An AI can simulate an economy millions of times to create fairer tax policy | Hacker News Reader

An AI can simulate an economy millions of times to create fairer tax policy

technologyreview.com

159 pointsasamant150 comments

> In the simulation, four AI workers are each controlled by their own reinforcement-learning models. They interact with a two-dimensional world, gathering wood and stone and either trading these resources with others or using them to build houses, which earns them money. The workers have different levels of skill, which leads them to specialize. Lower-skilled workers learn they do better if they gather resources, and higher-skilled ones learn they do better if they buy resources to build houses.

So basically they're tackling one of the hardest macroeconomic problems with a simple and completely abstract microeconomic model which sounds less like cutting edge AI and more like a fusion of iterated prisoner dilemma models and the Edgeworth box.

> the AI’s policy cobbled together aspects of both, applying the highest tax rates to rich and poor

Sounds like they've overfitted to the model's assumptions so much their agents are choosing to not be poor primarily to reduce their tax bill...

It reminds me of the covid models, particularly the infamous Imperial model.

https://en.wikipedia.org/wiki/CovidSim

The question is whether it's useful to have a model at all, or whether it's better to just reason about it from first principles. If a basic simulation gave you a tool to explore emergent properties, perhaps some theories could be generalised to apply to real economies? Or maybe not, I don't know.

Models are supposed to be based on first-principles reasoning. A "model" that's not designed to capture some aspect of reality is called a computer game...

The Imperial model gives reproducible results, so most of the noise was professional programmers being salty about the fact that ugly code can still be accurate code.

The fact that it predicted things like 80% of the population becoming infected was simply unrelated to code quality.

Regardless of code quality, are these models really useful if, as you say, they are "accurate", but still make predictions wildly out of step with reality?

Usually, because they help narrow down the sources of error. If you've tested your code so you're sure the various bits work as intended, any remaining somewhat crazy results are down to things you don't understand properly.

A somewhat validated code allows you to do by computational experiments what you can't by real life experiments. Very handy if one of the common end states in the experiments is "death".

The real imperial story is that there were two wrong parties: - The scientists who were not including enough detail in their math. - The drive-by developers who had no idea what did or didn't make scientific simulations work.

Together they produced both an inaccurate estimate, and a github page full of uninformed speculation about what the problem might have been. ;)

Honestly I think the main problem was they choose overly pessimistic estimates, or maybe that their model is so expensive they could only investigate a few choices of the parameters.

same way that a given implementation is better than none. Its a concrete implementation that serves as an anchor. Its a fixed point to discuss implementation issues around as well as working as a harness that others can build on/around.

Wasn't it more professional programmers being salty that policy decisions were being made by code with absolutely no tests or peer review?

Code with no _unit tests_ or _code reviews_ possibly, it was obviously tested by the scientists that it made somewhat sensible and self-consistent predictions.

In addition, predictions by earlier versions of the code had been peer reviewed, which is about as good as it gets outside of computer science.

The report itself wasn't formally peer reviewed either, though in the crisis mentality of March this year not many people had the patience for peer review. Frankly, most still don't if you consider the amount of news about unreviewed science that get published.

gives? perhaps it does now, but it was certainly non-determinstic when it was first released, which is also when it was used for policy decisions.

See for example https://news.ycombinator.com/item?id=23099386

But how useful was the model? It showed that the virus would spread exponentially, but we knew that. Is there anything that it showed that we didn't know when they programmed it, with constants tuned to what they expected?

That was the point. They used stochastic partial differential equations, they're by definition non-deterministic.

No, they should be reproducible if run with the same seed.

Reproducibility doesn't mean accurate or good, it means it can be recreated. Its a crucial first step, but a model can be reproducible and also totally worthless.

Yeah, but as you can see in this thread the rock star programmers of HN only know enough epidemiology to complain about code formatting and improper RNG seeding.

An interesting article. I was unaware that John Carmack worked on the code before it was released.

An issue I see here is that while punishing the poor is clever way of reducing poverty, a monetary punishment risks trapping them in a negative spiral. Beating them up (but not so much they can't work) or even killing them might work better.

I hope people using AI for public policy are careful what they wish for.

Agree with these comments. Like maybe the "AI" found some "Hand of God" move for the economy....

Like many solutions that AIs come up with—such as some of AlphaZero’s game-winning moves—the result appears counterintuitive and not something that a human might have devised.

...or maybe their model doesn't include the fact that when you raise taxes on the poor too high, in eras of tremendous inequality, you get bread riots, political violence, and eventually, revolutions.

Not to say this kind of modeling isn't potentially useful!

Just modern monetary policy already uses algorithms etc (Taylor rule optimization etc), so the question is how can ML/AI supplement those humans with better tools, rather than thinking a computer is going to be able to deal with the complexity of the real world.

Put another way, what does the machine do when interest rates hit 0?

The so called A.I. was not working with anything that resembled a real economy. Standard economic models also have this problem, they also have other problems, but this particular example is just doubling down and then some. What is particularly unfortunate is that there are people out there trying to create real agent based models based on actual simulations of the Economy that are far more advanced than this (realistic markets, accurate simulations of the Banking System) but are not getting any of this kind of exposure.

Salesforce obviously is backing its own product, but the MIT Technology Review should really know better.

If the AI workers can't get sick, die, starve, suffer reduced productivity from chronic health issues, or care for children, then I don't see what lessons they can really expect to learn from this model. If their poor AI workers can just keep working hard gathering wood and stone forever, being fully productive without ever eating any food so they can afford to pay their taxes, then it's disconnected from reality beyond anything I'd even consider a "toy model".

I wonder if it's not punishing the poor, but trying to encourage them to earn more. Of course, this might not take into account the nuances of the real world where some people just can't because of constraints like children, location, disability, etc. That's where AI simulation might break down.

The more I think about it, the more I think the reason is that the policy network simply doesn't have training data for incomes that low so the shape is basically random.

And yet, as simple as this is, four agents is more than you see in most "complex" DSGE models... This looks to me like they're applying deep RL to agent-based modeling. It does seem pretty silly now, but it could be promising. You don't have to restrict these agents to any neoclassical assumptions and there isn't any reason you couldn't introduce firms as well. The only limit is computational power. Macroeconomics tends to look down on agent-based modeling because it doesn't provide neat, closed form solutions. But I've never bought that the results from DSGE models are anything more than products of all their simplifying assumptions. After all, they seem to be the very definition of overfitedness with absolutely no predictive power. At least they're trying something new here.

DSGE models are designed be solved on your laptop in 0.0001 seconds. There are more complicated models, like Krussell-Smith, but it's not clear that they help explain macroeconomics better.

Agent-based modelling isn't more popular because they assume that the agents are very, very stupid. Also, not only do they not produce closed-form solutions, they don't tend to lead to clear statements at all. Also also, the leading agent-based modeler, Cars Hommes, is infinitely better published than most economists, so he's not suffering too bad.

Whereas, in the real world, the poor might not get the chance to not be poor because destitution would cause their death.

I'd be interested in playing with such a simulation where the agents must buy or make food to eat and must have shelter or suffer.

And, also, that the other agents know this and can play their own hand accordingly.

I think this would change where the redistribution scheme would land quite a bit if the idea is to maximize value creation.

Yeah, it sounds more like they're training an AI for resource management games (like idk, Dwarf Fortress and co?)

The headline is misleading (click-bait). Reading the headline, I imagined they have simulated the entire economy of the world and have used reinforcement learning to find optimal policies, but they only simulated 4 workers and are making generalization to the entire economy.

How did I know that this would be the result from the headline? It's just like the old saying. Garbage in, garbage out.

An important distinction: publishable garbage!

Luckily, there is now a baseline, the next person can only publish improvements. Maybe one day we reach Hari Seldon levels of AI.

TBH I'd trust a describable model much more than "cutting edge AI"

Yep. Looking into their model in detail, whilst it has all the simplifications of a typical economic model designed for tractability and algebraic solutions (two good economy, four agents, tax officials have perfect information) , it also adds plenty of unusual assumptions like agents moving round a grid to find resources. Which as others point out might tell us a lot about how AI can optimise strategy games (more precisely that AI optimises for values in a strategy game better than simply using the Saez formula or US federal tax brackets), but very little about the relationship between debt, taxes and investment.

I would trust a monkey throwing darts at the stock page over "cutting edge AI".

I highly recommend Jeffrey Winters' book "Oligarchy" to understand tax policy. Though an academic book Oligarchy is very easy to read and surprisingly neutral politically.

Winters basically says that oligarchy is when a particular economic group has asymmetric power in a society, arguing that the main use of the power is wealth defense. He then outlines how this has four major forms:

- Warring oligarchies where warlords fight each other to protect wealth

- Ruling oligarchies similar to Italian city states

- Sultanistic oligarchies were multiple oligarchs align to support a single oligarchical ruler

- Civil oligarchies which are basically like the US and Singapore where oligarchs mostly use tax law and investment strategies for wealth defense.

The main evolution between these oligarchies is how much energy needs to be expended in wealth defense, the earlier stages requires constant violence against both the lower class and competing oligarchs, evolving to more cooperation between oligarchs and finally when the wealth class agree to the civil law so long as it can still defend its wealth.

In the case of civil oligarchy, Winters argues that nearly all of wealth defense and manipulation of the law is basically around tax codes and loop holes that allow wealth defense. The oligarchs in general are fine surrendering their military power (which is expensive anyway) so long as they can still defend their wealth. They agree to adhere to the rule of law, so long as the rule of law doesn't impact their ability to maintain wealth.

The conclusion of course being that we will always have complex tax law because it is required by the ruling oligarchy.

This is why we need to get rid of all deductions/credits/exemptions. If the society wants to incentivize something, the government can pay cash for it and account for it properly.

Exactly, eliminate tax shelters too! Expose the true financial reality to the tax system and a more equitable taxation policy can emerge. Although you'll have to mitigate a subset of people passing their income through foreign banks.

Americans basically can't use foreign banks in the way you say; in fact it's quite difficult to use them at all and remain a US citizen. The primary way the wealthy keep their money is by putting it in trust funds; non profits basically. That's where zuckerschmuck and Gates keep their dough. I'm all for getting rid of those, but most people will whine it nukes their favorite charity.

I was thinking along the lines of funneling income payments through corporate shells in foreign countries in order to evade personal taxes.

There are tons of ways techniques. The tax code enables this abuse increasing complexity and establishing loop holes. These benefits become virtually meaningless to the greater population and only works to the advantage of a few.

> "you'll have to mitigate a subset of people passing their income through foreign banks."

just tax it like any other income, which will likely allow the overall rate to be a bit lower. there's no moral or legal reason not to. the wealthy might threaten to leave (and take their tax payments with them), but it's largely a hollow threat, as people really want to live among their friends, peers, families, communities, their familiar amenities, and their accumulated stuff.

There are a ton of loopholes involving foreign banks. It’s more complex than “just tax it.” How do you tax income in a foreign bank if the bank doesn’t respond to IRS’s request for information?

You tell the taxpayer that you will just stipulate a very large sum until they instruct their bank to hand over the details. If they are willing to pay the stipulated amount instead of handing over the data you raise it because it was obviously too low.

you fine both bank and client for that behavior, and use the same might to pressure change that we've used against multiple small countries. it's not that hard policy-wise, but political will is certainly harder to muster.

I disagree that if a bank in _generic foreign country_ doesn’t work with _us agency_ we should start politically pressuring the _generic foreign country_.

Sure, while we do this we will start catching loop holing billionaires, but we will also begin effectively putting the IRS in a position where it knows (or can learn about) about literally every transaction in the entire global banking system. This seems like an undesirable outcome.

That simplified framework literally isn't actionable given nitty gritty details of valuation and deductions like business expenses - without them no low margin business may exist without a very low baseline tax rate as earning $100 on $90 of input would mean no to negative profit on a shockingly low total tax rate of 10%. The decision if it counts or not is a matter of semantics if it is "grossly overtaxing/giving a massive subsisy to the unprofitable".

Let alone all of the other "not technically a tax" semantics like licensing fees that get utterly stupid like - "Is it an exemption that license plated bicycles don't pay for emission inspections or gas taxes? That an original Model T hardly driven doesn't need a registration fee?"

I dislike the favortism and complexity but not having it is incompatible with the current allocation of power to self-govern and imposition of any new status quo would be transient and/or near impossible to get passed in the first place.

This is where most people would say "Just tax on profits" and then we get into a big rabbit hole of wtf is profit anyways?(just to kick start it "GAAP vs Non-GAAP")

Thank you for this very useful summary.

I would be curious to understand how a flat tax would interact with this premise.

It's an alternative strategy for oligarchies to defend their wealth

Or for people to keep the money they earn.

Yes, areed. Without taxes there would be no more state to defend private property, and the productive instruments could be managed by the people using them.

> It's an alternative strategy for oligarchies to defend their wealth

Not necessarily. A flat tax could have an arbitrarily high rate and be used to fund programs (e.g. UBI) that result in net transfers from the rich to the poor.

In practice what would likely happen, as happens today, is that people would make the case for tax exemptions for things like housing, medicine and education, arguing that they tax the poor, but then the exemptions in practice enrich landlords and drug companies and university athletics departments and lead to transfers from the poor because the portion of the exemption not going to the customer is also not going to fund the UBI.

That’s possible of any tax policy, but presumably a flat tax eliminates significant complexity from the tax code which reduces the possible avenues for wealth defense.

What other thoughts do you have on the issue rather than a simple dismissal?

A flat tax simply eliminates tax brackets; usually the upper one is one that goes. It's entirely possible to have a flat tax system with an incredibly complex and irrational set of deductions, just as it's possible to have a progressive tax system with zero deductions.

As for Winters' book, which I haven't read, it sounds like a more pessimistic version of Olson's Power and Prosperity, which essentially argued that all forms of security arrangement involved those with power extracting money from people, but kings had less incentive to pillage and more incentive to protect than bandits, and democracies more interest in using extracted wealth for public benefit than kings. I doubt Winter would disagree with my characterisation of campaigns for flat taxes as being one of oligarchs' wealth defence strategies in a democracy.

Thank you for the response.

A flat tax means all income is taxed at the same percentage rate. Or I guess it could mean everyone pays the same flat dollar amount in taxes.

A marginal income tax is just as simple, and not regressive.

A marginal income tax is dramatically more complicated because the rate depends on your income from other sources. If you take a contract for $500, whether you pay 15% or 25% depends among other things on whether you take a totally different contract with an independent party which pushes you into a higher bracket. It makes every transaction depend on every other one rather than allowing them to be independent, and requires a corresponding invasion of privacy to ensure compliance. Compare this to where the taxes are deducted by the employer without even having to disclose who the employee is to the tax authority, and a total elimination of tax paperwork on the part of the employee.

It also creates arbitrage opportunities to minimize the rate, e.g. hiring your nephew who is in a lower tax bracket and then deducting his wages as an expense from your own taxes even though the money remains in your family.

This is way overstating the complicated-ness of marginal income taxes for 99.9% of people. Adding up your income and performing 2 or 3 basic arithmetic operations is not "dramatically" more complicated, nor are marginal tax rate increases dramatic enough to affect any decision about doing more work or not.

And with the use of computers and electronic databases containing individual's unique ID # (e.g. social security #) and the internet to transmit payroll information, even a marginal income tax does not require employees to do any paperwork.

>It also creates arbitrage opportunities to minimize the rate, e.g. hiring your nephew who is in a lower tax bracket and then deducting his wages as an expense from your own taxes even though the money remains in your family.

This is irrelevant and I don't see why anyone would have a problem with people "minimizing" taxes in the way shown as it's above board, and tax policy taxes individuals, not families. If you want to consider money distributed to other members of your tribe as yours, then go for it.

That kind of stuff doesn't last too long in my experience though, once people get a taste of financial independence, the individuals in a "family" start looking out for themselves first.

> This is way overstating the complicated-ness of marginal income taxes for 99.9% of people.

Only because it is a cause of 99.9% of people operating that way. The amount of tax and other paperwork involved is one of the primary reasons people don't commonly take on small side jobs (or do but don't report them). That is one of the major costs of the complexity.

> And with the use of computers and electronic databases containing individual's unique ID # (e.g. social security #) and the internet to transmit payroll information

You don't see how this is, itself, a dramatic increase in complexity? With a flat tax individuals wouldn't even need to have a unique ID.

That we might be able to dump the complexity of this on businesses instead of individuals isn't making it go away, it's just making it harder to do business because now you need complicated tax software and central databases.

Also, the individual is still doing paperwork, because how else is the employer getting their ID number?

> This is irrelevant and I don't see why anyone would have a problem with people "minimizing" taxes in the way shown as it's above board, and tax policy taxes individuals, not families.

It's clearly a violation of the spirit of a progressive tax system. If a rich person buys his nephew a car, he has to pay his own marginal tax rate on it. If he instead pays his nephew the value of the car as a salary, he pays no tax on the money and the nephew pays the lower marginal rate. And that assuming the entire thing isn't just a kickback system where the rich person ends up with use of the car anyway and they privately agree to do it that way to split the "savings" from the marginal rate arbitrage.

It's subverting the intent of the system, in a way that can't be done if they were both paying the same marginal rate and progressivity was achieved by giving everyone a large fixed starting credit (i.e. UBI).

> That kind of stuff doesn't last too long in my experience though, once people get a taste of financial independence, the individuals in a "family" start looking out for themselves first.

But it's still a dependent relationship. The nephew only continues to get the money by staying on the good side of his "employer" -- it's the same situation where the wealthy use their money to control those in their social circle, but now you're giving them a tax discount for doing it.

That would depend on the tax rate. If the ruling class has a marginal tax of 20% and the flat tax is at least 20% it doesn't help them "defend their wealth."

Those frameworks sound a bit vapid and conspiracizing in taking mundane goals and trying to make them sound sinister as possible. Akin to racists taking immigrants and minorities having 3 or 4 kids on average instead of 2 as "planning long term white genocide".

Whether something preserves a given group's wealth is utterly independent of it being good or bad for everybody else. Rule of law and not having bandits choking trade also help to preserve their wealth. So would a hypothetical vaccine for all diseases. The definition of oligarch is also very non-standard as it would not only mean college educated technically qualify as they use their disproportionate earning to earn more to build and preserve their own wealth but so would teotoller factory workers over their peers as they would spend less wealth by not drinking and with compound interest/average stock marker growth and time could become a shocking fund.

To be frank in absense of other details in the summary the "framework" is so vaucous that he can conclude anything he wants including "the agricultural and food producers have ultimate control of the army, navy, and airforce" or "that dentists are the gatekeepers to being upper class".

From the article:

"In the simulation, four AI workers are each controlled by their own reinforcement-learning models. They interact with a two-dimensional world, gathering wood and stone and either trading these resources with others or using them to build houses, which earns them money. The workers have different levels of skill, which leads them to specialize."

So I'm sure they can create a fairer tax policy ... as long as our world is equivalent to a simplified version of age of empires.

This is just someone's hobby that they somehow convinced other people was serious work.

> So I'm sure they can create a fairer tax policy ... as long as our world is equivalent to a simplified version of age of empires.

Greetin. Correctus? Estorer.

The article seems link free but the below seem to be related to the project.

https://blog.einstein.ai/the-ai-economist/

https://einstein.ai/the-ai-economist

https://github.com/salesforce/ai-economist

[edit: Sarcasm removed. I think the project is exciting. I don't think the stated value of data/ML being fairer than politics is very strong because (a.) politics is reflective of culture and represents a sophisticated method of developing consensus among many different ideas of what fair means, (b.) tax accounting work will just shift to GAN and the end result won't be much different just more complex.]

GAN https://en.wikipedia.org/wiki/Generative_adversarial_network

As the saying goes:

To err is human, but to really foul things up you need a computer.

-- Paul R. Ehrlich

I've come to believe the most dangerous thing in the world is an expert with a computer model.

Maybe. I'm more worried about hundreds of thousands of amateurs with computer models (Excel spreadsheets.)

I'd actually be more okay with that because their models don't influence policy.

I don't see it that way; The aggregate effects of cost optimization by amateurs with spreadsheets at 100 F500 companies, 1000 mid-sized firms, 10,000 cites, 100,000 small firms has got to look a lot like policy.

FTA

> “It would be amazing to make tax policy less political and more data driven,”

Seems like an ...optimistic... viewpoint. IMO, the model, the data, and the notion of fairness are all political.

> "the result appears counterintuitive and not something that a human might have devised."

Will a legislator be willing to enact a policy that doesn't meet their intuitions?

> fairness are all political

Exactly. I came to say exactly that. Think a hypothetical of when you're a child and your mom has to decide how many cookies your mom gives you and your brother:

* Should mom give you equal cookies?

* Or should she your brother 2x the cookies because you're 5 and he's 15 and has 2x the caloric needs?

* Or should she give you 10x the cookies because he's had so many cookies in his life and you've had comparatively so few?

* Or should he give you more cookies because you're 1/2 his weight and we want to equalize those.

* Or maybe she should give you less cookies because you're fat for your age and he's not.

* Or maybe she should give you more cookies because you like them more than he does and would get more out of them.

* Or maybe she should give you chips because you're allergic to the chocolate in cookies and it would kill you. But what's the right chips to cookies ratio that is fair? And what if your brother also wants chips too, but we only have so many cookies and so many chips and him eating chips means less for you and he can eat cookies, so why should he get chips? But if you're getting 10 chips, and he's getting 0, how is that fair?

WTF is "fair"?

I don't know, that seems to match real-world policies, where there are unemployment aids but they're usually contingent on you being employed some of the time (e.g. runs out if you haven't been employed in X months).

Perhaps the specific example policy seems intuitive to some. I'm wondering about the general case. There's a problem of interpretability in AI models, it seems like there's also a related problem in terms of how to make decisions based on uninterpretable (unintuitive) models.

Yes, and there's also the possibility of over-fitting, finding a local minimum, or the population size being too small...

It can't do shit until you have a computable definition of "fair"

I think a better metric would be net value creation.

There are two times to care about inequality:

When the low-end creates physical suffering and when the high end controls enough they can force the low end into suffering.

But, between those markers there's a /lot/ of room to maneuver.

I really feel like the term inequality is a simple placeholder for a more complex idea that has morphed into it's own, mis-aimed, creature.

Indeed. Some view fairness as "I worked more, therefore I deserve more" and some view fairness as I should get the same as everyone else, regardless.

> In economics, the Gini coefficient is a measure of statistical dispersion intended to represent the income inequality or wealth inequality within a nation or any other group of people.

I'm guessing this is their metric.

https://en.m.wikipedia.org/wiki/Gini_coefficient

You can fix all wealth and income inequality with a simple 100% tax on everything, but that doesn't make it fair.

Gini just measures inequality doesn't it? So a perfect Gini coefficient would be quickly achieved with a 100% tax on all wealth and income. I'm not sure I'd say that was fair or a preferable way to live though...

It also assumes that "fair" is the desirable outcome.

The problem is not that we don't know how to create a fair tax policy. The problem is that the unfair winners under current policy spend their wealth to prevent one from being implemented.

It's a human greed problem; an AI cannot help us with that. This AI's advice world either be ignored, or its algorithms corrupted before it began.

> The problem is not that we don't know how to create a fair tax policy.

I would say this is a big chunk of the problem. people don't actually agree on what "fair" means. depending on who you ask, "fair" might mean anything between "I get to keep most of what I earn" and "everyone gets equal access to resources".

regardless of what "fair" means to you, you still need to understand tax incidence (who actually bears the burden of a particular tax) to devise policies that achieve your goal. tax incidence can be a very controversial topic.

I think you underestimate the complexity of tax policy. There are problems even with the definition of "fair".

What is a fair tax policy, then? Because the vast majority of tax policy debates I've seen are trying to craft a fair policy. Even folks I disagree with vehemently are generally interested in fairness.

> Even folks I disagree with vehemently are generally interested in fairness.

I tend to think this is the case as well...it seems that, if you can manage to get people into a proper state of mind, or carefully phrase questions to avoid triggering heuristics in the subconscious mind, in my experience "most" people have fairly substantial intuitions of compassion, group solidarity, a desire to do the right thing, and all sorts of other good stuff.

> What is a fair tax policy, then? Because the vast majority of tax policy debates I've seen are trying to craft a fair policy.

I wonder if a substantial part of the problem might be that the mental concept of "fairness" is inherently and extremely subjective - if we were somehow able to view the source code of the mind, I suspect each individual's implementation of FairnessCoefficient() for any given subject would involve an absolutely massive number of variables, some that would seem to have no obvious relevance to the matter, and the code would be utterly riddled with bugs. I estimate this to be the case based on observations of how people discuss ideas, and how their minds will produce different (and logically inconsistent) answers to the same question by making very simple changes to the variables you ask them to consider when asking a question - or, even using the exact same variables, but changing the wording of the question.

Another part of the problem of developing "fair" tax policy is linguistic: what does this word "fair"[1] mean? Like really mean, in the real world, the aggregate of all people's opinions? How might we know such a thing? Tax policy debates is one approach, but how accurate/optimal are the calculations these policies produce? Do they take into consideration the psychological impact it will have in a population, which in turn drives their behavior in a wide variety of ways (including voting), which in turn drives the real world success of the recommended policies? And, even if one assumes that these think tanks are producing fairly optimal (for the real world) recommendations, to what degree are they implemented consistent with their design? And where incomplete implementations occur (approximately always I would estimate), what are the downstream consequences of that - how far do the real world results skew from the model estimates when ~random changes are injected, or portions left out? How would we know the answers to such questions? How many people realize (like, really realize) how important these unknowns are? How many people actually think this level of detail is even worth acknowledging, let alone devoting some reasonable amount of effort into exploring? Perhaps some golden knowledge is located not too far below the surface? But how how would we know, if we don't even look? [2] I have no idea how, but it can easily be observed that most people seem to hold the belief that we do know such things (which is kind of interesting, if you think about it for a while.)

Regarding fairness being inherently subjective (a matter of opinion), and the obvious folly in outsourcing calculations to academics (and ultimately politicians) rather than letting those whose lives are effected decide, people often remark that "people aren't smart enough" to properly assess what is "fair". I think there's actually a lot of truth to this. But then, if one was to study a feral child, you might discover that they do not have the ability to do many things that we take for granted, like perform the most basic of mathematical operations. Similarly, if you rewind time sufficiently, you can also find a point where even the brightest minds on the planet were unable to design a machine that can transport people through the air. This suggests that human beings have a unique characteristic compared to other mammals: the ability to learn and understand very complex multivariate ideas, as well as the ability to pass this knowledge down to subsequent generations. Considering this, just as we discovered a way to consistently teach people how to perform mathematics (and some other things), might there also be a way to teach people to at least somewhat skilfully consider the complexities involved in economics?

But even if we did achieve that, there's still a problem in determining the aggregate of all people's opinions on the notion of economic fairness...how do we collect those opinions for analysis? As far as I can tell, this is widely considered to be a "solved problem" via two means: a) Democratic elections b) Public opinion surveys. My suspicion though, is that these two approaches may not actually be as optimal as we believe, in that several shortcomings should be fairly obvious with the mildest of examination. Despite this, it seems like these shortcomings are not considered noteworthy enough to do something about. Or, perhaps people don't actually consider ideas such as this, at all.

This structural/epistemic shortcoming (and the substantial unawareness of it) in our society seems to be not only limited to the topic of economic fairness. For example, I often read online (the news, forum discussions, serious academic papers, etc) assertions of the form "People believe <X>" - or even more generally, "<X> <Y> <Z>", where the values of X,Y,Z can be near anything describing the real world, and people seem to typically label these assertions as "the facts". A bit of problem with this system though, is that most people seem to not realize (and very often seem to have a very strong aversion to realizing, or even discussing) is that most of these things that we have categorized as facts, are nothing of the kind. Rather, they are estimates. Some of them are surely fairly accurate, and many of them are likely amazingly inaccurate (even backwards, the opposite of what is true)...but what they aren't, is "facts". And yet, we seem to believe the opposite of this (that they are facts), and we then act upon these false beliefs, and then proceed to complain endlessly online that "things suck", and attribute it to <x>, or <-x> (say, political parties), typically inline with other beliefs, which are also not true.

I wonder what a disinterested alien from another planet might think about the human species. Here we have these people who are able to achieve the most wondrous technological things, like splitting the atom, for example - and then with this knowledge, proceed to use it to design a weapon that can be dropped on other members of their species who live in far away lands, who happen to have different ideas than them (or something like that, the true causation may not be well understood).

They have the ability and the drive to see and understand the form of matter down to the quantum level, spending many of billions in the process, with no particular real world application in mind for this knowledge, but then simultaneously seem unwilling (perhaps, unable) to provide basic food and shelter for a substantial number of their species, or devote any widespread, serious/skilful mental energy into that arguably far more "important" problem (and this is just one example, there are many others).

They have the infinite curiosity required to develop technology that can see deep into the subatomic level of material reality, or light years into outer space - things that are at best mildly interesting to most people - yet simultaneously seem to have very little curiosity (if not an outright refusal) for attempting to understand that which causes widespread material discomfort (at home or abroad), things that have been complained about in widespread abundance, for centuries. They have seemingly endless energy and time for complaining about certain aspects of reality, but little time or curiosity for what actually causes it.[3]

What a curious species, these humans.

[1] https://www.merriam-webster.com/dictionary/fair

[2] https://en.wikipedia.org/wiki/Streetlight_effect

[3] http://media.moronail.net/images/stories/dg_pictures/0905/13...

It could for sure. But that assumes it is set up correctly. And even if it is setup correctly and runs realistic simulations, there are often in real life unforeseen variables like hurricanes, earthquakes, new discoveries, etc.

But how we define what is a fair tax policy is incredibly important. If the goal is not well defined how can an AI reach it?

What about other economic factors. What if a fair tax policy is not complete or possible without some other economic element(s) as a counterpart. What if a fair tax policy could only exist when Universal Basic Income is established.

But let's say that somehow magically all unexpected circumstances are accounted for. And the AI does successfully create a fair tax policy.

There will be those whose stand to loose a large amount under a fair tax policy. They are few, but they are powerful and wield massive influence over government. Would they go along with such a policy? It doesn't seem likely. It seems like even If you get a fair tax policy that is created, there will be people honestly questioning it's validity and people who are against regardless of how valid the result is.

It just seems unlikely to succeed. It may be useful economists might dissect what result it gives hoping to find new insights. But beyond that, I'm not sure how much good would realistically come from it.

Something like "fair" tax policy is impossible because the world isn't fair. It's like trying to find mathematical formula to make 2 x 2 equal 5. Anyone trying to shape tax policy is biased one way or another and is causing detachment from effort and reward which is extremely demotivating. It looks innocent on a spreadsheet, but tell someone who worked hard all their life that now they need to part with half of their salary to fund some vanity projects meanwhile they neighbour gets paid for watching telly all day.

In my opinion the fairness of the world is irrelevant as no matter what search space of possible and impossible combinations and get you will never find universal consensus on what qualifies as "fair" making the asking futile. Is a world where any form of violence is literally impossible unfair to others in worlds wracked by it or who desire it consensually?

With the little tiny weeny omission of debt from the model. “The economy”

And spending resources to maintain existence.

Simulation has been used extensively by economists going back decades. One example is the Dynamic Stochastic General Equilibrium model.

Often the output is determined and easily predicted by the input assumptions and the results you get back aren't surprising given an understanding of those assumptions as well as the process that generates the output. Quite often it's a case of garbage in garbage out. Worse yet is when the results are spun as a finding/discovery in the empirical sense when in reality it's very common for researchers to fudge the inputs in order to get an output that is coherent and aligns with both common sense and extant theory.

Simulation (Monte Carlo) is much more reliable in financial product research since the market does seem to obey some stable distributional dynamics, and the consequences on product pricing are well understood.

"Father, why are we only allowed 2 portions of stew a day?" "Hush, my child. Econatron 1.3 Nightly 20310201 dictates it. It sees all. It knows all."

The linked blog post doesn't mention anything about modelling the scarcity of land, so I'm a little skeptical about the results. In the modern world, one of the most biggest inequalities and barriers to efficiency is access to land, so not modelling that is a big oversight.

It looks like they also only considered income taxes, and ignored all other taxes (such as resources extraction taxes and land taxes), which is also a big oversight.

> However, finding a tax policy that optimizes equality along with productivity is an unsolved problem.

"A land value tax is generally favored by economists as (unlike other taxes) it does not cause economic inefficiency, and it tends to reduce inequality.

Land value tax has been referred to as "the perfect tax" and the economic efficiency of a land value tax has been known since the eighteenth century.", from https://en.wikipedia.org/wiki/Land_value_tax

> was 16% fairer than a state-of-the-art progressive tax framework studied by academic economists.

I wonder how fairness was measured.

This sounds rather similar to a lot of the work being done in macro/public finance right now.

If you look at a model from someone like Mendoza you’ll see that they have several parts of the economy characterized as recursive optimization problems + various constraints.

As others have pointed out, however, it is quite difficult to interpret these model results. The best you can do oftentimes is argue for why your setup makes sense then provide some intuition for what is going on. The problem is that slightly tweaked models are capable of delivering qualitatively divergent results.

I do think that reinforcement learning brings to the table some novel techniques for solving models with larger state spaces, though.

What's amusing is that you can probably do it without any 'fairness' tax at all - if there is sufficient saving.

You just need to ensure everybody can sell their labour to the central bank for a fixed amount and that the central bank can credibly threaten to create alternative production at the low end.

https://new-wayland.com/blog/look-ma-no-tax-from-zero-to-ful...

Denis Merigoux used formal methods to find discontinuities in the French tax code. This should be a matter of course, done but the OBM for every proposed tax law change.

https://blog.merigoux.fr/en/2019/12/20/taxes-formal-proofs.h...

> maximizing both productivity and income equality > highest tax rates to rich and poor and the lowest to middle-income workers

This somewhat obviously achieves both goals by encouraging productivity at the bottom and applying redistribution from the top.

But... real tax systems are designed to maximize revenue, just like any business.

> AI ... simulate ...

Wow, so now the whole of CSE (computational science and engineering) is AI.

Great, I can call myself AI researcher.

I know there's a lot of nuances in this, but I really with the US would head towards computer driven policies (with judiciary oversight). Think of it as Census v2.0. There's so many things that should not have human involvement like district lines (gerrymandering).

I'm not saying I disagree with you, but how would this work? At some point there has to be human decision when making the computer program that runs the policy. What's to keep the creators of the program from programming in their bias? Wouldn't the humans still teach the computer how to draw the district lines in a way that benefits their political party?

If politicians wanted things to be fair they could get rid of gerrymandering right now without the need for a computer program. So why would they be ok with a computer that got rid of gerrymandering for them?

Totally agreed, that's why some sort of judiciary oversight is so important. But the important thing is that the cost function you optimize has to be "constitutionally" correct. Similar to the Census, more population gives your more House or Representative seats, etc. I would expect you can create something similar for other policies.

E: And I also agree how today's problem should be able to be solved if policy makers wanted to. But if we went towards a data driven approach like the Census, these crooked policies should never been able to been in place to begin with.

The word fair has no meaning in a political context. To one it might seem fair that we all can afford the same things, to someone else it seems fair that we all keep what we make. And so on. A fairer tax policy is entirely in the eye of the beholder.

At the end of the article they say they’re releasing the source code, does anyone know where?

I think it may be this:

https://github.com/salesforce/ai-economist

Foundation is a flexible, modular, and composable framework to model socio-economic behaviors and dynamics with both agents and governments. This framework can be used in conjunction with reinforcement learning to learn optimal economic policies, as done by the AI Economist (https://www.einstein.ai/the-ai-economist).

The only "fair" tax policy is no tax policy. The function of taxation is to punish an activity, which is inherently discriminatory. In the context of a monetary sovereign, it serves no revenue function.

An AI youtuber who explains papers has covered it: https://www.youtube.com/watch?v=F5aaXrIMWyU

Highly recommend.

Opaque and complex models used to dictate economic policy. What could go wrong? I know Trump and Covid suck a lot of oxygen out of the room, but heck, 2008 wasn't all that long ago. Remember when a bunch of sophisticated models told us that a mass default on U.S. mortgages was statistically impossible?

Taxes are about tradeoffs, not fairness. You're talking money from people under the threat of jail, there's no fairness there, only a weighing of individual rights vs. the needs of the many. Every tax policy is a compromise of some sort with varying degrees of winners and losers. This is a political problem not a technical one.

The issue is that the bottleneck is not coming up with the solution - but actually getting the political capital to pass it.

Haven't we known for years which parts of the tax policy are outdated and unfair? (All of them).

it’s very cool work; i wish we could ban the use of “AI” as a noun. it only ever confuses people.

We can't even agree on what "fair" means. Setting aside tax avoidance, which is a different issue, people are upset Trump didn't have to pay taxes on years his businesses had massive losses. Many people think fair means "if you are rich, you should have to pay a large amount of taxes every year no matter what". But actually, you can be rich and bleeding money left and right. If you had negative annual income, even if you are rich, I don't think the government should also get in on the action and bleed you some more just because you have money. Tax money when it changes hands.

It's partially that everything about the man feels like a fishy facade and that the combination of running everything as a pass through but also having business bankruptcies. There's also the question of stories like [0] where there are huge loses claimed that on closer examination seem dubious and are later banned. I think it's the mixing of personal and 'business' finances that really muddle the waters here and that the twin facts of him being 'rich' and 'losing gobs of money for decades' feels unfair.

[0] https://www.seattletimes.com/nation-world/trump-lawyers-advi...

True, but take any <rich person>. If that rich person doesn't pay taxes because they legitimately had no positive income because their business is hemorrhaging, the media can easily spin that into public outrage.

"Mr. Moneybags over here paid fewer taxes last year than Working Class Joe! The injustice!"

Generally that doesn't happen for decades though, and when it does it's still an indictment of the idea that capitalism/America is meritocratic. The original business will eventually fail or the person runs out of money isn't that rich any more. When neither of those things happen it starts to look fishy, especially when it's known that the person uses legal at the time but dubious or even now illegal tax holes.

On top of that even if they're losing a lot of money they're still getting all the benefits of the government they're just not paying into the system in a proportion to the amount they're getting from it.

> On top of that even if they're losing a lot of money they're still getting all the benefits of the government they're just not paying into the system in a proportion

That's such a weak argument undermining your deeper potentially stronger retrospective. I've seen that argument before and it requires basically brainwashing that a particular government (the US, if we're being honest) is offering sooo much infrastructure that was required to make a profit. "Roads, schools, and such an amaaaazing court system". And that's just not true to domestic and heavily international business, and even if it was there would be a diminishing return that this argument doesn't even begin to account for.

I think a lot of the difference is just how expansively I see the benefits of things like schools. For instance no matter how poorly they perform relative to international standard they're still essentially free childcare for every person in the US which frees up so many people into the workforce. Just look at how much panic there was over the fact that keeping schools closed or very limited was the best way forward for public health, it really highlights how much work they do just letting more people into the workforce.

Exactly, governments are incentivizing transactions: “if you borrow and spend in the real economy then it offsets what you pay us, here are a few perks to help that along.”

People also dont seem to understand that a limited partnership can borrow other people’s money, have overhead costs and also pay you a salary+distributions, but the partnership tax returns passes all the overhead costs to your personal tax return showing a big “loss” while not showing you personally made money that year because the losses naturally eclipse that. You can still buy a mansion in all cash and not be in debt and not have anything to pay governments.

Doesn't take an army of accountants and lawyers to understand that. Well, except for the people that think they need an army of accountants and lawyers to form a Limited Partnership.

As one hedge fund manager put it: what exactly is "fair" anyway, and how will you know when you've attained it?

If I understood correctly "Good economics for hard times", this statement is not correct:

> taxing people too much can discourage them from working or motivate them to find ways to avoid paying—which reduces the overall pot.

What's the evidence baking that claim?

There seems to be a few studies which have attempted to derive the shape of the Laffer curve, and at least a few of them display a maximum below 100% taxation.

https://en.wikipedia.org/wiki/Laffer_curve#Empirical_analysi...

What's the evidence baking that claim?

This is called the Laffer Curve https://en.wikipedia.org/wiki/Laffer_curve

And again, where is the evidence for this claim? I've gone through the last 50 years of US tax cuts, and every time the taxes are cut, revenue decreases. The Laffer Curve may be an interesting theoretical model, but using it to influence any policy decision is laffable.

There is an entire section in the Wikipedia article describing empirical analysis.

"In 2017, Jacob Lundberg of the Uppsala University estimated Laffer curves for 27 OECD countries,..."

"In the early 1980s, Edgar L. Feige and Robert T. McGee developed a macroeconomic model..."

"A 1999 study...found no evidence that the United States was on the wrong side of the Laffer curve."

"Writing in 2010, John Quggin said...presented a model...A draft paper...presented a model...estimated...a stylized reduction..."

And there's Kansas...

As you point out the effects proposed by the Laffer curve have never been observed in the real world and instead the opposite has been true. It's an article of faith among many American conservatives though and faith cannot be overcome by facts.

This seems unnecessarily dismissive. It sounds like you're arguing the Laffer curve doesn't exist, which is a pretty bold claim not backed up by scholarship. This is a very different claim than "the US is on the left side of the curve" which even a lot of conservatives agree with. If you're arguing against the existence of the curve, it would mean that a marginal increase from 99% to 100% would increase revenue which seems unlikely.

The laffer curve works decently when there is high inflation. Currently we have low inflation. It is an interesting way to control inflation through the use of taxes. However, it has also been sold as a way to make taxes 'fair'. Fair is everyone pays the same no matter what. No freebies, no deductions, nothing. Once you introduce incentives or sin taxes you skew the system. There is now lots of money on the line and people will game that system.

Economics also has lots of math that 'works' but in the real world flies apart because of secondary effects. For example the 18th amendment. It sure got rid of booze but created an interesting underground of violence and ways to smuggle/hide it. A secondary effect. No tax issue exists in a vacuum. Anyone who says 'my way is the most amazing way' has not thought through all the secondary effects. For example lets say you want to get rid of something, million dollar cars. Lets tax it at 99%. The rich were going to buy it one way or the other. The tax is meaningless to them other than an annoyance that they will try to work around. You have not eliminated anything. You can get the same effect on the lower end. You want people to stop doing something but by putting a higher tax on it you have just basically said 'only rich people get this'. Many would not consider that 'fair' either.

The word 'fair' is also used many times. It is being used in place of equity. Because it sounds nicer.

The laffer curve is a way to introduce perfect price discrimination into tax law. Getting it 'right' is the same issue that most economic models have. They can define pretty well their supply curve. The problem comes when you try to define that demand curve. That thing changes from day to day and whatever I ate for breakfast or I slept too long and feel kinda groggy your ceo said something on twitter. Economists can kinda get it close but it moves around too much for you to make good decisions you can make kind of hand wavy ones. The demand curve changes! As you want MR=MC. But that point is more like an area. If you get that wrong you can over/under charge and mess things up later on. The laffer curve can plot its supply very nicely. But it has the same issue as all other models defining the demand curve. The laffer curve demand plot is usually conspicuously missing when shown to people. As the laffer curve would also be almost by definition a monopoly position as only one thing controls it. So getting MR=MC right is even more important. That is just the 2d curve on that thing. The 3d would be some definition of equity. I do not know of anyone looking into that sort of thing other than the quants.

I am not sure what you mean that "the Laffer curve doesn't exist"? It exists as a concept that has never been shown to be valid.

"It's an article of faith among many American conservatives"

Isn't it more accurate to say that they believe their country is on the right hand side of the curve?

Rise the price of your product 10x overnight and watch customers disappear.

You should consider the very basic economic concept of price-elasticity of demand. I might be able to raise it to 20x, watch customers disappear and make more revenue. So please, educate yourself a bit and next time try to put more effort into your response.

> I might be able to raise it to 20x, watch customers disappear and make more revenue.

No doubt, but that trick works only up to a point, after which you start losing customers rapidly and the whole business collapses after a critical mass switches to cheaper alternatives. The Laffer curve is an illustration of that idea. Push vehicle taxes to absurd levels (85-150% in Denmark) and you'll have half the population cycling to work. Some such effects might be desirable and socially/politically worth the loss of revenue, others - like a switch from industrially produced alcohol to moonshine, and resulting deaths - not so much.

Looking from Europe, the American obsession with low taxes over healthcare and general welfare seems short-sighted, but the expectation that you can infinitely raise taxes without ill side-effects isn't much brighter.

Here's the wisest economic saying I have ever heard: governments should run the economy like a housewife runs her household budget. That was Margaret Thatcher.

The second-most insightful saying on economics I have ever heard: The curious task of economics is to demonstrate to men how little they really know about what they imagine they can design. -- Friedrich August von Hayek

Forget the AI, forget the experts, forget the professors with their PhDs in economics. They're all just guessers, making guesses that are as likely to be wrong as they are to be right. Got that?

I strongly disagree with running the government like a household budget. It seems like a nice simple reduction but they're very different things.

Corporations don't run as household budgets, neither should governments.

Two main differences, growth and taxation. Growth means that economically you can outspend your "budget" every year and still come out ahead because that spend this year makes next year's budget larger. Taxation means your can spend money on any one initiative and if it increases economic activity in any way you can make it back via unrelated taxes without increasing any tax rates or fees.

Households have neither of these options.

A household could also just spend money hoping to increase next year's income (because you bought a bigger car or a new suit/dress). If that works, you'll also "come out ahead". But it usually doesn't, I guess. For taxation, you might consider stock market investments a similar thing "oh sure, I spend money here, but by doing so I stimulate the economy and I get dividends from my shares". Of course, they can't just declare themselves shareholders of other companies and send in the cops to collect money, that's a state-level privilege.

It seems to me that its not really that different from a household. A household can also end up ahead if it outspends in ways that contribute to future growth. In fact, most people do to get an education, buy a house, etc.

It seems to me that it's not a really question of whether or not its a household, its whether the spending actually contributes to growth. Fundamentally, that depends on how its spent. If you spend the money bombing countries in the middle east, probably not.

Nation is very different from a household. There is lot more people, they are anonymous. You need money to track who owns to whom, in household everybody knows. The worker specialization is much more relevant in the national economy. There is labor and trade and different political interests. I could go on and on.

Addendum: It might be true that in the nation as in the household, you should be aware if you are entering an external contract with someone, like borrowing money. But the question of internal debt between the citizens of the nation cannot be reduced to household economy, there is no such thing as an austere husband lending money to his wife.

Obviously, a nation is very different from a household.

But they aren't different in the way that the post I was responding to suggested, that a household couldn't possibly overspend for future benefit whereas a nation could.

No, that's not what the quote was about, the quote was saying that their economies should be managed the same way. Your interpretation of that quote is much narrower and in particular, it only applies to foreign debt.

I don't know what quote you are referring to as no quotations are involved in this discussion.

The post I was responding to said:

> Growth means that economically you can outspend > your "budget" every year and still come out ahead > because that spend this year makes next year's budget > larger."

I pointed out that's not different between a household and a nation, both can do this.

Yes, nations are very different from households. And yes, you shouldn't run their economies the same way. But to make that point you really need to point to criteria that differ between households and nations, and not to things which aren't actually fundamentally different.

Thatcher's government also required the assistance of a third-world military dictatorship to recover a group of near-uninhabitable rocks in the south Atlantic from another third-world dictatorship, in part because they had severely cut military spending.

"British people still believe in loyalty to their friends." (https://www.theguardian.com/world/1999/oct/06/pinochet.chile)

The US would've never been a country had it followed Thatcher's advice. It took borrowing to fund the Revolutionary war. It also misses or misunderstands the value of a central bank and a reserve currency.

That belief is only held by those who are hard of accounting.

The government's red ink is our black ink. Treasuries are savings assets for the private sector.

It's the desire the save in aggregate and be in the black that drives the red ink on the other side of the balance sheet, not the other way around. So to balance the budget you have to prevent anybody saving in aggregate, which is impossible to do without completely collapsing the economy 2008 style via an unsustainable private debt bubble.

People like to save. That has to be accommodated by the currency issuer.

> governments should run the economy like a housewife runs her household budget.

This is such a bad way to think about the economy, and cited as such so often, that I'd started to think of it as a strawman. Housewives don't print their own currencies.

You just described science if, after the guess, they test the guess and observe the results.

That wasn't wisdom at all, it was just a folk tale to justify a right-wing government pushing forward a slew of unnecessary budget cuts.

After all, a household does not have to invest in jobs, fund healthcare, maintain infrastructure and so on, as governments do. And it's been repeatedly demonstrated that there are damaging second-order effects from not doing this, which particularly hit the poorest and most vulnerable in society. Not that your average selfish right-winger would give a shit about that, of course.

Unfortunately, fans of right-wing politics tend to be too ideologically blinkered or just too stupid to realize how much of a fallacy it is, and the deleterious effect this has on governance.

Someone should read some of Richard Thaler's works. In other words you cannot simulate population's irrational behaviour. The fact that someone wants to take other person fruits of work and redistribute them using arbitrary rules, means anything but fair.

Any specific recommendation?

I would recommend reading "Misbehaving" that talks about economy and people behaviour. It is quite an eye opener.

Reply on news.ycombinator.com