation of any type of business contract.
- Mujicv generates CV optimized for recruiters from simple inputs.
The usual suspects in media and content generation:
- Ernie-ViLG (Baidu) generates images from text with a Chinese cultural point of view and with (seemingly) better performance than Dalle or Stable Diffusion.
- "Caiyun Xiaomeng" is a story generator that only requires a title or short text input to generate lengthy fiction.
- Pan-Gu Coder (Huawei) focuses on generating and debugging code from simple natural language inputs. Unlike ChatGPT, this AI tool focuses solely on solving mathematical and software problems.
- Cambricon (founded in 2012!), just like Nvidia, develops and packages high-end GPUs for cloud and edge AI infrastructures.
- Ubtech (2012) develops automated education and industrial robots.
I could go on and on... And I didn't even list Baidu and Alibaba as prominent cloud AI solutions providers.
What's the key takeaway?
Just like with mobile e-commerce ten years ago, and the way Weibo and Facebook became two parallel universes of the same principle, the next few years promise to be a gigantic A/B testing of AI between the East and the West. Our societal and business divide will grow further on the one hand. Still, on the other hand, understanding how China's deep learning AI ventures take off and what they focus on will prove invaluable in building our own market's edge.