I am a fan of the Economist magazine, so much so that I recently subscribed to receive print copies. My first issue arrived yesterday, and I was shocked to see that they have also just published an article about China and Artificial Intelligence (AI). Much to my surprise, they disagreed with my post on a very similar subject, and the difference pretty much starts at the top, with a title of “China may match or beat America in AI”.
And its not just the title. My thesis is that trade policies which protect domestic industries come at a price, and the price in the case of software may be self-defeating as it limits the ability of domestic AI companies to build large data sets. By contrast, the Economist asserts that China can ‘win’ in AI because it already has such large data sets. So, yeah, there is a difference of opinions here.
Not afraid of a fight with capitalism’s preeminent proponent, I read the article to see how we ended up on different sides. This is doubly so, because my post offers a defense of ‘protectionist’ trade policies in most cases, which contradicts a core editorial principle of the magazine. So we seem to be disagreeing where we would normally agree.
As with most things nowadays, once you get past the headline and read the details, their position becomes more nuanced. First, they acknowledge that one of the reasons China is leading the way in AI is because it has immense support from the government. AI clearly has the potential to disrupt large swathes of the global economy, and many people agree that robotics (a subset of AI, sortof) poses a huge threat to China’s labor-intensive economy. I agree that this is a big priority in China’s policy circles, but I am not sure that this support is going to help the same way it has in other industries.
Second, the Economist points out that China is turning out a huge number of professionals trained with AI. At least, that is the plan. The article concedes that “Finding top-notch AI experts [today] is harder in China than in America”. No one questions China’s ability to throw huge numbers of trained professionals at a problem, but that has not yet taken hold. There are AI experts all over the world, and in the global race for talent the US has many advantages.
But the crux of their argument is that China has the world’s largest online audience with more Internet users than most of the rest of the world combined. And, they argue, that audience will allow domestic companies to build the biggest datasets.
My first reaction to that was to notice that it confirms a basic tenant of my thesis – the competition in AI is going to center (for now) on owning the largest datasets.
In my post, I used autonomous driving as an example. Companies like Uber and Lyft have the ability to (someday) add sensors to their cars and collect the largest datasets for real-world driving. Their competitors in China are (for now) limited to a smaller (but still large) data set of China drivers only. The Economist is assuming that the datasets that matter most will be collected from data that is currently collected online. I would argue that online datasets are just one piece of the data universe, and data collected in the offline world are both larger in number and over time more important.
Their argument about China’s online population is valid. For many AI applications this will be an important audience. For instance, e-commerce and online banking offer a ripe ground for AI-based improvements (fraud detection, lead generation, etc.). But there are going to be many, many more offline applications for AI. It is also unclear if the data collected from observing Chinese consumers will offer a model that fits to consumers into other retail markets, dulling the advantage of accessing that market.
Boiling it all down, our difference really comes down to the importance of this large, China online audience. My view is that this advantage will only offer a small benefit as AI models become widespread in offline industries. I have learned over the years to never underestimate the dynamism of China’s economy, but I continue to believe that in this area, traditional policy mechanisms are going to face a very different economic reality which will be heavily shifted by software realities.