What the Death of A Chinese AI Creator Reveals About American Capitalism's Limitations With Big Data
Why you'll likely need a license for government data and AI
The sudden death of Tang Xiaoou, the fifty-five year old founder of SenseTime, occasions more than a little introspection about the U.S.-China relationship, in light of the thawing between the world’s first and second economies.
SenseTime is the world’s number one most successful facial recognition app. (Clearview, which I cofounded, is number 2, according to the Commerce Department’s National Institute for Standards and Technology.)
Excellence deserves applause and excellence, at least here, was Made in China. When we were setting up Clearview we read quite a number of papers about Chinese facial recognition.
China, you see, has a lot of people and their state can readily take that data and press it into the service of the state. (Clearview had to be a bit more clever and scrape the Internet to build its own database.)
This power of the Chinese population is something which cannot be ignored in a geopolitical context. Quantity has a quality all its own.
As with all companies tied into the Chinese intelligence state there appears to have been more than met the eye.
There are reports about misstated revenue.
And the U.S. Department of the Treasury rightly noted the ghastly ways in which SenseTime had been deployed on the Uighur people. If you think the Turkic peoples don’t notice this, you have another thing coming.
And there appears to be a coordinated strike on SenseTime over its alleged inflated revenue.
All of that will likely be sorted out in time but it is interesting to see how far SenseTime has come in a relatively short period.
To be sure there’s an alternate history in which America, relying on the CIA grants, beat China to facial recognition development. That didn't happen but perhaps we can take some solace in Tang Xiaoou’s education — University of Rochester for undergrad and MIT for grad — and what that means.
Here’s the Washington Post discussing the differences in January 2018:
More than anything else, experts say, deep learning technologies need huge amounts of data to come up with accurate algorithms. China has more data than anywhere else in the world and fewer constraints about mining it from its citizens.
"Now we are purely data driven," said Xu Li, CEO of SenseTime. "It's easier in China to collect sufficient training data. If we want to do new innovations, China will have advantages in data collection in a legal way."
Add in Chinese hacking and you can see why China is poised to be quite powerful at AI in the 21st century, especially when Chinese-backed genomics companies like 23andMe or Chinese-backed social media companies are decidedly lackadaisical about security.
America capitalism has essentially two forms — royal charters and the free market. You can think of the state has decided with the mob which form should operate in where in a kind of informal regulatory dance. You can think of Netflix killing mobbed up Hollywood or Uber crushing mobbed up taxi world.
The free market is more or less well understood. This is why we’ve had such high antitrust enforcement as of late. Competitive markets reduce prices for the public.
But when you’re trying to do something very big, you need a government. We don’t call it royal charter anymore but that’s essentially what happens when NASA doles out billions to SpaceX and what doesn't happen when the FCC denies $900M to SpaceX.
Royal charters will likely give way to AI licenses. Courts are going to sort what material from the public domain is useable and what’s not. My guess is that they’ll draw a distention between AI that enhances public safety like Clearview.AI and AI which cannibalizes the work of artists or creatives like Open AI. We might call this Deep State AI versus Mob AI.
Want to work on AI? Get a license.
Here’s how I suspect the future licensing will go:
All of government will be hosted in either Google Cloud, Amazon Web Services or a newly cleaned up Microsoft. This’ll be the subject of another post but I suspect that’ll happen over the next year or so with renewed focus by the British and American intelligence services.
That massive government data — and organizing it — will be mighty tempting for startups. “Data centers are expected to lead growth adding close to 200,000 GWhs of new energy consumption by the end of the decade. Even if the entire industry met the Microsoft and Google 24/7 clean energy commitment by 2030, we’d still need 22GW of new power,” notes Jigar Shah, director of the Department of Energy’s Loan Programs Office. Avoiding wasting that compute power will be a serious issue, especially as the grid has proven notoriously unreliable.
Startups will apply to get, say access to the Department of Interior data, and have to go through a thorough vetting.
Startups that do well with this data will be readied for an acquisition. In this way, the nexus of power between the private empires (Microsoft, Apple, Amazon, and Google) and the American empire will be complete.
We might have a different fate, old friend.