Tuesday, June 13, 2017

Hayek was wrong about central planning

Friedrich Hayek is known for making this argument against a centrally planned economy:
When Hayek explains the obstacle to effective central planning, his claim is not merely that information is widely dispersed and therefore hard to acquire. Rather, it is impossible to acquire (Hayek 1973, 51). When prices are set periodically by a central planner, rather than instantaneously by consumers and producers who are the first and typically the only people to have that information in reliable and timely form, prices inevitably carry less reliable, less timely information. ...

A central planner could have the world’s most powerful computer, beyond anything imagined when Hayek published “Use of Knowledge” in 1945. No computer, however, could solve the problem that Hayek was trying to articulate. The problem is not lack of processing power so much as a lack of access to the information in the first place. ...

Although computers cannot solve the problem, Hayek thought radically dispersed decision making by buyers and sellers can and does solve the problem, so far as it can be solved. Sellers who charge too much end up without customers; they learn to be more efficient or else go out of business.
I never found this argument convincing. Nowhere does he estimate or compare the computational resources needed.

Sure, it was obvious in 1945 that computers of the day could not do it. He may not have been aware of any computers at all. But what about today?

If radically dispersed decision making by buyers and sellers can solve the problem, then surely computers can if enuf info is digitized. Amazon could soon understand the buying habits of consumers better than they understand them themselves, if it has not already.

Hayek would have said that Amazon is impossible, as Amazon is essentially an empire that is bigger and more complicated that the centrally planned economies that he contemplated.

Amazon seems to be more efficient than the markets it is replacing. Isn't it a counterexample to Hayek? Wouldn't Hayek have said that dispersed independent bookstores would be more efficient than Amazon?

There are other arguments for the freedoms that Hayek advocated, but I am not buying this one.

I had this exchange with a Hayek buff:
You should send this to Russ Roberts of Econ Talk. He is a big Hayekian and is very interested in this type of discussion. You've got some credentials... get on the show!
How can he be a disciple of a guy who wrote in 1945 about the limits of computers, and there aren't even any numbers or formulas in the analysis?
Well, he would argue that you are never going to be able to run the whole economy that way. There is an infinity of markets, prices and preferences, always changing. I' m serious. Russ is a good guy, I've met him. Put your argument in a concise form and get a dialogue going. He loves this kind of stuff.
This is sometimes an argument against AI: a computer can never make an optimal choice among an infinity of possibilities.

First, it doesn't have to make an optimal choice. It only has to outperform humans.

In 1945, these arguments said that computers would never play chess.

Look at Amazon today. I don't have any hard data, but I bet its complexity is already greater than that required to manage the whole economy of more than half the countries of the world.

I log into Amazon, and its seems to already know what I want to buy and what I am willing to pay. In some cases it is better, as it recommends a product that I would not find on my own. And it gets better all the time. It may soon be the case that its predictions are better than my own
shopping skills.

If you look at its distribution system, where it can buy products in China or wherever, ship them to warehouses in my state, and deliver them to me on a 2-day order. That is already better than what humans can do. Only a centrally managed computer system can do that.

I can't keep up with a serious Hayekian tho. He'll probably point to the failure of managed economies like N. Korea. But there has never been one with the ruthless efficiency of Wal-mart or Amazon.
Well, there is no doubt that computers have come farther, faster, than most people predicted. Kasparov has interesting things to say to Harris on this, and points out that a man-machine blend will beat the strongest computer. But ultimately, you have to bet on the pure machines (I don't think Kasparov does, yet).
Kasparov is engaging in wishful thinking. As I understand it, the best chess computers are now rated above 3200. Kasparov at his best was about 2800, while the average grandmaster is 2400 and the average expert is about 2000. So the computers will beat him every time, and man is no help to a man-machine blend.
Amazon and Walmart are a little different story. In principle, they are like huge supermarkets (which were themselves a tremendous innovation). They know what you like, and what you might like (like a good store clerk who has waited on you before). They also have a very efficient delivery service. They are outstanding at reacting to local information, like shipping out salt when there is a sudden outbreak of snowstorms. As a book addict, it is truly magical how Amazon has "gathered" all the used books out there and made them available for about $5 each. But to manage an entire economy... well, that is very different. There is so much local knowledge that would have to be constantly gathered, updated, collated, predicted-upon, reacted-to.. and it is a lot more complicated than what Amazon is doing, because you have the whole logistics and manufacturing chain behind it, not to mention agriculture, which is dependent on weather... Layer on top of that finance, investment, capital markets... Anyway, we have a history of being scared by bigness. And of course the biggest problem of all (forget the system becoming to vulnerable to hacks/viruses, etc.) is the control of the system. Big government anyone? I think a healthy fear of centralization is a good thing... who tweaks the inputs to that vast computer algorithm? Thanks, I think I'll opt for a chaotic, dispersed ecosystem that contains some huge megafauna like Amazon, but leaves plenty of space for the little guys. Don't you know some fancy complexity/chaotic systems theorems that would give you pause?
Let's figure out what is possible before discussing the policy implications.

If Hayek is right that big govt central control of the economy is impossible, then why fear it? If it is possible and harmful, then we need to take steps to prevent it. If it is possible and beneficial, then we should welcome our new robot overlords.

The new high-tech economy appears to be a winner-take-all economy. Look at Google, Facebook, Amazon, Apple, Microsoft, etc. have all thrived by using massive economies of scale and central planning to dominate their markets. The Libertarian-Hayekian dream of a diversity of vendors is needed to address a diversity of consumers has turned out to be false.

We are finding out that at the billion-user scale, it is a whole new game, and computer AI takes over. Humans and Hayekian bargaining are as useless as humans trying to team up with that 3200-rating chess computer. The systems are just too complex for humans to tweak. It is big-data and AI all the way.

There are a lot of professors who projected limits to AI based on their experiments will millions of data items. But a lot of those problems evaporate when you jump from the million-scale to the billion-scale.

There are some AI programs used by Google and others with as many as a billion parameters that are determined by training on billions of data items. No one knows what the parameters do. No human can eyeball the parameters and tweak them for better results. It is impossible for any human to even understand 1% of it.

Look at this recent WSJ article:
Facebook is launching new tools to help marketers optimize their ads to target the people most likely to buy their products.

Some advertisers already share purchase data from their websites with Facebook via a tracking pixel so they can measure whether ads on the social network are generating sales. But the launch of the new tools, announced in a blog post Monday, will be the first time advertisers can use this data to optimize their campaigns toward the highest-spending customers.

Advertisers spend huge amounts of their budgets on researching the desired target audiences for their products. The idea for both new tools is to reduce the amount of money advertisers spend targeting the wrong people—known in the ad industry as wastage.

Facebook is hoping its algorithms, which take advantage of the rich set of data it has on its 1.9 billion users, will be able to do some of this job for them, or at the very least help them focus their ad campaigns on driving real purchases rather than just clicks. If Facebook can prove it is driving sales, advertisers are likely to increase their spending with the platform.

“This is a priority product for the company,” a Facebook spokesman said.

The first of the new tools, Value Optimization, uses prior purchase data to estimate how much money a person might spend in an advertiser’s store over a seven-day period. The tool then adjusts the advertiser’s campaign to send ads to the people it anticipates are most likely to actually spend the most money.

The second new product is an add-on to an existing tool called Lookalike Audiences, which launched four years ago and finds people on Facebook who display similar traits to advertisers’ existing customer lists. The add-on, called “Value-Based Lookalikes,” as the name would suggest, finds the Facebook users among those with similar traits who are also most likely to make a purchase after seeing an ad.

Both tools are rolling out from Monday and are free for marketers to use.
This is only possible because Facebook has a billion users, has millions of computers in server farms, and has sophisticated big data AI program managing everything.

The Hayek vision of a farmer bargaining with a vendor in his local village is obsolete.

Sure, there is room for little guys selling products, but more and more of them are going to end up advertising on Google or Facebook and selling on Amazon. Soon the little guys will not even be setting their prices, as they find it more efficient to let Amazon AI programs set them. They will just be cogs in a massive centrally-planned machine.

Maybe the Hayek fans have answers for all this, but I doubt it. What do they say about what Amazon and Facebook have already accomplished?
Here is the classic article: http://www.econlib.org/library/Essays/hykKnw1.html

Sowell wrote an outstanding book on this called Knowledge and Decisions. ... Always thought it was the best econ book I ever read. It re-launched my interest in free-market economics.
If they are thinking of markets on the million-scale, they might be right. But on the billion-scale I think that they are wrong.

They would probably say that the larger the market, the more you need independent human decision-makers. That is probably true up to the point up to the limits of the human brain. When you get to the billion-scale market, humans are useless and computer central planning becomes much more efficient.

No comments: