Thoughts in Between

by Matt Clifford

Matt's Thoughts In Between - Issue #44

Forwarded this email? Subscribe here. Enjoy this email? Forward it to a friend.

China and the future of universities

We talked recently about social mobility in China and separately about why there are so few computer science grads (in the US and UK), despite the rewards attached. This week the indispensable Jeff Ding has a fascinating translation of a Chinese essay in the style of advice to an ambitious young Chinese person that ties these threads together.

The author, Saidong, is convinced that “computer science is still the major in which it’s easiest for the working class to improve their situation”. Most interesting, though, are his thoughts on how ideas of prestige are changing as China's economy and society are transformed. In particular, Saidong argues that the economic vibrancy of the city in which you study is now more important than the history of the specific university:

Shenzhen and Suzhou, which do not have much accumulated history of higher education, are rapidly turning economic advantages into educational and scientific research advantages

It’s interesting that this didn’t happen in the UK or US. Oxford, Cambridge (both UK and MA), New Haven, etc, are all small economies (though arguably in the orbit of larger metropolises). One driver Saidong points to is that the sheer weight of numbers of Chinese PhDs educated in the top Chinese and/or Western universities means that even new institutions can accumulate a critical mass of top scholars. Prestige matters a lot in academia - and technology tends to warp what we see as prestigious. China is here, as so often, an interesting glimpse of the future.

How to teach AI human values?

DeepMind has a new paper on AI safety. AI safety - ensuring that artificial intelligence does not cause harm to humans - is a controversial topic, particularly the claim that, left unchecked, AI poses a potential existential risk to humanity. Lots of smart people think that’s nonsense - but equally lots of very smart people think it’s the most important topic in the world. Nick Bostrom’s book Superintelligence and this short introduction are good starting points for learning more.

A key topic in AI safety, and DeepMind's paper, is alignment. How can we ensure that powerful AIs have similar values to humans (and so don’t end up innocently destroying the world when asked to make as many paperclips as possible)? DeepMind has made astonishing progress in AI through reinforcement learning - i.e. where the AI has a clear goal (or reward function), such as the score in an Atari game, and learns over many iterations how to maximise that score. Their proposed safety approach echoes this: have the AI learn its own reward function (i.e. values) by interacting with humans, at the same time as learning how to maximise that function. 

This raises interesting possibilities. DeepMind argues that applied recursively this technique allows us to train AIs to assist humans in evaluating other AIs, which should allow the technique to scale to superintelligent systems. There’s more commentary in Rohin Shah's excellent Alignment newsletter and Jack Clark’s must-read Import AI. Clark argues that this approach could allow us to scale the number and diversity of people involved in shaping AI, by providing a straightforward way for non-technical people to shape AI values. Perhaps this is the path to AI with distinctively national - or even international - characteristics

The missed opportunity of pre-Brexit foreign policy

With Theresa May calling off today’s planned vote the Brexit chaos continues, as this excellent graphic shows. I'm intrigued by the evolving views of moderate Leave voters - those who are strongly Eurosceptic, but deeply disillusioned with the actual process of Brexit. Roland Smith calls this (small-ish?) group “Alt Remainers” and cites this post (“I’m a Leaver who would be happy for a second referendum”) by Ed West as a pragmatic statement of the argument. 

This seems particularly interesting now, given the rise of Eurosceptic - or at least "ever-closer-union"-skeptic - voices among European governments. The Economist has a piece on the “new Hanseatic League” of Northern European countries who want to see Eurozone reform from inside the EU. Without Brexit, the UK would be the natural leader of such a group (though Brexit also prompted them to form).

The UK’s failure to form and lead such a group pre-2016 was arguably its great strategic mistake of post-Iraq foreign policy. This David Goodhart essay from 2014 is remarkable to read from a 2018 perspective. Goodhart argues that the UK under Blair pushed for ever greater Eurozone integration because it believed that an integrated Europe was in its economic interests - but at great long term strategic cost. Perhaps unsurprisingly, this echoes the mistake Remainers have made in pleading their (my!) case: imagining the economic argument trumps all else. What are the consequences? No one knows, but we’re about to find out...

Quick Links

  1. Future perfect. What are the most compelling ideas in science fiction? (Actually, here's one in real life)
  2. You are what you speak. Great thread on personality and language use.
  3. It's not black and white. Stunning colour image of Pearl Harbor under attack.
  4. Importing inventors. Which countries have most immigrant (and emigrant) inventors? And which countries want more immigrants?
  5. Gut punch? Fascinating study mapping moral impulses onto physical sensations. (Click to see striking example of how liberals and conservatives differ)

Your feedback

Thanks for reading. I love getting your comments and feedback, so do hit reply if you have thoughts. Equally, I love getting new readers, so if you know someone who might enjoy TiB, just forward this on.

Until next week,

Matt