TiB 149: Who will own the AI future; building a more innovative society; what top ML talent wants; and more...

Welcome new readers! Forwarded this email? Subscribe here. Enjoy this email? Forward it to a friend.

The latest TiB podcast episode is a conversation with the brilliant Jade Leung on AI governance. Jade’s doctoral thesis, “Who Will Govern AI?” was one of the best things I read in 2020. Jade looks at the power struggles between governments, corporations and researchers that accompanied previous general purpose technologies (“GPTs”) and asks what this can tell us about the future of AI.

Jade’s work has caused me to rethink my ideas about a topic we’ve discussed a lot in TiB: the political power of talent in AI. I’ve talked before (see, e.g., TiB 80) about how the ethics and worldview of the elite research community are shaping the behaviour of big companies. Jade argues, however, that we shouldn’t expect this to persist: it’s a common theme in the early stages of strategically important GPTs, but soon gives way to fragmentation and state dominance (“States win in the end” is one of the main reasons I expressed skepticism last week about the ascendance open protocols)

We cover a lot of other topics, including:

I highly recommend this conversation. Jade is an extraordinary talent.

Historian Anton Howes has a great post on how we can increase the rate of innovation in society (His related book on the history of the Royal Society of Arts is superb). The core argument is that most policy designed to stimulate innovation focuses on the (relatively) tiny number of people who actually become innovators - but the big win is upstream from that: getting many more people to innovate in the first place.

Anton makes the important point that necessity is not, empirically, the mother of invention. Innovation is not instinctive. The idea of innovation - or the “improving mentality” - first has to get into the heads of potential innovators. (We’ve talked about the importance of the improving mentality before, largely in the context of the work of Dan Wang - see, e.g., TiB 28, 92 and 138). 

I’m obviously sympathetic to this point: my whole career is premised on the idea that the world is missing out on its best founders precisely because starting things is not natural. So how do we spread the “improving mentality”? Anton admits it’s not straightforward. His best answer is to work to increase the social status of invention, perhaps through popular culture (The post has some great movie recommendations). That seems right to me. See, for example, this new paper on the (very) long run effects of stories: cultures where risk taking is rewarded in folklore are more entrepreneurial. Innovators, get storytelling! 

A key theme in my conversations both with Jade (see above) and with Ian Hogarth in the first TiB podcast episode is the importance of global competition for machine learning talent in this (presumably) early phase of AI’s development. What can states do to attract the top talent needed to bootstrap a strategic capability in AI? An interesting new paper by scholars at Perry House and the Future of Humanity Institute looks at the immigration preferences of around 500 top AI researchers to try to answer this question.

The results are encouraging for the US and, to a lesser extent, the UK. Nearly 60 percent of top researchers not based in the US say they think there is a good chance they would move there in the near future - compared to 35 percent for the UK and just 10 percent for China. Broadly, in terms of intrinsic attractiveness, the US is well positioned to be the destination of choice for the world’s leading AI talent, with China having to rely on its - obviously large - homegrown talent pool (though the sample was rather underweight Chinese researchers).

The biggest barrier facing both the US and the UK is their own immigration policies. Visa and immigration concerns were the most cited reason for reluctance to stay in or move to the US - and higher than for any other country. Meanwhile, the UK has the lowest proportion of foreign researchers who say they are likely to stay put for at least the next three years. Given the stakes and the small numbers involved, making immigration hassle-free for PhDs in machine learning is about close to a free lunch in technology policy as I can imagine.

I’m grateful to have such a great set of readers. If you enjoy TiB, please share it on social media or forward this email to someone who might like it too. Podcast reviews also extremely welcome.

Feel free to email me or find me on Twitter if you have any feedback or questions.

Until next week,

Matt Clifford