Lore

Robots Weekly ?: Behind the Bamboo Curtain ??

Posted on 12/18/2018 by Kyle Mensing

Last time on Robots Weekly I talked about all the fun (terrifying) stuff China is doing with AI. This week let’s look at what China dreams of accomplishing and some other fun things happening there at the moment. ?

China for World Leader 2030 ?️

China is putting it all out there when it comes to AI. The government announced last year that they want the country to be the world leader in AI by 2030. So, you know, small goals. ?

While there isn’t a clear picture on what the government plans to spend to make this plan a reality, city governments are already pledging billions (with a b) of dollars to fund AI research and development. ???

How do they envision getting from the present to their dream of ~~world domination~~ AI leadership? ?

By 2020: be solidly in the mix as one of the top countries in the field ?
by 2025: make major breakthroughs and lead the field in at least some areas ?
By 2030: claim the crown and do a victory dance ?

By many measures this plan is humming along nicely as Chinese companies and organizations have produced more papers and patents than their contemporaries. Granted, these are fuzzy metrics since content farm type entities can probably churn them out. However, Chinese AI startups lead the world in funding. This metric seems a little more solid, but I could see how it might be manipulated (see: amount of money being invested by city government’s). ?

Still, China has big goals for AI and they are hustling to make them happen. ?

Side note: that article does a nice job covering some facets of AI at a high level

AI: brought to you by Humans ?

“I used to think the machines are geniuses,” Ms. Hou, 24, said. “Now I know we’re the reason for their genius.”

China’s biggest advantage in the AI space race is a huge labor pool. This fact is behind many of the country’s advances across the board. They can simply throw bodies at a problem until the tipping point is reached. Access to labeled data is the biggest barrier to entry in machine/deep learning right now, a problem that can be solved by throwing humans at it. Therefore, using the transitive property, China = lots of labeled data. ?

Another advantage is a society used to communal life and government ~~control~~ oversight as opposed to privacy. So you get attitudes like the one reflected in the below quote. “For now” is never a good strategy. ?

“Cameras make me feel safe,” she said. “We’re in control of the machines for now.”

Must Be Patriots ?

The Beijing Institute of Technology has launched a track focused on the development of intelligent weapons and admitted the first class that will specialize in the field. So, yikes. ?

This marks the first university program designed to train students to become experts in this field. China isn’t making any secret of their desires to be a military AI power. ?

I’m really confused about the stance they appear to be taking in a position paper submitted to the UN:

“As products of emerging high technologies, development and use of lethal autonomous weapons systems would reduce the threshold of war, and the cost of warfare on the part of the user countries. This would make it easier and more frequent for wars to break out,” Beijing said, appealing for more discussions.

“Until such discussions have been had, there should not be any preset premises or prejudged outcome which may impede the development of AI technology,” it said.

So, intelligent weapons could make wars more common and easier to start so we need to discuss it but we shouldn’t stop development and wait for those discussions? I feel like that’s saying “yeah, we know there are a lot of potential problems with this, but we’re going to go ahead and develop as much of this tech as possible until the sector is regulated or monitored.” ?

Google Who? ?

Remember Google Duplex? It’s finally moving from testing to deployment, in limited stages. Heard of Alibaba? They have a similar service that “is already servicing millions of customer requests a day.” ?

There are 3 features of the Alibaba assistant that are especially impressive:

Interruption – the bot appears to handle interruption pretty seamlessly (I say “appears” because I only have 1 transcript to go off of)
Nonlinear conversation – like interruption, the bot handles a conversation diverging from the script very well (Google chose the fields it did for Duplex because the conversations are typically pretty structured, a.k.a. linear)
Implicit intent – the bot can correctly interpret indirect answers and keep things moving

Basically this bot appears to handle conversations more like humans than other voice assistants. So while we in the West look at companies like Google, Microsoft, and Amazon as being on the forefront of AI, they may already be behind their Chinese counterparts. ?

TL;DR:

What are China’s biggest advantages when it comes to developing and deploying AI systems?

a ton of data
massive government support
general cultural acceptance of things that would raise privacy concerns elsewhere
a huge labor pool to handle the manual aspects of the systems (data tagging, etc.)
close ties between companies and government for low regulatory friction