Via the AGI list from the New Yorker:
Before Trump took office, the Chinese government was far outspending the U.S. in the development of the types of artificial intelligence with benefits for espionage and security. According to In-Q-Tel, an investment arm of the United States intelligence community, the U.S. government spent an estimated $1.2 billion on unclassified A.I. programs in 2016. The Chinese government, in its current five-year plan, has committed a hundred and fifty billion dollars to A.I.
So the Chinese government is going to spend $30 billion dollars a year on AI research in the next Five Year Plan. This may or may include the amount to be spent by large Chinese companies such as Baidu, Alibaba, Huawei, ZTE or Chinese signals intelligence agencies. In a response to the original AGI list post, someone points out that there is a lot of company research in the US. But again there might be in China. And there are a lot of AI professors in public and private universities in the US. But these academics are dependent on grants from the NSF, the DoD and other bodies even if there salaries are paid for the states and from student fees. These days a British academic wouldn't last long in they weren't generating research grant income. Perhaps things are different in the US if you have got tenure and you really can just sit around all day thinking great thoughts, although you probably aren't going to have many PhD students or postdocs.
So, it's very hard to compare the efforts of the two countries directly. A(G)I is a broad subject and there are a lot of rabbit holes to go down. I'd guess in the US, $1 million gets you 2-3 FTE researchers, which is 2000-3000 per $1 billion. You might get mote in China, I suspect. So, $30 billion/year would get you 100,000, perhaps twice than in China. But you are going to need to train a lot of those people from scratch. Which means huge numbers of undergraduate, master's and PhD students as well as postdocs. It takes a long time to train people, but if you already have a trained engineer/scientists, you can stick them through a master's degree in a year and then, as desired, a PhD, so you can certainly have a large cadre of people in five years. Quite possibly a lot more people than the US.
Of course, whether this leads to A(G)I depends on all sorts of factors. If you have lots of researchers, you can let a thousand philosophies bloom, so rather than push everyone down the currently fashionable avenue (supervised deep learning, say), you attack many problems from many sides and go back and look for ideas from the existing literature to revisit. As with any kind of R&D spending, there's the question of how much is actually D. Lots of AI R&D can end up being people doing traditional ML stuff like linear regression on whatever dataset they have. With big data, there's a lot of exploratory data science that can de done.
In the US, there's still relatively little funding going directly in AGI research. That might be starting to change, but it might still be very much a minority area of interest. are the Chinese any different. Are they developing an AGI roadmap and then systematically implementing it? They could have hundreds of PhD students trying to crack StarCraft (or Civilization), but will they? It could be that we are on the cusp of a SpaceRace-like AGI Race or this could just be lots of money either being squirted away or going on making better fraud detection and recommender systems. And if there is to be an AGI Race, do we need a Sputnik 1 or has that already happened?
Add comment
Comments