Ryan Beck

science

  • Why Nations Fail

    02 Mar 2020

    I just finished the book Why Nations Fail and really enjoyed it, and I highly recommend it to everyone. Its theory is that institutions are the most important factor in whether a country is prosperous or whether it’s poor and struggling. I think it makes a strong case for capitalism based on inclusive institutions. It uses some really interesting history to argue its case, and while most theories aren’t as strong as their creators make them out to be I do think it seems like a likely explanation for a big part of the difference between rich and poor countries.

    Why Nations Fail is written by two economists, and before publishing the book they wrote a really interesting related paper that I think is worth discussing and learning about. They don’t talk about this paper at all in the book, but it’s directly related to the subject.

    The paper is called “The Colonial Origins of Comparative Development” and it uses an interesting approach to show that colonialism (which directly affected the vast majority of countries on this planet) and the institutions established by it are responsible for why some countries are rich and some are poor today. Their hypothesis is that settler mortality determined whether an area was set up with extractive or inclusive institutions, and that those institutions persist to modern times in most places and determine whether a country is rich or poor. They use an instrumental variables approach to determine if their hypothesis is correct, which is a useful approach for determining causality when you can’t run an experiment.

    A common example of the instrumental variables approach is cigarette taxes and health. Though it seems obvious, it’s hard to prove with evidence that smoking causes bad health because you can’t force a random group of people to start smoking in order to compare their health outcomes to a control group. Since you can’t do that, it leaves open the possibility that maybe people with bad health choose to smoke as a comfort. In other words it’s difficult to determine which way the causality goes. So instead you can choose an instrumental variable, like cigarette taxes, and see if places with high cigarette taxes have better health. An important factor with instrumental variables is the variable can’t be a direct cause of the final result. The tax itself is going to have no direct effect on health outcomes. It’s only through the tax’s effect on the price of cigarettes that the health outcomes change. Another important component is that your causal chain has to be accurate. If places with higher cigarette taxes don’t actually have less smoking then even if their health is better your theory is busted. The causal chain is broken because smoking isn’t less than in other places. So using that instrumental variables approach you can indeed find that places with higher cigarette taxes have less smoking and better health. This is strong evidence of the link between smoking and bad health outcomes, and my understanding is instrumental variable approaches like this are kind of the gold standard in studying things where you can’t run a randomized controlled trial.

    The paper uses settler mortality as their instrumental variable. They theorize that settler mortality determined whether institutions became inclusive or extractive. In South America and Africa settler mortality was high due to disease. So the theory is that instead of settling there and establishing colonies where settlers would come to live, they set up extractive institutions to get slaves and gold. They established institutions based on forced labor, few or no rights, and the colonists using power and violence to rule with an iron fist and extract the wealth for themselves. In contrast, places like North America, New Zealand, and Australia had lower settler mortality. The colonists moved there to live, eventually setting up institutions where they would have political power and rights. The quote below is from the paper, discussing places where extractive institutions were developed.

    This is in sharp contrast to the colonial experience in Latin America during the seventeenth and eighteenth centuries, and in Asia and Africa during the nineteenth and early twentieth centuries. The main objective of the Spanish and the Portuguese colonization was to obtain gold and other valuables from America. Soon after the conquest, the Spanish crown granted rights to land and labor (the encomienda) and set up a complex mercantilist system of monopolies and trade regulations to extract resources from the colonies.

    Europeans developed the slave trade in Africa for similar reasons. Before the mid-nineteenth century, colonial powers were mostly restricted to the African coast and concentrated on monopolizing trade in slaves, gold, and other valuable commodities–witness the names used to describe West African countries: the Gold Coast, the Ivory Coast.

    But wait, couldn’t it be that places with high mortality are poorer just because they have more people dying making it harder to succeed? The paper addresses this point by noting that the mortality of native people was a lot lower than that of settlers. Many Africans develop a resistance to malaria so that while it’s still dangerous in childhood, adults are often not affected by it. Settler mortality often differed drastically from native mortality, and therefore using settler mortality is still valid.

    Something interesting the paper notes about settler awareness of mortality rates is that the early pilgrims who arrived to settle the United States had originally planned to go to Guyana until they learned of the high mortality rates in Guyana. People had enough information to have a good idea which places were deadly and which were safer to settle in.

    They also present evidence to prove their theory that early institutions are hard to change and often still persist to this day, which is why those early institutions are so important for modern day wealth.

    So the causal chain the paper uses is this:

    (Potential) settler mortality –> settlements –> early institutions –> current institutions –> current performance

    They find that their hypothesis is strong and accounts for a good portion of the difference in modern day wealth and performance. Places with high settler mortality are much poorer on average than those with low settler mortality, and they show that the causal chain holds up and that the type of institutions (extractive or inclusive) matters a lot.

    They check their work by controlling for or including a number of other factors like geography, colony origin (British, French, Dutch, etc.), prevalence of malaria, and so on. They find these other factors have little effect on the results and conclude that settler mortality and its effect on institutions is the most important factor.

    Like all theories it’s not perfect, and I know there has been some back and forth between the authors and another economist who criticized the quality of the settler mortality data. Getting accurate data from a few hundred years ago can be difficult. But it does seem to be a pretty convincing theory that seems to me to be respected by the authors’ peers. Even if it doesn’t explain all of the difference between rich and poor countries I think it makes a pretty strong case that the type of institutions a country has are important for its development and that those institutions depend on historical factors and can be difficult to change.

    Link to paper: https://economics.mit.edu/files/4123

  • Technology and Jobs

    13 Feb 2020

    People have been afraid of technology taking jobs for a long time. Apparently in the 1500s William Lee invented a knitting machine and went to Queen Elizabeth to apply for a patent. The story is second or third hand so it’s not a direct quote, but apparently part of her response went something like this:

    “My Lord, I have too much love to my poor people, who obtain their bread by the employment of knitting, to give my money to forward an invention which will tend to their ruin, by depriving them of employment, and thus make them beggars.”

    This isn’t entirely about technology because apparently it was common practice to grant monopolies to well-connected people, kind of like a more powerful patent, and part of her concern was about letting Lee have a monopoly in a time when there was some political opposition to the many monopolies that had been granted. But still it does seem like part of her concern was the technology taking jobs issue.

    Technology doesn’t eliminate jobs and make more people unemployed, it just changes which jobs people do. It also makes us all better off. In the short term we should be concerned about making sure people whose jobs become obsolete are supported and can still live comfortably, but demonizing technology just makes us all worse off in the long run.

    The quote and more info here: http://conversableeconomist.blogspot.com/2019/03/the-story-of-william-lee-and-his.html?m=1

  • Robots and Rubik's Cubes

    15 Oct 2019

    This is some really cool stuff. A group called Open AI has used machine learning to teach a robotic hand to solve a Rubik’s cube. They trained the hand to overcome a variety of interferences with its mechanics as well, such as having some of its fingers tied or having a stuffed giraffe try to move the cube around.

    According to their article it can solve a Rubik’s cube 60% of the time, and it can solve a Rubik’s cube starting at the maximum number of moves from completion (26) 20% of the time. If it drops the cube or times out it’s considered a failure. The movements aren’t programmed in directly, they trained the hand using a simulated model of the hand so that they could do thousands and thousands of iterations under a variety of random environments (different cube sizes, parts of the hand disabled, different finger friction, etc.). Then they applied the resulting program to the actual mechanical hand.

    It’s pretty amazing how much movement this hand has and how it adapts to the “perturbations” they apply. They can randomize a Rubik’s cube and place it in the hand and it’ll work out how to solve it. Check out their article about it for more information, it’s really fascinating and I definitely recommend checking out the part about the perturbations, the videos of that are really cool: https://openai.com/blog/solving-rubiks-cube/

  • Putting a Price on Carbon

    09 Oct 2019

    Hey Democrats, do you want to cut carbon emissions in half in just ten years? Do you want to increase benefits to the least fortunate among us? Do you want to increase development of green technology so it spreads to other countries?

    Hey Republicans, are you concerned that cutting carbon emissions means damaging the economy? Do you believe the vast body of evidence that we should be concerned about our carbon emissions, but don’t think bans on fossil fuels or the many non-environment related provisions in the Green New Deal are the way to do it?

    Well you’re in luck, because a carbon tax is perfect for everyone. Overwhelmingly supported by economists as the free market solution to climate change, a sufficiently sized carbon tax could cut our emissions in half in ten years with only minor impact on the economy. If you refund the revenue from the tax equally back into people’s pockets, the least fortunate get more back than they pay in.

    Here’s a neat tool that you can play with to show what effect certain carbon tax designs might have on carbon emissions and the economy. Seriously, everyone needs to be talking about putting a price on carbon and making sure our representatives know it’s important.

  • Uploading the Brain

    03 Oct 2019

    I posted a poll on Facebook and Twitter yesterday asked how long everyone thought it would take for us to be able to copy the human mind to a computer and I wanted to expand on that a little more. In my opinion it will be at least 100 years, though I do think it will happen eventually. Some philosophers and scientists are skeptical that it can happen at all, but for now I’m going to assume that it is possible (I do think it’s possible, I’ll post some discussion of some philosopher’s arguments and why I disagree in the future).

    For starters I think it’s helpful to look at where we’re at in our understanding of biological brains. The human brain is extremely complex, so scientists have started out small and are attempting to work their way up. One of the biggest achievements in this area so far has been the mapping of a roundworm brain. The roundworm brain map was created by scientists and is now being used in the OpenWorm project, which is a really fascinating project. The OpenWorm project wants to recreate a roundworm entirely within a computer. To do this they have to understand every function of a living roundworm, such as the exact configuration and connections between neurons and the exact muscle layout and behavior. Then they have to build it within a computer model. If they can do that, it’s possible that the computer simulation would essentially be a real roundworm, or at least act identically to one.

    Replicating a roundworm within a computer would be one of the first main steps on the path to copying the human brain into a machine. The first task necessary for modeling a roundworm was to map the roundworm brain, which means figuring out the location of every neuron and how they connect to each other. This map, or wiring diagram, of the brain is called a connectome, and the roundworm connectome has already been fully created. That’s a big and exciting achievement, but it’s also only the first major milestone of the OpenWorm project. Replicating the function of that connectome within a computer is likely several years away at the very least.

    So to get an idea of how far away we are from transferring a human mind to a computer, we can make a very rough estimate based on progress on the roundworm connectome. The roundworm connectome has 302 nuerons. Let’s just say it took about 10 years to develop. This is a very loose estimate, but part of the development process is creating software that aids in more quickly mapping the connectome and developing better techniques, and we’ll ignore the years of work that first went into scientists’ basic understanding of roundworm biology before work began on the full connectome. 302 neurons over 10 years is roughly 30 neurons per year. One of the next big projects currently being worked on is creating a connectome of a fruit fly. A fruit fly has 135,000 neurons. Recently scientists have imaged the full fly brain (about 21 million images) and turned that into a full 3D model. You can even explore a 3D model of the images and slices online. But that’s at the macro scale, or the big picture map of the brain. The full neuron-level connectome will require mapping and connecting those millions of images, and that will probably require the development of some new technology and/or software.

    So this is again very rough, but let’s assume the fly connectome might take 10 years to complete, which would be an average of 13,500 neurons per year. That would be about 450 times the rate of the development of the roundworm connectome. Let’s also assume that every 20 years our average speed at creating connectomes increases at a multiple of 450 times (the roundworm connectome was being worked on in roughly 2010 and my guess is the fly connectome might be mostly complete near the late 2020s). This multiplier represents the pace of improvements in technology that enable us to create these connectomes faster. So in the first 20 years you’re doing 30 neurons per year, the next 20 you’re doing 13,500 per year, and the 20 after that you’re doing 6,075,000 per year. The human brain is estimated to have about 100 billion neurons. So if we’re at 13,500 per year right now, 40 years from now we would be at 2.7 billion neurons per year. It would still take us 37 years at that rate to create a connectome of the human brain. But 60 years from now we would be at 1,200 billion neurons per year, easily fast enough to have already created the 100 billion neuron connectome of the human brain by then.

    Where that leaves us based on this very crude guess at the pace of connectome development is about 50 years to develop a connectome of the human brain. And that’s all just for the first step, that’s not even including trying to replicate the connectome within a computer. Sure, some of the development of computer systems that can model functional connectomes will happen in parallel to the creation of the connectome itself, but recreating the connectome in a computer will likely take much longer than creating a connectome of the human brain.

    It’s also important to note that all of the above is a bit simplified as well. Neurons aren’t just dots that turn on and off and are connected to each other by wires. They’re cells, and there may be more complex behavior going on within them or within the way they send signals. They may not just be pulsing on and off. Additionally, neurons communicate with each other through synapses, and synapses are constantly changing. So even if you map the brain and all its synapses, you have to understand how these synapses might change over time in order to more accurately replicate a living brain.

    For all of the reasons above, I think it will be well over 100 years before we’re able to copy a conscious brain into a computer. We have a long way to go to get there and our understanding of the brain is still limited. It seems likely to me that even if we are able to create a human connectome within 50 years or so, we’ll hit roadblocks or be limited by technology or a lack of information about how the brain works when trying to replicate it within a computer.

    My understanding of biology and the current state of brain research is very limited, so be aware some of what I said above is probably inaccurate. But I definitely recommend checking out the OpenWorm website to read more about it, and the 3D fly brain is really cool as well. It’s an exciting time to be alive, I hope within my lifetime we’ll see a lot of progress made in this area.

    OpenWorm: http://docs.openworm.org/en/latest/faq/
    Fly brain: https://tinyurl.com/y4xf46w8