ChinAI Newsletter #32: Huawei Goes the AI Way
Welcome to the ChinAI Newsletter!
These are Jeff Ding's weekly translations of writings on AI policy and strategy from Chinese thinkers. I'll also include general links to all things at the intersection of China and AI. Please share the subscription link if you think this stuff is cool. Here's an archive of all past issues. *Subscribers are welcome to share excerpts from these translations as long as my original translation is cited.
I'm a grad student at the University of Oxford where I'm the China lead for the Governance of AI Program, Future of Humanity Institute.
Huawei’s AI Strategy
Two translations on Huawei’s pivot to AI, publicized at its big annual Huawei Connect conference this past week. The first one, leans more toward a press-release, but has some juicy details. Stay with me for the second one though: it’s a thorough examination of Huawei’s prospects in the field of AI+Security, a nice connection to the past couple of issues.
“A huge AI supernova is rapidly evolving,” says this piece from the S&T media publication xinzhiyuan (AI era). Huawei announces two AI chips (Ascend 910, which they claim has the highest density of compute for a single chip, surpassing Google and NVIDIA in that respect, and the Ascend 310). In addition, they want to create a full-stack AI portfolio, planning to launch a unified deep learning open source framework for integrating software and hardware.
Most interesting passage from the piece on commercial samples of Huawei’s new AI chips and potential customers: “In the past few months, Huawei executives have been recommending it to Microsoft, expecting the latter to use the chip in its cloud computing service Azure China. According to informed sources, there is a dedicated group of Huawei engineers who are developing the software running on the chip to meet Microsoft standards. Engineers are also running Microsoft's algorithms on the new Huawei chips, such as the Bing search engine's speech recognition software, for testing.”
*Microsoft Azure is notable because it was the first foreign public cloud service provider offered in China in compliance with government regulations.
Huawei’s Move into the Security Industry
Huawei has elevated security to be one of the company’s “strategic business lines” — a phrase mentioned 23 times over the course of Huawei’s release event in Guangzhou of 30 new cameras, in August 2018. Ambition is to be the top 3 in the world in video monitoring/surveillance. Piece is nicely structured into three key questions:
A. Why does Huawei want to get into the security industry?
The government agenda is key here: “A cloud for the entire province,” “A cloud for the entire country” are sayings that reflect government prioritization of cloud-enabled public security functions
After government proposed “billion-level” projects such as “Safe Cities,” everyone rushed to respond: still, AI-enabled video surveillance products account for a small proportion of the actual applications (My view: perhaps because everyone rushed too quickly and put in infrastructure that is difficult to scale and integrate across platforms)
Huawei also needs to find industries with room to grow, its product share in a lot of the current industries it’s involved in has reached a ceiling
B. Can Huawei do a good job of security?
Huawei’s core competitive competencies:
CHIPS CHIPS CHIPS (Huawei’s Hisilicon chips account for more than 60% of the global share in the security industry, stronger and more efficient chips empower cameras to have more functions
Overall technical strength - just a giant in terms of overall R&D ($15 billion USD annual budget)
Standardization/Integration - current issue for video monitoring is “data chimneys” — “the data of each police unit and the video data across different regions cannot be shared” — Huawei has the capacity to standardize the market, its open architectures could allow any camera data to be transmitted for unified analysis
C. How does Huawei do security?
For me, this was actually the most interesting section because it got into the technical details of implementing facial recognition/smart security that I don’t think many policymakers/commentators actually understand. Let’s walk through one quick example, more in the full translation:
Many facial recognition cameras, as well as the demo videos you see online, only work in constrained scenarios (good lighting, no obstructions of the face, etc.). This is the case for license plate recognition which is constrained when there’s low light, blurred motion, and large tilt angles. Huawei claims its camera can remove those constraints because it simultaneously captures the linked face+body/license-plate+car body so that even if someone obscures her face/changes her hairstyle, she can be identified, and it gets >85% recognition rate for license plate recognition in low lighting, blurred motion, and large tilt angles.
Full Translation: Tackling Security, does Huawei have a Chance?
This Week's ChinAI Links
Chinese idiom of the Week: 巧妇难为无米之炊 (qiao3fu3 nan2 wei2 wu2 mi3 zhi1 chui1): even the cleverest housewife can’t cook without wife. Used in context of transitioning to the “How Does Huawei Do Security?” section, what are the tools that Huawei has to enter the industry.
CBInsights has done some really good research on China’s AI scene. Here’s their brief on the growth of surveillance tech in China.
My boss at the Governance of AI Program, Allan Dafoe, testified in front of EU Parliament this past week on international governance of AI and what he calls the need for a “cooperation metanorm.” Definitely worth a listen (his section is from 10:17-10:33 of the video)
From our friends at Import AI, Huawei’s Hislicon chip won an AI Benchmark competition that tested 10,000 phones for the performance of AI software.
Fascinating discussion on this Twitter thread surrounding a past ChinAI translation on China’s chip strategy. The hook is the rhetorical/joking question posed by the author: “Can we only rely on peaceful reunification?” (to catch up to Taiwan’s TSMC in chip fabrication). The resulting discussion is a great exploration of the complex supply chain in semiconductors as well as the importance of accumulated “learning by doing” gains in this field.
Thank you for reading and engaging.
Shout out to everyone who is commenting on the translations - idea is to build up a community of people interested in this stuff. You can contact me at jeffrey.ding@magd.ox.ac.uk or on Twitter at @jjding99