China Strengthens Ethics Reviews on AI, Life Science; Users Can Turn off Recommendation Algorithms; Chinese Self-Driving Startup Raises $400 Million

Weekly China AI News: March 21 to March 27

Dear readers, all 132 passengers and crew on the China Eastern Airline’s flight MU5735 that crashed earlier this week in southern China have been confirmed dead. The miracle didn’t transpire. I wish we could restart 2022 as we have been experiencing a tragic, heartbroken three months in 2022. R.I.P. to all souls on the board. 

Also starting this week, I am making a few changes to my newsletter, including more uses of bullet points and brief explanations of my news selections. I wanted to thank one of my friends and a newsletter master, Cai, for her terrific advice on taking my newsletter to the next level!

News of the Week

China Sets Ethics Red Lines for Life Sciences, AI Research

China’s top authorities this week released a set of guidelines to strengthen the governance over ethics in science and technology, “the first state-level guidelines on sci-tech ethics” and “another milestone event following the establishment of the national committee on research ethics”, the state media described. Life sciences, medicine, and artificial intelligence are three key areas where oversights and regulations are expected to be tightened.

Why it matters: Any future sci-research studies, especially those on human subjects or animals, should seek approvals from a qualified own-party scientific review committee, the document mandates. The national sci-tech ethics committee will compile a list of high-risk research areas so any future scientific studies that fall into this category must receive the green light. The list will be rolled out by the end of this year.

With this guideline implemented sooner or later, the “CRISPER Babies” created by Chinese biophysics researcher Dr. Jiankui He three years ago are unlikely to transpire again in the future.

Other highlights:

• The sci-tech ethic governance in China started late and some cutting-edge research is getting into no man’s lands, said a deputy minister of the Ministry of the Science and Technology explaining the background behind the guideline in a press event.

• The guideline is centered on five ethical principles: serve the well-being of humanity, respect people’s right to life, adhere to fairness and justice, control risks in an appropriate way, and maintain openness and transparency.

• Universities, research institutions, medical and health institutions, enterprises, etc. are the primarily responsible subjects for any ethics violations.

Users Can Turn off in-App Personalized Recommendation

You swipe a short video on TikTok in three seconds and the app’s recommendation engine already learns your interest better than yourself, putting the next content in front of you that AI thinks you’d like. Recommendation algorithm is embedded in almost every app, from Amazon presenting products based on your search history to Spotify discovering music that is spot on. You get addicted to apps, but the protection of your privacy is compromised as well.

Starting mid-March, Chinese Internet users can turn off in-app personalized recommendation algorithms as the new rules on algorithm recommendation services have been implemented on March 1. The new rules, which were released in earlier January, required all recommendation algorithm service providers to:

• Inform users in a conspicuous manner of its recommendation services;

• Provide users with services that are not tailored to their personal interests;

• Offers users a convenient option to turn off the recommendation algorithm service.

Why it matters: WeChat, Taobao, Baidu, Weibo, and other apps have followed the suit, a monumental step towards a privacy-focused mobile Internet. It’s only the beginning as users are endowed with rights to view what data have been collected and delete any personal data. 

Papers & Projects

Time for CVPR’22 & ICLR’22 papers!

AI estimates 3D poses of multiple people from one single image

Researchers from Meitu’s MT Lab and Beihang University’s CoLab proposed a novel model that can predict the 3D pose of multiple people based on one single RGB image. The proposed model, named Distribution-Aware Single-Stage (DAS), is a fully Convolutional Neural Network that can “simultaneously localize person positions and their corresponding body joints in the 3D camera space in a one-pass manner.” Read the CVPR’22 paper Distribution-Aware Single-Stage Models for Multi-Person 3D Pose Estimation

How to train a text-to-image generator without text data

Training a text-to-image generator in the general domain (e.g., Dall.e, CogView) requires huge amounts of paired text-image data, which is too expensive to collect. A team of researchers from ByteDance took advantage of OpenAI’s language-to-vision model CLIP and proposed a self-supervised scheme named CLIP-GEN for general text-to-image generation. Read the paper CLIP-GEN: Language-Free Training of a Text-to-Image Generator with CLIP.

Autonomous vehicles predict trajectory from car-passenger relations

Predicting future motions of road participants is an important task for driving autonomously in urban scenes. A team of researchers from Tsinghua University and MIT exploited the underlying relations between interacting agents and decouple the joint prediction problem into marginal prediction problems. The proposed approach M2I first classifies interacting agents as pairs of influencers and reactors and then leverages a marginal prediction model and a conditional prediction model to predict trajectories for the influencers and reactors, respectively. Experiments show that their simple but effective approach achieves state-of-the-art performance on the Waymo Open Motion Dataset interactive prediction benchmark. Read the paper M2I: From Factored Marginal Trajectory Prediction to Interactive Prediction.

Rising Startups

WeRide, an autonomous driving upstart, has raised almost $400 million in its latest funding round joined by GAC Group, Bosch, China-Arab Industrial Investment Fund, and Carlyle Investment Group. Founded in 2017, the Guangzhou-based company is now valued at $4.4 billion, up from $3.3 billion a year ago. WeRide owns a fleet of 300 autonomous vehicles — robotaxis, robobuses, and robovans — that have accumulated over 10 million km of L4 test mileages and now operate in Guangzhou, Shenzhen, Shanghai, and Abu Dhabi. You can visit the company here.

Rokid, a developer of AI/AR tech gadgets and robotics software, has raised $110 million in its Series C funding round. Founded in 2014, the Hangzhou-based company has rolled out three types of AR glasses targeting industry clients with various work environments, from high-risk areas such as oil & gas, electric power to exhibition, education & training. You can visit the company here.

Tsing Micro, a company in reconfigurable computing chips, has raised several hundreds of million yuan in its Series B funding round. Founded in 2018 as the reconfigurable team from Tsinghua University, the company has mass-produced three chips based on coarse-grained reconfigurable arrays (CGRA), a reconfigurable chip architecture with a shorter reconfiguration time compared to FPGA. You can visit the company here.

Rest of the World

Nvidia CEO Jensen Huang introduced the new NVIDIA H100 Tensor Core GPU based on the new NVIDIA Hopper GPU architecture at the company’s annual flagship event GTC 2022. H100 is built with 80 billion transistors using a cutting-edge TSMC 4nm process designed for NVIDIA’s accelerated computing needs. Read the deep dive about the new GPU architecture here. Nvidia also announced that BYD, the world’s second-largest EV maker, will use the NVIDIA DRIVE Hyperion computing architecture in its new vehicles for automated driving and parking, starting in the first half of 2023.

EU this week introduced the Digital Markets Act, a new set of rules to ensure fair competition and more choice for users. As The Verge explained, DMA could be the EU’s “biggest ever legislative effort to balance competition in the tech world.” Tech giants like Google, Meta, and Apple will be under stricter scrutiny. For example, the largest messaging services (such as Whatsapp, Facebook Messenger, or iMessage) will have to open up and interoperate with smaller messaging platforms, if they so request. The financial penalties are steep as “if a gatekeeper does not comply with the rules, the Commission can impose fines of up to 10% of its total worldwide turnover in the preceding financial year, and 20% in case of repeated infringements.”