Would the communist central price-fixing have benefited from AI? I mean, could the fixing of consumer prices have been automated by an AI system? Maybe using machine learning?
It's simple compared to modern machine learning, but linear programming [1] was initially developed by Leonid Kantorovich partly with the goal of automating central planning. Most of his ideas weren't implemented in practice though.
Here's a recent paper surveying some of that work and how one might approach the central allocation problem with the benefit of modern techniques: https://arxiv.org/abs/2005.01539
Allende tried something similar in Chile with project Cybersyn. Not AI yet but it shows how the economy could be centrally planned using technology. Shame that Pinochet's coup ended the experiment.
There are dozens of reasons. Free markets have deduplication of labour, concepts like patents and IP that are inefficient are necessary, there are economic crises that wreck compound growth in the long term, there are issues of inequality, humans may be lazy and not attempt to allocate their capital in the most appropriate way (see index funds), socially useful endeavors may not be profitable, etc...
But central planning is insanely hard, so it's not easy.
That said, the USSR which literally used pen and paper for planning was the #2 economy in the world, so it clearly has some potential.
Now there can be other reasons to not want planning.
> There are dozens of reasons. Free markets have deduplication of labour, concepts like patents and IP that are inefficient are necessary, there are economic crises that wreck compound growth in the long term, there are issues of inequality, humans may be lazy and not attempt to allocate their capital in the most appropriate way (see index funds), socially useful endeavors may not be profitable, etc...
There's no reason to think that economic crises and inequality would be absent in a planned economy.
The USSR was unaffected by the Great Depression, unlike all capitalist countries. It’s because it used planning for use instead of markets for profit.
The USSR didn't have the great depression, but they did have the HoloDomor or terror famine of Ukraine between 1932 and 1933 and the wider Soviet Famine of that period.
The history of the famine later went on to inspire the book Animal Farm and were a pure artifact of the central planning economy forcing the export of the very food required to feed millions of people.
Wasn't the Great Famine in 1932 and 1933 a direct result of central (mis-)planning? PS: also I wonder if you actually experienced living in a centrally planned economy, the East German economy for instance stumbled from one crisis into the next, not as bad as the Soviet Union under Stalin obviously (as in: people didn't starve to death), but a healthy economy is pretty much the opposite of the "real socialist" economies of the Eastern Bloc.
Doesn't it have more to do with the fact that they didn't even recover yet from the 1st World War? Also don't forget Holodomor. It literally happened after the Great Depression yet unlike Western Nations, 3.5 million people died from disease or starvation. Unlike in said Western countries that just had an economic collapse.
Of course there are. If you have good knowledge of the economy and of production mechanisms you can simply not do anything that could cause a crash. The issue is getting the knowledge to behind with, obviously.
Inequality is necessary to some level but empirically it's possible to make it much smaller.
Empirically the USSR had a single economic crisis without exogenous cause which was caused by incorrect accounting of agricultural production at its heart.
> If you have good knowledge of the economy and of production mechanisms you can simply not do anything that could cause a crash
But we don't have that knowledge. Economists disagree on many foundational questions. At some point, economics shades into psychology and we certainly don't have theories that predict human behavior.
I don't discount the possibility that knowledge can help economies run more smoothly but you're understating the difficulty here.
> Empirically the USSR had a single economic crisis without exogenous cause which was caused by incorrect accounting of agricultural production at its heart.
The idea that the USSR had one economic crisis is a bizarre joke. The only way you can justify this comment is to ignore all the shortages and poverty and make some tortured semantic argument about what constitutes a "crisis".
Empirically speaking, free markets trounced centrally planned economies in the 20th century. If you can't recognize that, I'm not sure why anyone would take you seriously.
In absolute numbers. But planning economies becomes so much more difficult at scale, though if we looked at the communist country with the highest per capita economy we would still see numbers that are middle of the pack.
The USSR never published a GDP figure. It couldn't, because GDP makes no sense without a market.
But yes, the USSR as a whole had a third of the US's average GDP. I agree, it was less than the US. The thing that is surprising is that they got there by planning an economy with pen and paper, which is insane to me.
There are potentially plausible reasons such as that a free economy system could take longer to achieve an efficient market equilibrium compared to a centrally planned system.
So long as a monopolistic centrally planned company acts on a free market do you consider your point unchallenged, or does the fact that the free markets tend to become dominated by large centrally planned corporations cause some tension?
There's no reason to think a free market economy is the single best solution we could ever hope to stumble upon. We keep evolving, we will evolve out of the "free" market economy as well.
The central planning will always produce better results, but only if you know the optimal strategy. Since it's impossible to know the optimal strategy a free market is a better solution in practice. Until a sufficiently strong AI arrives.
Of course, it can be planned - Lenin and Stalin showed that. Just make an AI that emulates those two geniuses.
But the results would be equally great like they were in Soviet Russia or Mao's China. Why would anyone in a centrally planned economy work beyond the bare minimum needed to survive?
It's hilarious that someone has to call you out on that. HN...
> Pinochet's coup ended the experiment.
I won't say that the coup was great, but it likely did less damage to the people than Cybersyn would have done.
> But the results would be equally great like they were in Soviet Russia or Mao's China.
That's your HUGE assumption. While I didn't make any assumption, you are making a big leap of faith here without anything to back it. Pretty much what I would call nonsense.
> Why would anyone in a centrally planned economy work beyond the bare minimum needed to survive?
Sigh. I don't know, maybe ask the many artists, intellectuals, sportsmen, scientists, etc that came out of URSS? How can you be so certain about human nature when nobody can agree on that?
Hayek did not, actually. His argument for it can't work but it can work using markets assumes P=NP and that humans are capable of running algorithms that computers can't, but only when they act as perfectly self interested rational actors.
It does not assume P=NP as far as I can tell, unless you understand something I've missed? I'd be very interested in an explanation for how a market assumes P=NP, if I got one I'd basically be done with the concept of markets because I do not believe P=NP.
A market economy is a distributed network, a self healing network and a self organizing network that optimizes itself for resource distribution, an emergent property of human populations with asymmetric distribution. The network itself runs algorithms much more efficiently than a centrally planned economy which does not have these traits. A sufficiently powerful AI could perform as well as a market economy, but it would be redundant and have other effects, such as taking agency from the benefactors of the economy. I don't know that it could outperform a market economy, it's possible that it could but so far I'm not sure.
The problem of maximizing your self-interest in a market is general enough that finding a solution to it can require solving an NP-hard problem. This is actually true, real financial problems are NP-hard.
The market does not optimize for resource distribution, it optimizes for profit. It is tyrannical in that it removes your agency, as a member of the market you have no other choice but to optimize for profit in many important situations.
This is already dysfunctional enough to completely destroy a society, so we have to implement fictions like IP and regulations to tame the beast.
As far as central planning goes it's certainly possible to plan an economy while allowing people to choose their jobs and start small companies.
A sufficiently advanced AI would inevitably be more efficient, simply releasing all IP restrictions, all binning that is not necessary, ending all unproductive landlordism, and the compound growth from the lack of crises would do it.
Now you could be opposed to it from an ideological point of view, but no matter what you're misstating Hayek's point, as he contends that it is not, actually, possible to perform the economic calculation outside of markets, due to computing brain magic of humans only when they act that way.
Also, the mere principle that a market can run the economic calculations more efficiently than another computer is a violation of the Church-Turing hypothesis. There is no machine that we know of that can do any computation with better complexity than a Turing machine, save for quantum computing.
Do you have an example of an NP hard problem a person would have to solve on the process of optimizing for self interest?
A market optimizes for whatever the participant wants, profit is simply a mechanism by which the overall market performs price discovery. You always have a choice to not optimize for profit, I make trades all the time where profit is not my goal.
Just choosing a job and being kindly allowed a small local marketplace is not agency.
Hayek's point would obviously violate the church-turing hypothesis (unless there's some hitherto unknown quantum-esque computation process going on in the human brain which I doubt, or you're mischaracterizing his point, I'm reading his book as we speak) but I believe that a sufficiently advanced machine capable of outperforming the market given equatable conditions (same information availability and processing) would probably perform those functions with less resource efficiency, not to mention the resources to build a redundant machine we don't need. If you build one that is better by using more energy and resources to scale its capabilities up, well now you're not optimizing resources on the whole because you're dedicating some for this machine to exist and operate, the resources to build the machine and maintain operation are then distributed suboptimally.
If a machine that uses the same net resources for this task and with equal capabilities could outperform the marketplace at efficiently distributing resources, and there are NP hard problems in personal finance, wouldn't that mean P=NP?
I suspect that if project cybersyn had succeeded, it would have eventually joined human central planners with statistical models, and then replaced the planners with ML once it became powerful enough (probably in the last few decades).
As far as I'm aware, the soviet system was bureaucratic rather than computerised, so they wouldn't have benefited as readily.
I've actually done some deep thinking on this topic, and I've come to a conclusion on the topic of the efficiency of central economies vs decentralized (market) economies, that is, a market is a network where the task of resource allocation is distributed dynamically to nodes best suited to make those decisions, the network is constantly reorganized to maintain low latency information channels, and the computational load is also spread to all the nodes, whereas a central economy has latency to the computational node(s) (government agencies and bureaucracies) and also reduced computational power (people making financial decisions with their minds). As a result, market economies will always result in more efficient and therefore more prosperous societies than centrally planned economies.
AI can change this, if it is sufficiently powerful enough to perform the computations needed for resource allocation, has the necessary information input (it can never have all of it but there's probably some place where it gets close enough), has low latency feeds for that information, and has very efficient distribution algorithms. That would basically need to create a system redundant to the economy itself, and redundant to a system that there's no worry about disappearing unless the humans for which resources are being managed disappear, so it seems pointless anyway. Also there's the problem of turning humanity into a cold machine and taking agency from every individual.
In sum, perhaps. But how much more prosperous, and to what end is that prosperity used? If prosperity were expressed in terms of apples, what's more beneficial for humanity: 80 apples in a shed that the owner won't let anyone touch and 20 for everyone else, or just 50 apples spread around? What about 80 apples?
There's a point in scarcity where inefficiency is the enemy to conquer, certainly. But there's another point where perfectly maximizing how many apples get put in sheds isn't winning a useful battle.
Where those points are, where we sit in relation, how do you convince people to pick apples in the first place, etc. are interesting questions. But without us being in a pretty specific spot for how hard apples are to come by, absolutely maximizing for prosperity across a population is not necessarily maximizing real utility for the population.
All those distribution problems are resolved optimally by the distributed system.
Optimally does not mean perfectly of course but there's no reason to believe that an artificially enforced system can distribute resources more efficiently than the naturally emerging system of markets.
They tried computer-assisted central planning and it mostly failed because of data quality. There were a lot of incentives to lie and not a lot of reasons to be honest.
When my grandma was in last class of school they were sent to help force farmers to use publicly funded artificial fertilizers. Farmers didn't wanted to because "their fathers didn't used them and it was fine". So communists had school kids with some physical workers deliver fertilizers to farmers and gather the data (how big area, how much fertilizer, what was the yield, etc.) And if somebody didn't wanted to take it they were supposed to call the militia on them.
Of course nobody at the ground level actually had any reason to gather the real data, check it, do the expected work or bother militia. So it was all a big exercise in faking data and passing it higher up :)
That was how most communist initiatives worked in Poland :)
The computer planning system that was suggested, GAS, in the 50s, included an internet that would automatically gather/collect and send data. The most that was ever done was to apply the algorithm manually to a few industries and see if it outperformed manual planning (it did).
The planners weren't stupid. They were aware of the data quality issues. That's why they basically wanted to build the internet in the 1950s to avoid that. Even Lenin knew this well hence the NEP.
The issue is that this threatened the bureaucrats so they invented problems to stick to the old ways of doing things that were easier to fraud.
I'm talking Poland specifically not USSR and there was some real computerization done by ZETO and CEOI from 60s. Poland had a few homegrown computer architectures and also since 70s had access to western IT industry (for example it used IBM mainframes pretty much as they were released).
For example PESEL system was designed then (each citizen gets assigned a number on birth, it encodes sex, date of birth, serial number from batch [so it doesn't need to be synchronized constantly], and a checksum so it's easy to find mistakes). And every database in country used that PESEL as "business id".
At first protein-based calculators were executing this algorithm but it was computerized and it's used to this day and works pretty well (even if there are some gotchas from the times when people were calculating them manually).
I'm talking about computerization of the planning process. I don't think PESEL counts. As far as I remember Poland didn't computerize any input/output economic tables or forecast consumption from purchase data or anything of the sort that was proposed in the 50s.
There was a little of that too, for example SIZ, SPIS and CENPLAN. They were introduced partially and never functioned as designed of course, after all it was a communist country :)
Yes, they were introduced partially as experiments. As far as I'm aware though there was never an automated input/output calculation which is the real step #1 of cybernetic planning.
Exactly, the bureaucrat caste became a wall against an efficient and democratic socialist planning, just as Trotsky predicted. They blocked the soviet cyberneticians plans and when the USSR collapsed they became capitalist themselves by seizing the state assets. The problem wasn't socialism but the stalinist bureaucracy. This article is on the point:
It wouldn't even need AI for massive benefit. Simply automating and authenticating data about which factory is producing what using what inputs would have been absolutely massive.
IMO capitalism and communism would meet with big data driven AI, it would benefit both: for capitalism algorithms would eliminate wastes, and drive cost down, so that the supply need not to far exceed demand; for communism central planning would actually work, you don't need market to adjust supply and demand. In the end these two system would work in a very similar way.
Here's a recent paper surveying some of that work and how one might approach the central allocation problem with the benefit of modern techniques: https://arxiv.org/abs/2005.01539
[1] https://en.wikipedia.org/wiki/Linear_programming