Matt's Thoughts In Between - Issue #11

This week: is AI investment worth it, why wage stagnation begets Trump, and how central planning triumphed in the end…

The New York Times ran an article this week looking at the extraordinary salaries earned by researchers in artificial intelligence. It mentions the $1.9m paid to the top researcher at OpenAI and the $345,000 average cost per employee at DeepMind. 

People are expensive in AI - and the costs don’t end there. There are now 19+ companies designing their own chips for deep learning acceleration, according to James Wang. Barrons has a piece looking at some of the contenders in this space. It’s clear it’s a race that is going to consume enormous sums of capital in the next few years.

So why might it be worth it? Even if you dismiss the likelihood of general intelligence, what’s already happening is extraordinary - not least in software engineering itself. This tweet mentions that Google used machine learning to reduce the lines of code in Google Translate from 500,000 to 500 - i.e. the software “learned” how to write more efficient code. This is just one instance, but it will happen everywhere. Jeff Dean (who leads Google’s AI work) gave an amazing presentation on this topic late last year at NIPS (one of the top AI conferences). Software that can write software will change everything; it’s worth spending a lot to invent it.

Many people have argued that the stagnation of median income in rich countries is one of the primary causes of Trump and Brexit. Branko Milanovich, creator of the famous Elephant Chart, certainly believes so. 

But this seems puzzling. Politics is a two-dimensional competition: parties compete on left-right economic issues and “open-closed” cultural issues. So if populism is a response to economic stagnation, why do its policies seem so much about culture? Dani Rodrik published his answer this week: the centre-left chose to make cultural issues the main political battleground and failed to develop any real economic policy alternatives. (Rodrik links to Thomas Piketty making a similar point about liberal “Brahmins” competing with conservative “merchants”)

This seems to me the critical question for advocates of a new centrist party in the UK and, indeed, for the Democrats in 2020: what are those alternatives? Rodrik points to the challenge that must be solved first. There’s evidence that low trust in government (and it has fallen consistently) makes voters - even those that are worried about inequality - skeptical of anti-poverty policies. A centrist platform without a plan to reverse that will struggle to make headway.

EconTalk, one of the most consistently interesting podcasts, has a great interview with the philosopher Elizabeth Anderson on her new book, Private Government: How Employers Run Our Lives. Anderson argues that libertarians should worry much more about corporate - and especially employer - power than they do. Interestingly, this is a major motivator for many people to get into startups. I listened to the podcast the day after I stumbled across this (startup-focused) post on Why We Hate Working for Big Companies. 

It feels like there’s something in this. One of the blogs I’ve been reading longest, over ten years, is Chris Dillow’s Stumbling and Mumbling. One of Dillow’s favourite (provocative) arguments is that the Soviets actually won the Cold War - how else can you explain the triumph of central planning that exists inside almost every corporation? (More here and here). 

Even if you don’t want to go that far, it does seem as though there’s something broken in hierarchically run companies that is a little less broken in startups. At the very least, Anderson notes in a fascinating historical segment (start at 41:44), before hierarchical production took over in the 19th century, bosses and workers used to go out for drinks together a lot more often than they do today. Not a high bar, but perhaps a starting point for today’s entrepreneurs…

If you like this, please share it. The easiest way is to tweet in two clicks here or to forward to a friend or two. Hit reply if you have articles you’d like to share.

Until next week,

Matt