ChinAI #226: Regulating AI won't set U.S. back in competing with China
Plus, part 2 of AI birdwatching
Greetings from a world where…
the best thing I read this week: a scathing review of a Pablo Picasso exhibit at the Brooklyn museum
…As always, the searchable archive of all past issues is here. Please please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay support access for all AND compensation for awesome ChinAI contributors).
ChinAI Links (One to Open)
Must-read: The Illusion of China’s AI Prowess — Regulating AI Will Not Set America Back in the Technology Race (Foreign Affairs)
Last week, along with co-authors Helen Toner and Jenny Xiao, I published a Foreign Affairs essay that picks apart claims that the U.S. shouldn’t regulate AI because China would race ahead. Big props to Helen Toner for taking the lead on this piece and distilling our arguments in a summary thread:
This essay draws on evidence from a GovAI report by Jenny and me on recent trends in China’s large language model landscape, published last month.
Feature Translation: Counting Birds (part 2)
Context: Okay, let’s pull ourselves away from the vortex of great power competition in AI and get back in the world of counting birds! Last week, we learned about the promise of computer vision for identifying migratory birds in China, from a longform jiqizhineng article (link to original Chinese). In the second half of the translation, the author digs into the wetland weeds of why the road ahead is long and arduous.
Before I get into the details, I want to highlight a debate I’ve been having with Miles Brundage, OpenAI’s head of policy research, on the pace of general-purpose technology diffusion. Miles contends:
“General purpose technologies take decades to diffuse” as a basis for a relaxed view of AI’s economic impacts doesn’t sufficiently take into account the Internet as diffusion mechanism, the speed in diffusion generally in recent decades, and *gestures broadly* what’s happening”
I disagree. So, let’s see “what’s happening” in the spread of AI to species identification.
Key Takeaways: For each particular application of AI — say, bird identification — you need customized data, which can be very hard to collect and clean.
Consider the red-crowned crane. At birth, it is gray and ugly. As it flies south from Heilongjiang province to Jiangsu province, its feathers gradually turn white, and the distinctive red spots emerge on the top of its head. To get effective data for identifying these cranes, “phased training is done every 3-4 months, and some small adjustments are made to the system continuously, so that the system can fully recognize the whole appearance of red-crowned cranes from juvenile to mature.” As the article states, “It can take a year for each new species added to the database.”
Camera set-up also complicates data collection. If there are issues with the camera angle, you can’t capture the key features of oriental storks or red-crowned cranes. “It is difficult to take a clear picture when observing a palm-sized spoon-billed sandpiper from a distance of 300 to 500 meters.”
General-purpose technologies demand infrastructure investments, complementary innovations, and organizational adaptations. All of these take time. Specific examples from bird identification illustrate each of these three features:
Many protected reserves are located in remote mountains and forests, without access to electricity and internet. Obviously, this slows the implementation of AI solutions.
The high costs of infrared wildlife cameras limited the ability to collect animal data. Innovations in camera technology are necessary to reduce the costs of these cameras.
An older generation of conservation personnel are hesitant to use AI in species identification.
In sum, companies like Chuangshi Zhineng have been trying to deploy AI in this domain for the past six years, and they have not made much progress. The company’s marketing director estimates that the current diffusion of AI in wildlife protection is “not even 2 or 3 percent.”
I’m very willing to concede that I am constantly just trying to catch-up with all the new advances in AI. At the same time, it can be very easy, especially if one is fully immersed in the inside world of all the exciting things happening on the technical front of AI, to underestimate the timeframe of diffusion. Sometimes, it’s helpful to just take one specific application sector like AI birdwatching and take stock.
More details in FULL TRANSLATION: Counting Birds
Thank you for reading and engaging.
These are Jeff Ding's (sometimes) weekly translations of Chinese-language musings on AI and related topics. Jeff is an Assistant Professor of Political Science at George Washington University.
Check out the archive of all past issues here & please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay for a subscription will support access for all).
Any suggestions or feedback? Let me know at chinainewsletter@gmail.com or on Twitter at @jjding99
I also enjoyed the Foreign Affairs article. One other key reason China won't win the AI race: their systems aren't open with their people. In other words, the feedback loop won't work as designed due to CCP restrictions.