I’ve been reading AI Superpowers by Kai-Fu Lee and some of the points he raises are worth debating.
Firstly, let me say that this is a good book and I am enjoying it. Kai-Fu Lee is a uniquely credible commentator on how a China vs. United States competition to dominate in the AI space might play out. Unique because he has spent most of his career in the States but, for the last decade or so, firmly embedded in China.
He raises two points:
(1) We are now in the Age of Implementation and the Age of Discovery (which we have been in for the last 400 years) is now less relevant to economic success. His assertion for this is that it is the speed and scale at which you can implement AI that is now critical to competitive success. He claims that we are now in this age courtesy of the success of machine learning techniques such as deep learning. What is now less relevant for competitive success, he claims, are those attributes which were born in an Age of Discovery and that, traditionally, the West have dominated with the focus on the discovery of new knowledge and the commercialisation of inventions. China appears to be better placed to succeed in the Age of Implementation vis-a-vis the rest of the world (including the United States).
Here we get to one of the premises of his book – there are only two AI superpowers: China and the United States. They are the only AI superpowers by virtue of their inherent characteristics such as geography (literally the size of their landmass allowing large populations to exist to generate huge markets and, of course, huge amounts of data), entrepreneurial culture, and access to the technology necessary to engineer and deploy machine learning and AI products and services at a massive scale.
(I have not yet explored his claim here and, for now, I’m assuming that this isn’t a crazy assumption. What is interesting is that the AI readiness index 2020 report places United States in 1st place and China in 18th place. However, as the authors themselves admit, their research evaluates readiness for AI, not the scale of implementation in which, arguably, China would score significantly higher)
(2) We are now in the Age of Data and we are leaving behind the Age of Expertise. Lee believes that as long as you have access to data, we now have the capabilities (such as machine learning), to quickly learn to get close to, if not surpass, the performance of human experts. He cites the well known example of a deep learning algorithm classifying breast cancer scans. What he doesn’t mention, however, is that – at the moment – the best performing algorithm is 92% accurate. A human expert is about 94% accurate. And that a deep learning algorithm + human expert is 99% accurate. For these kinds of applications near perfect performance is essential. So, the equation ‘more data + machine learning + compute’ does not always result in better performance than a human. I am a strong supporter in the idea that we should engineer AI to complement human capabilities, not supplant them, which is why I love examples like this.
I am also reminded of Centaur AI teams in international chess competitions. As Garry Kasparov describes in his book ‘Deep Thinking’ the performance of a two amateur humans players with two laptops team saw off the competition which included a supercomputer+grandmaster team.
For the foreseeable future, and even when we develop responsible AI, we will still want humans in the loop. Especially as the criticality of a decision increases. (A talk by IBM at NeurIPS 2020 last month, December 2020, made a similar point).
I think we can see where Lee is going here. He has deftly set out two arguments against why the traditional strengths of the Western world – discovery, expertise – are now less relevant. But I think this is somewhat disingenuous and, what is more, he doesn’t consider how the West might respond.
Let’s use one example to illustrate this. One of Lee’s claims as to why China will out-compete the rest of the world in an Age of Implementation is that Chinese culture (a) relentlessly embraces the ideas of others, copies them and improves on them thus focusing on the application of knowledge rather than the discovery of new knowledge, and (b) has a ferociously competitive domestic market with very few competitors left standing and those that are have discovered a defensible market position (so a very Darwinian view on the Chinese economy). Think of it as an accelerated evolution of a dominant design. And the driving force behind both of these? A culture that is relentlessly focused on being busy and industrious.
Now, I must admit, I don’t know a great deal about Chinese culture but I suspect that this is a fairly accurate description. Whilst the Chinese have shaped themselves into a phenomenal and extremely hard working society, the Western world is moving away from such cultural norms. Nowadays, hard work isn’t seen as the path to success. Personal satisfaction and having a more balanced lifestyle are the new harbingers of Western society now. We need look no further than the experiments in Scandinavian countries for 6-hr days and the 4 day working week (Scandinavian countries are always a good barometer for what is coming down the line for other countries in the Western hemisphere).
As I say, it is hard to dispute that – on effort alone – China is already winning. However, Kai-Fu Lee is writing a book about AI and yet he doesn’t appear to have considered how a more automated world will mean that human effort is not only less necessary, but actually will not be able to keep up. What use is it to have a nation of hard-working, ‘gladiatorial entrepreneurs’ if AI can out-code, out-innovate, and out-experiment them? Those countries that can automate the entrepreneurial process from inception of an idea, to trialling a multitude of different products/services in parallel, and rapidly refining and re-deploying new variants on a steadily improving trajectory, will be those that will win. We don’t have such technology at the moment, and deep learning alone won’t cut it, and so we will be reliant on the discovery of new AI capabilities which we will require elite expertise to develop. Those same two competencies that Lee claims are now less important!
(Microsoft’s ‘low code and no code’ power platform is a good example of where this may be happening already.)
I’ve not yet finished Lee’s book and its certainly thought provoking. We are just at the beginning of an AI revolution that has been ignited by deep learning. But our future AI capabilities are likely to take us well beyond a reliance on a technique that was first developed in the 1980’s. Don’t get me wrong this is an important book and I’ve encouraged my mentees, colleagues and clients to read it.
And, on a final note, I’ve just started to read Ian Hogarth’s excellent blog post on ‘AI Nationalism’ which dovetails quite nicely with Kai-Fu Lee’s thinking. I’ll have more to say on that too.