Can An AI Design Our Tax Policy? 💰📊

❤️ Check out Perceptilabs and sign up for a free demo here: perceptilabs.com/papers
📝 The paper "The AI Economist: Improving Equality and Productivity with AI-Driven Tax Policies" is available here:
blog.einstein.ai/the-ai-econo...
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Haro, Alex Serban, Alex Paden, Andrew Melnychuk, Angelos Evripiotis, Benji Rabhan, Bruno Mikuš, Bryan Learn, Christian Ahlin, Eric Haddad, Eric Lau, Eric Martel, Gordon Child, Haris Husic, Jace O'Brien, Javier Bustamante, Joshua Goller, Lorin Atzberger, Lukas Biewald, Matthew Allen Fisher, Michael Albrecht, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Ramsey Elbasheer, Robin Graham, Steef, Taras Bobrovytsky, Thomas Krcmar, Torsten Reil, Tybie Fitzhugh.
If you wish to support the series, click here: / twominutepapers
Károly Zsolnai-Fehér's links:
Instagram: / twominutepapers
Twitter: / twominutepapers
Web: cg.tuwien.ac.at/~zsolnai/
#taxpolicy #taxes

[@cloudsquall88]

I just saw the title and said "Yes, please!" loudly

[@antman7673]

Me too. -I want a democracy based on scientists entering bits of data, all people to vote for the direction to set sails and an artificial intelligence to compute the way.

[@duncanw9901]

Equity is injustice

[@cloudsquall88]

@@duncanw9901 Why?

[@quebono100]

thats one of the evilst things on earth. :D because to make people productiv, just get rid of taxation. Problem solved!

[@emrahyalcin]

@@cloudsquall88 cf.kizlarsoruyor.com/q8520629/3323f416-b4c2-4252-a3ae-c7fad071d3a3-m.jpg

[@johntheux9238]

"AI will replace the most hated dangerous jobs" -> AI replace politicians. Sounds legit.

[@dan_loup]

The US president job is literally the most lethal job in the world. 20% chance of dying on it either of disease or murder.

[@johntheux9238]

@@maxsmith8196 Because they are so old? xD

[@rowboat10]

@@johntheux9238 well I guess that's another factor

[@Kuumin]

more than 70% of US presidents are dead

[@Relics]

@@Kuumin lol

[@scaredyfish]

Can we just appreciate the fact that they didn’t just simulate, they made graphics and animations too!

[@priyapepsi]

you should check out Primer. he makes stuff like this

[@MCRuCr]

And sooooo CUTE!

[@jwonz2054]

Nothing worse than a charismatic wrong answer.

[@quincy3367]

@@priyapepsi Wow, cool channel, imma binge watch all of them now.

[@MagicGonads]

@@jwonz2054 exactly what I'm thinking, they are spending resources on aesthetics for what should be extremely precise and separated from aesthetic design constraints

[@martinhenriksson8617]

It would be very useful if you could summarize what sort of tax policies suggested by the AI system led to this improved result.

[@stevep8490]

I think he said the variables were the tax rates and redistribution split. I don't believe it was specific

[@quirkyturtle4896]

After reading/scrolling the paper [search; Wealth Transfer and Tax Impact] there were two models, one with 'AI workers' and trade and another model with human workers and no trade. So there are some limits to the conclusions here. BUT MORE importantly wealth was equally redistributed between all workers (think UBI) as opposed to the system we have irl (pay non-workers). The net tax was negative for low(-180), middle (-150) and high earners (-80) with the top/highest earners paying all (400) after redistribution taxes. The AI workers (w/ trade) worked best under a model with effectively a flat 60% tax with two exeptions; 20% marginal taxes for middle bracket income. quote:"Compared with the baselines, the AI Economist features a more idiosyncratic structure: a blend of progressive and regressive schedules. In particular, it sets a higher top tax rate, a lower tax rate for [middle earners], and both higher and lower tax rates on [low earners]" The no trade model (w/ human and AI workers) worked best under a REGRESSIVE tax with one exception: no tax on the lowest income bracket. The Saez model is also regressive but they HAVE a very high income tax on the lowest bracket. Other problems include the AI 'gaming' the system to precisely control their income (high one year, low the next, repeating).

[@HaloDude557]

It's pseudoscience

[@waltercapa5265]

@@HaloDude557 It's probably more of a proof of concept. Which is true, we could probably create better economic systems by using technology instead of just letting the markets decide. Markets are really good servants, really horrible masters.

[@Wakish0069]

@@HaloDude557 Incorrect

[@o2dyt]

AI: "Politicians produce nothing and take a lot. Too inefficient. Engage kill switch" Now that's the future I want.

[@ExplosiveBolts]

Same with people on welfare.

[@MiguelAngel-fw4sk]

@And-Nonymous Body Mass Index?

[@kaplansedat]

creating an engage kill swith is a policy. if you decide to do that. Then ??? you are a politician

[@X-3K]

everyone gangsta til the AI commits tax fraud

[@andresmartinezramos7513]

jajajaja

[@cefirodewinter9086]

Ahhhh that's what I love about robots... They just can't do that

[@genericpseudoname160]

@@cefirodewinter9086 yes, the programmed ones can't, but with enough resources and time, the AI can change it's code in order to do whatever it needs or wants to do. Unless you add a "morality" or social score to it's actions (which might even get ignored by it's algorithm if it really feels like it), a flexible AI (basically any AI that can learn) would probably do whatever is in it's best interest.

[@thatoneguy9582]

@@cefirodewinter9086 i mean given the BullshitTM that the AI could do in the hide and seek video, I doubt

[@adrianmach7952]

@@genericpseudoname160 How does "morality" nullify the possibility of tax fraud? Everyone who did something evil most of the time had a perfect explanation for it, which made sense to them and people who followed. Racism can be easily explained as people cleansing our genepool from inferior ones, in the long term in their eyes this is beneficial to the mankind. I don't really see how any system that we have can not be twisted into justifying crimes whether that's religion, morality or ideology, I think the closest thing we have is actually law.

[@recrewn]

I thought for a second that this was Primer XD

[@mateusvmv]

Yeeah the blobs look a lot like Primer's

[@Falstad88]

Same. I'm disappointed it wasn't. If it were Primer we would get a full blown explanation, a series, and probably even the source code.

[@darkenergy7291]

same lol

[@justarandomsquid3568]

lol you watch him to

[@HyperIonMake]

I thought I was clicking on a primer video.

[@johannesschroter8984]

This was a very interesting paper and I hope to see more work in this direction. The thing that strikes me most is: Equality and Productivity is a fixed valued desired outcome in this calculation. The sweet spot between productivity and equality lies in the real world in the eye of the beholder. It is subjective. So my question is, the AI can give us the best answer to a question, but are we giving the AI the right questions to answer?

[@MrAlziepen]

Underrated comment. A sim to discover the right question to ask would be cool, like if productivity and equality are a false dichotomy or not. Subjective variables often remind me of the Donnie Darko lifeline featuring "fear-love." I get that they have to pick something simple and tangible so it makes sense for this I think, but I agree other variables mixed into it would make sense, wich at the end I think they were alluding to like with sustainability. Personally I think it would be cool to bake in a happiness value, and if enough little guys are repressed that they start a revolution or something.

[@austinmengarelli7269]

In my mind, equality (in terms of wealth distribution) shouldn’t even be a considered as an outcome. Rather, it should be more like “fairness” where everyone owns an amount proportional to their relative output

[@popopop984]

@@austinmengarelli7269 That’s somewhat unfair, people who own companies don’t actually contribute much to the company itself, they just got the ball rolling and let other people handle all the other tasks. Yet, since they did get the ball rolling, they get paid immensely and can even control the direction of the company. Wouldn’t they get nothing in your system? And if you say they should get a lot, then it’ll be the same as it is today.

[@austinmengarelli7269]

@popopop984 I think saying that they “got the ball rolling” downplays their role. Besides, it is their own company. All they did was engage in voluntary exchange with employees, investors and most importantly customers. No one was forced to do anything. As you describe it, you make it sound like starting Apple, Google or any of those companies is extremely simple… if that’s true everyone should do it

[@Xiox321]

This reminds me of that one episode of Love Death and Robots where some yogurt became sentient and solved the world's problems.

[@matt92hun]

>implying real humans would do things just because they make sense

[@FyL43]

how to be control by A.I

[@KnightMirkoYo]

It's about the market, not specifically single humans. The market is a self-optimizing system.

[@IceMetalPunk]

@@KnightMirkoYo Except... it's not. As this shows, the free market fails at optimizing for equality in any way, and only ends up benefiting the top minority of people at the detriment of the lower classes.

[@KnightMirkoYo]

@@IceMetalPunk Sure, I agree 100%. I was not talking about equality, specifically. The OP questioned the ability of single humans to make "correct" choices, and I say it doesn't matter. The AI would optimize for equality and productivity, while entities on the market optimize for own profit.

[@IceMetalPunk]

@@KnightMirkoYo "Entities on the market" are individual humans. The market is just people, it's not something separate from individual choices.

[@BromoDragoonFly]

Let's bow to our AI overlords for being more humane than our human counterparts

[@alpharius6206]

AI will say: the fuck u need such thing as money. Go full commie, I will redistrubute all the wealth myself!

[@trancetuberevived1131]

As long they are programmed to do so

[@fuckjoebiden]

@grogdizzy fuck off please. bezos did not earn his wealth, his workers earned it for him. same with every rich person. we as a society chose to give a large portion of our produced wealth to our employers, but we could easily shrink this and still keep everyone afloat, as this AI does. the AI increased both productivity and equality across society. is that not progress to you? you enjoy seeing people starve? the human race as a whole benefits

[@matthewhubka6350]

@@alpharius6206 and then immediately see that nobody is doing work. That isn’t what this model did for a reason. Similarly to why the model didn’t decide on a free market

[@davidwuhrer6704]

Isaac Asimov predicted this. We expect our tools to serve us. They can serve us best by being better humans that us.

[@TheYargonaut]

Having just read "Seeing like a State" by James C. Scott, I'm going to go with "no" or at least "not nearly there yet". The simulation makes a lot of convenient, but unrealistic and potentially fatal, simplifications, most obvious of which being a highly legible society with 100% tax compliance and uniform neoclassical utility function for all citizens. Oh, and the presumption of a governance goal that includes equalized net worth for some reason. It will be interesting to see follow-up work introducing more complexity and possibly applying this work in simulation games (like Cities:Skylines). Some factors off the top of my head for future papers to consider: - tax-dodging and enforcement (both physically and in terms of legibility); paper does mention one form employed even in this simulation by varying high/low income to game the rates - different governance goals, like maximizing regime revenue - competing states with citizen mobility and regime change possibility (can we make the same agent serve both governance and citizen roles, with various political systems?) - diverse Austrian subjective value utility functions for citizens - utility functions with "satisfaction" (e.g. utility = log(wealth)) - increasing diversity of skills and tradable goods (noticed, strangely, that houses were not tradable and had to be built by each agent for a direct utility cost) - state goods and services separate from just redistribution - capital goods which increase production, including physical capital (machines, land) and human capital (agents can increase skill level)

[@Unethical.Dodgson]

Yeah there's a lot that this paper doesn't account for just yet. I mean in the UK - Rockstar managed to get away with paying less tax than the average working middle class citizen simply because they threatened to leave if they had to pay more fairly.

[@filipwolffs]

There's definitely a lot that still needs to be done before this system could even hope to create an effective tax system for a real country. Even so it's nice that progress is being made in the fields of large-scale regulations.

[@Dayvit78]

The good thing about these AI algorithms is that they always improve - as this channel says "in two papers' time." So all your notes are valid and would be addressed with further improvements in the design. Gone are the days where people can say, "well that doesn't work, so let's leave everything as it is."

[@basbekjenl]

I mean as a starting point this is actually already pretty involved, sure adding parameters would make it more or less accurate and changing the existing parameters would make a big change too. The beauty is with this paper we can play around with those parameters and compare them to real world data to find which parameters and values would be most accurate. Naturally like predicting the weather this is never going to be perfect but all it has to outperform is our current system which is hardly a difficult thing to do given how flawed we make our decisions. Like self driving cars if one in a million crashes that's thousands of times better then what humans do behind the wheel. It won't be perfect it's not a calculator, this isn't math there won't be a single right answer but trying to get a better answer is worth believing in.

[@HappyGick]

@Alexander Why create an AI? Simple: the AI will not account for ideologies or morals, it will just reflect what's technically the best system, and from there we can improve upon it, and study if there are ways to implement it better based on our societies. That's why AIs are very good for these sorts of problems.

[@xwector]

Sounds great at first, but the study doesn't translate to the real world, because it's missing a key point: - The citizens can not immigrate, they have a "Berlin wall" around them. Therefore the wealthy can not immigrate if they feel the policies are unfair to them, into a more favorable economic environment for them. Implementing this keypoint may result in much worse performance for this AI.

[@BattousaiHBr]

there's a lot more missing in this simulation than just immigration. the point is that research has just begun and is already showing results, it's only going to get better from here.

[@xwector]

@@BattousaiHBr That's true, but I found that the most important missing keypoint. If those economies can't compete with each other I think the paper showing nothing really.

[@MrYjgh]

@@xwector just kill them :)

[@andrasbiro3007]

That's just another excuse to give the rich everything they want. While this can be a problem in reality, it's a far more complex issue than it is usually portrayed. For example the middle class can move too, and even the poor, not just the rich. I assume that's what you were referring to, because this system seems to be pretty fair to the poor and the middle class, and "unfair" to the ultra rich. Then you have to consider that moving isn't that simple, especially when you run a business, although it's getting easier as communication technology advances. Another question is, do we need ultra rich at all? If there's enough upward mobility, than there always will be enough new entrepreneurs to replace the ones that leave. The model optimized for productivity too, not just equality, so businesses still thrive. Lack of capital could be an issue, if there isn't enough wealthy investor, but the same technology that let's them leave also makes it far easier to raise money from the middle class and even the poor. Investing is no longer the privilege of the rich, and there are other ways of funding a project too, like Kickstarter and it's clones. Also, extremely high income taxes for the rich isn't a theoretical thing, many wealthy country does that, and even in the US it once was the norm, and that was exactly the time when the economy grew the fastest. Of course many other factors were in play that time ('50s and '60s), but still we can conclude that high taxes don't destroy the economy. The best run economies today are probably Sweden, Norway and Denmark, which all have high taxes, low income inequality, low poverty, but still very high per capita GDP, and they are doing very well by practically any metric you can think of. kzbin.info/www/bejne/aHaTlYp-eNOcmZo kzbin.info/www/bejne/nnyqqHp3fd-hfLM kzbin.info/www/bejne/m4Oyf42Zop2rsM0 That said, it would of course be interesting to simulate multiple countries with different economic systems and see how they interact and how that influences the result. It's not just migration, but also trade, politics and even war, which are all extremely important. Without corrupt politicians I don't think tax havens would exist, because they can be forced in many ways to end their unfair practices. In reality they exists because politicians tend to be rich and don't want to pay high taxes either.

[@gorkemvids4839]

Also you're missing another fact which earth has a wall around it called gravity and loneliness.

[@ahmadprogramming1197]

Let us appreciate the creators of the 3D models for a second because they are cute :)

[@Fluffytree03]

I'm sure the creators are indeed quite cute. :)

[@maxdefire]

Don't be fooled by communists.

[@maxdefire]

​@@nemo-x after all, communists show you a bright future with everything for free, but they all want you to work for yourself and not be able to use other people's labor for your own good ...

[@maxdefire]

​@@nemo-x obviously because the communists are trying to gain trust by pretending to be cute and filming cute videos. Nah, the point of communism is that if you don't want to work, you are no longer part of the community and you will die. And under capitalism you can be bloody rich without producing anything, just by owning a means of production.

[@maxdefire]

@@nemo-x Who knows how it works, we never had the opportunity to experience communism.

[@12q8]

This is very interesting. I want to see this in more complex situations.

[@saldownik]

Two papers down the line.

[@Bat0541]

This is how I felt about it as well, this is a really cool idea, but I am more excited to see what it does down the road.

[@Erbmon]

It's sketchy because it acounts for almost no variables but still makes statements about real life policies.

[@Erbmon]

@Finlayson I know, but that's not how papers are perceived by the public it goes like "Scientific study shows Capitalism is the worst economic system to ever exist!" you can not make economic statements in a narrow system, the tax policies they get this way are naïve, and tax policies isolated of other policies like immigration or customs are just worthless. They could say low taxation to the wealthy with low job regulations get this result but they stated capitalism and that is just wrong, capitalism is not what they are actually testing.

[@mikafizz1022]

@@Erbmon didn't they use the current economic status in a real life year though? We live in capatalism currently

[@danielpealer3561]

I'm my experience, limited as it may be, AI has a nasty tendency to do exactly as we tell it to, not what we mean for it to do. Say give it the goal to maximize nominal GDP and it might start randomly shuffling the money supply around while instituting a hyperinflation of the money supply. After all M×V=P×Q is useful as an approximation of GDP

[@jaredgarbo3679]

So whats the fix?

[@danielpealer3561]

@@jaredgarbo3679 There may very well not be one. I'm not going to pretend I have one. A free market works because each agent is able to break down the insanely complicated problem they face into a series of relatively simple problems related to variables they are concerned with. Yes, anything that any given Turing machine can compute can be computed by any other Turing machine, that much is true, but let us take the case of the Entrepreneur, this is a person who introduces new variables into the economic sphere, how do we account for this increasing complexity in the centralized system? The economy may very well be undecidable, in which case how is the AI supposed to look forward? In the simulation above they attempt to maximize the product of productivity and equality, aside from definitional issues here which honestly might require a paper in and of itself, say Bob Jones invents a Star Trek Style Replicator, a device which in the long run would massively increase both productivity and equality, but the production of which in the short term massively increases inequality in such a manner that the product of equality and productivity absolutely crashes, how is the AI supposed to account for this? I'm not sure it can. EDIT: The mere existence of entrepreneurs may function as a continuous adversarial attack on any economy regulating AI, for all we know a dude showing up selling Cheeze-Thems (Thank you Sam-O-Nella) could spontaneously convince the AI that it has absolutely maximized productivity and equality, just as a single pixel can convince some classifier AIs that a picture of a B2 Bomber is a dog. EX: kzbin.info/www/bejne/iXKXinh3jLujmM0

[@aidanwarren4980]

Exactly. Artificial Intelligences are devilishly hard to instruct. So often what can be measured is not what is desired. We want an increase in our standard of living, not a money shuffling machine. There's also the tendency of AI to replicate human biases, like computers that use racially biased data on policing and conviction rates and arrive at racist policy conclusions. I'm not saying humans are better, but we should be careful about handing over our policy machine to computers (or, to be more accurate, to the people building the algorithms).

[@CrudelyMade]

@@aidanwarren4980 I think it could be okay if a town or state tried an AI system.. see how that goes and if it's good, another state might adopt the same technique. @Jared Garbo, sometimes "This will make things worse" should be enough to prevent implementation. don't force change for the sake of change. wait until you actually have a good idea, otherwise you're just going to end up burning resources that could have been used elsewhere, making things even worse than just the new bad plan. consider Henry Hazlitt. ;-)

[@yashaswikulshreshtha1588]

Yeah, because we've habit of assuming hell lot of things, this is part of language, if you could just say a sentence and the other person just knows all the story and things associated with it, that becomes hell lot easier

[@Wertsir]

Considering the human competition, I, for one, _welcome_ our new AI overlords.

[@MouseGoat]

the ironic thing is tho, AI won't rule over us like terminators or make us slaves by becoming our masters. they gonna serve us to oblivion, you might think what can go wrong if we gonna remain the overlords. but at some point you wil wake up and see that you have given you hole life over to computers, free willingly.

[@christophsteck531]

@@MouseGoat but as long as the don’t have the General intelligence (and maybe consciousness) to put it all together it’s just more power to your logic-rational mind, for example wearing medical sensors that scan your Puls, blood sugar etc. and combined with AI etc. will help you optimize your sleep, diet, workouts, productivity in actual work etc.

[@AlecMuller]

"Gaming the system" can either mean *setting* the rules to give you an unfair advantage or *breaking* the rules to give you an unfair advantage. Simulations like this are great for identifying rules that were set badly, but they'll be useless for identifying how people will break them. Special interests will *definitely* push for rules that appear to be fair but are still easy for them to get away with breaking.

[@danwbeeston6146]

Interesting though - like most models run on unrealistic assumptions, but this is a whole new breed of modelling system

[@remicaron3191]

You sound like someone who's in a good place financially and has figured out how to game this system and doesn't want change. No system is perfect but if this offers some bennefits I still say it should be looked at and implemented. After all the US is now less equal than it was in the 1900's when we had no rights and most of the wealth was concentrated in the hands of the few. The reason we can't have progress in this area is that the well enough off constantly block change while the wealth gap grows. CHANGE IS HARD, EQUALITY ISN'T!

[@davidwuhrer6704]

_> they'll be useless for identifying how people will break them._ Wrong. Identifying that is the very benefit of using AI in this simulation.

[@RialuCaos]

Neural networks can also be used to find exploits. In some cases, they're even better at it than humans.

[@chrisray1567]

The “gaming” of the system that the AI agents learned was to have alternating years of high productivity and low productivity because that resulted in a lower taxes overall than a constant rate of productivity. This strategy was effective in regressive tax policies, which tax higher incomes at lower rates.

[@Kevroa1]

I really like the idea of taking a more objective approach to this topic and having an outside and unbiased source give a solution that is meant to be equally beneficial to everyone. I really hope this topic goes somewhere in the future. Unfortunately, I feel that if it gains any sort of traction it would be immediately shut down by lobbied politicians because it would undoubtedly be less favorable to large companies compared to now. My main concern is how reliable the model actually is.. I feel like there would be waaay too many parameters to think about in order to accurately depict this sort of thing reliably. If it were easy to model it then the current models would be way more effective. It is also limited by the parameters we already use. There are probably things that are not currently tracked, yet unbeknownst to us, have a tangible effect on the market.

[@complexobjects]

I first got inspired to learn RL from your channel and I eventually built my own multi-agent reinforcement learning system! This work is incredible. Haven't seen anything like this model before. Thanks for always covering the most cutting edge research Károly!

[@Darkfishi]

I'm currently studying Media Informatics and Visual Computing at the TU Wien and just wanted to thank you for keeping up my fascination for the topic with your short informative videos. Keep up the good work! It really is a great motivation and makes fascinating science approachable in a fun way. Thank you!

[@TwoMinutePapers]

You are very kind. Thank you so much!

[@themiddleman9376]

A wiener? Na servas!

[@hedu5303]

Servus too. I am studying statistic in Vienna

[@polishedmeat6399]

@@themiddleman9376 Do you really call people from Wien wieners? god.

[@xXxXLoneWolf103XxXx]

"Wealth Transfer and Tax Impact Compared with the baselines, the AI Economist features a more idiosyncratic structure: a blend of progressive and regressive schedules. In particular, it sets a higher top tax rate (income above 510), a lower tax rate for incomes between 160 and 510, and both higher and lower tax rates on incomes below 160. The collected taxes are redistributed evenly among the agents. In effect, the lower-income agents receive a net subsidy, even though their tax rates are higher (before subsidies). In other words, under the AI Economist, the lowest incomes have a lower tax burden compared to baselines. "

[@irisdogma8174]

Sounds a little like a UBI (the subsidies).

[@oxey_]

@@irisdogma8174 Pretty cool, i wish this paper came out years ago so we could see the progress because I can imagine this going very far in the long run

[@ataraxia7439]

Isn’t this what most countries do?

[@robertwyatt3912]

@@ataraxia7439 Yes, but we United States citizens for some reason think that we’re special and unique, and that these policies would somehow not work.

[@massimo4307]

@@robertwyatt3912 The US has plenty of welfare programs. Too many actually. And no, they don't always work.

[@asedtf]

This seems to assume several things: 1. Equality of outcomes is inherently good even at the cost of productivity 2. Opportunity is irrelevant 3. Social mobility is impossible 4. Closed system 5. Full compliance 6. Objective desires 7. No economy of scale 8. No mutual gain of value from trade 9. Government is efficient at distributing wealth 10. Taxation doesn't affect your choice to work harder I could think up more, as well as the obvious edge case spoilers.

[@matthewbennett7435]

Well yeah it's a simple model, that's the point. It operates on a lot of assumptions. It's not meant to be an ultimate guide to public policy. It's research. They even state in the video the outcomes were, "within the constraints of this simulation"( 3:45). The point is that with more research, simulations like these could be developed into valuable tools for policy making in the future. This isn't a map for all of our economic problems. It's a compass, one that we can set to point in a certain direction. And even then it's just a prototype compass. The hope is with time we can have this be a more accurate model of an economy and use it to help us come up with policy decisions.

[@DasGrosseFressen]

Just the animation where you have equality being orthogonal to productivity said everything...

[@KnakuanaRka]

What do you mean by that?

[@w-asabi2784]

That this is stupid... stupid conclusions only interesting part the reinforcement learning but this application is so unbelievably biased... as if capitalism and productivity was a sum zero game lol... I fear this propaganda... more productivity means more goods for everybody not just the richest part of the system, and I’m not even touching the point of how many problems are being solved by services they are producing... cheaper and better all the way up to the point everybody can eat different foods, travel the world at reasonable cost, have car, tv, air conditioning etc etc

[@w-asabi2784]

This looks like living in a world that is more equal than ever... I know it’s paradoxical for many people but that’s because they prefer to look bank account instead of quality of living and what we actually can buy... this is the more equal world ever and it’s only going to be better until socialism😢

[@DasGrosseFressen]

@@KnakuanaRka I mean that this assumption reflects the idea that productivity is independent of distribution of wealth (I question it and even argue that distribution has a net positive effect if done smart)... Evenmore and contrary to the first statement, at least in the video, the tradeoff is introduced as a given law. This is what it is optimised for, introducing an enormous inductive bias for the idea that somehow wealth distribution has a negative effect on productivity. This is the type of flashy, unscientific papers that CS loves. Given the hype of AI and the issues we have been seeing, it is dengerous...

[@ToyKeeper]

@@DasGrosseFressen Yeah, I was wondering about that too. From what I've seen when looking at a bunch of studies over the past couple decades, it seems very likely that spreading wealth more evenly (compared to the current state of the world) has a positive effect on overall productivity... and spreading wealth less evenly does not increase overall productivity. So the model could probably afford to prioritize wealth equality more (i.e. tax the rich to help the poor), without having to worry about it having a negative effect on productivity. Here's one recent study on the topic. It came to the same conclusion: www.lse.ac.uk/News/Latest-news-from-LSE/2020/L-December/Tax-cuts-for-the-rich Basically, rich people aren't special. They're not smarter or better than everyone else. Sure, sometimes they have good ideas, but so do other people. If they hoard resources, like if the 3 richest people have a net worth equivalent to the entire bottom 50% of the population (which is real, not hypothetical)... that basically means half the population can't afford to develop their ideas. We're missing out on half of the innovation we could have, because half of the people with the good ideas are financially unable to do the things necessary to turn those ideas into reality. In a situation like that, spreading wealth more evenly would actually increase productivity, because it would remove the barriers preventing half the population from innovating. So, er, I hope the societal model used in the AI will take this sort of thing into account.

[@OregonburlsStephen]

as a programmer many of the biggest mistakes come from missing factors that affect your simulation. something like this would have to reliably match up to the real world before any of its predictions could be given weight. even then its recommendations could not be blindly followed and would need to be use as guidance but not law.

[@coder0xff]

I thought the same thing. Maybe first make an AI that predicts the outcome of policy based on real-world data... though any novel policy would be problematic because it falls outside the domain of the training data.

[@Muskar2]

One of the easy tells about its limitations is that its model of the US tax system doesn't match the real wealth distribution in the US.

[@YASxYT]

This is gonna be your most controversial video haha

[@cakersthecake5337]

True

[@BigFatSandwitch]

Why controversial?

[@cakersthecake5337]

@@BigFatSandwitch p o l i t i c

[@pbj4184]

@@BigFatSandwitch Because he supports (or atleast seems to be ok with) redistributing wealth through taxes which people like me would like to disagree with. The video makes it look as if taxes are the obvious way to guide everyone into prosperity but of course, it isn't so simple

[@masterchief7301]

@@pbj4184 and would you like to explain why?

[@AlexH-nt1gg]

I am just amazed how high the productivity is in a free market system.

[@AlexH-nt1gg]

@Christian Moore Well, I don't want to engage in this comlex topic. But please note the goal of a free market system is to maximize productivity and efficiency. Not to imporve the live standards for all, which is way in the early stage of the industriy revoluation, people even kids were working 10+ hours without any day off. The labor movements made the employers to sacrfice productities to provide better protection to the workers, which had eventually resulted today's standard 8 hours a days and 5 days a week. Unlike this stimulation, people are not bots, if higher productivities does not translate into better life, it is pointless to most of people. This is why the pure free market system will never happen.

[@AlexH-nt1gg]

@Christian Moore Workers and kids in the 18th century were not "coerced," but they had to, becuase the communal lands were seized, they had to work in the factories to survive, and if you decided not to work there, there were plenty of poors who would take the job. "free" does not mean "free" if you don't even have a choice.

[@fado3057]

@Christian Moore "free to run multiple or no job", do you means that in your free market definition, people doesn't need money to live? In this case : yes, people will only work if their job make sense for them, but with the variable Money, you can't talk about freedom without assuming the voluntary bondage. A drug addict is only free if he is not obliged to take drugs. The need for productivity is about the limited time we have on earth, the need to trade living time for money and then for free living : is nothing but slavery.

[@fado3057]

@Christian Moore no this is called a tree. Maybe do you see your relationship with fruit slavery, and that you are enslaved by trees. But it's not free of charge, you just have to give awaythe seed you have capitalised inside you.

[@fado3057]

There is multiple example in nature where equilibrium exist between one that give and one that receive. This is named peace. Exchange with one taking more than he give is named Wear or attrition. complaisance in this is named sado masochism. Now don't keep your seed, go through the field to refill your alliance with life. 😉

[@adamphelps2369]

This was a very fascinating paper! Thanks for the video! I’d love to see a second paper on this :)

[@CaptainYesz]

I'm optimistic as well BUT I'm also wary that "hey look, we have a model of human behavior!" is a mistake that's been made time and time again Edge case scenarios can ruin certain people so modeling requires deep humility before applying it and toying with people's lives

[@saldownik]

Governments kill and ruin daily, it isn't a big of a concern.

[@AtticusDenzil]

can still be used as a suggestion and then further analyzed to see what the implications are

[@emanuellandin7403]

Mmm is not a model of human behavior.

[@NatnatXS]

politicans, lobbyists and bankers toy with peoples lives everyday. The chance that an AI will be better than money hungry beasts is pretty high though.

[@willguggn2]

They've had humans play who were overall less productive, in part because of their adversarial playstyle blocking each other.

[@HebaruSan]

At this point my hand reflexively moves toward my papers when I see there's a new upload from this channel

[@jwonz2054]

He has devalued his slogan just like the AI tax model will devalue currency.

[@ObjectsInMotion]

I need to put my papers down otherwise I'd drop them from the sheer excitement!

[@MrGreglego]

I don't know if it's the volume I'm listening at, but it's 1:00am and your whisper voice is sooooo relaxing

[@jacobringgaardmikkelsen5103]

Its pretty simple for me. you keep what you make

[@HappyGingerWolf]

But then how does anything get done by the government and how do rich people not end up owning everything?

[@Koubles]

@@HappyGingerWolf Some tax policy is fine, particularly VAT taxes for both buyers and sellers are good revenue generators that reward the government on economic activity and growth. However, modern income taxes and the various sub-taxes under them have become a massive government overstep over the last 100 years. The job of the government first and foremost should be to preserve the basic human rights of every citizen. Life, liberty, property, voting, etc. Everyone is viewed equally by law, and lobbying wouldn't do anything since government economic control only expands to preserving human rights. This means the government should only receive the funding to perform these tasks, which mainly would consist of public police and military. This is pretty much the Libertarian view of government and was for the most part the main political ideology that influenced the constitution, albeit they ended up making some compromises with the federalists like with the house and senate, henceforth why it was called the great compromise. Currently, the US has a very complex tax policy where pretty much everything is subject to tax except for certain key things on the average tax file, which can be difficult for the average person to navigate properly. With all the corruption that comes with a congressional body that has granted itself near-total economic control aside, the federal government currently spends about 20 Billion dollars every day. To put this into perspective, liquidating the companies of the top ten richest people in America without it devaluing or worrying about the hundreds of thousands of valuable jobs that would be lost, would raise enough money to run the federal government for around a year or so depending on spending increases. The united states government has a massive tax and spending problem, and if it does not get under control eventually a huge economic disaster will occur between the United States and her allies she supports. It's certainly a problem that needs to be discussed more, and it's unfortunate that a huge majority of US political figures from both parties are calling for more programs, spending, and taxes which does not bode well.

[@MushookieMan]

In other news, AI simulated a spherical cow.

[@sageinit]

Too true. Underrated comment.

[@C.I...]

And why productivity? Why not a happiness metric?

[@filipwolffs]

@@C.I... Happiness is a lot harder to quantify than productivity and wealth equality.

[@andrasfogarasi5014]

@@filipwolffs And then comes the philosophical question of happiness distribution! Should we subject a single innocent man to an eternity of unimaginable pain if it makes the rest of the world significantly better off? If not, how about only a lifetime of suffering? How about a year of pain? A month of discomfort? A day of inconvenience? What balance of total happiness and happiness equality should we strive to achieve? This is why Stirner was right. Assuming that other people matter, impossible questions arise.

[@antonf.9278]

spherical cow in a perfect vacuum

[@tomdip2094]

I think people really look too much into this. There is no possible way to accurately model something like this which would be easily transferable to an economy or people. There is zero mention of saving or investing, and the AI people are in a closed system with relatively few variables to account for. Another flaw is the assumption that reaching the pareto curve is ideal at all, which can only be determined by peoples individual values, and certainly not a computer. You really can't quantify this stuff without exactly modelling human behaviour on an extremely detailed level, variance in personality, generational changes, and the value system of the people themselves.

[@saldownik]

it's not about accurate modeling, it's about approximate modeling, AI just needs to be better than us

[@agustinpizarro]

@@saldownik but maybe free market is better than AI

[@stevengreidinger8295]

Tax policy simulation is an ongoing field under examination in multiple projects, and it's getting a lot better. See the long run of work done on microsimulation, as well as a potentially great improvement on this work by the team of Selmer Bringsjord at RPI.

[@13MrMusic]

well if the real economy was as simple as rock, wood and building houses this thing would make some sense.

[@gorkyd7912]

Should work in any universe in which the government simply redistributes all tax money... Which is none.

[@finn8518]

you do realize the animations were just for showing something, the actual AI itself takes into account way more complex structures… did you really think a scientific paper working with highly advanced techniques like AI would calculate using rocks and wood???

[@13MrMusic]

@@finn8518 Yes man, I know its hard to believe that this is a serious paper but yea, they only use wood and stone. Its really pathetic. Read the paper and youll see

[@Krokant17]

It still makes sense in what you call "real economy". The cumulation of wealth can be simulated in many minimalistic models and it also shows up in real world. It is not the point to be a perfect model, but that doesn't mean it makes no sense, I.e. physics is much more complex than Newton's mechanics but for many phenomena it is enough to know them and I wouldn't say "if real physics would just be adding some forces, Newton would make some sense", if you know what I try to say

[@jaikumar848]

Economic explained channel need to see this😀

[@Edkahmed]

Great crossover

[@A_B_1917]

Considering the fact he seems to be Keynesian, I think he would be interested in this.

[@martiddy]

That would be a great collab haha

[@KishoreKumar-uz8ir]

Yeah if anyone who is in his Discord Channel sees this comment please let him know about this video. I would definitely love to watch him do a video on this.

[@GKS225]

Yes! Let him scrutinize this!

[@mrleemrleeohmrlee]

This is a great idea, except we're still defining what we think is most important, i.e. income equality. What if we're wrong about that? Can we get an AI to determine what's better: focusing on income inequality vs. pulling people out of poverty?

[@davidwuhrer6704]

We just don't agree on what is most important. To the top player, equality puts his position in jeopardy, decreases his relative wealth, and is bad for productivity, so not only is it not important, it is downright harmful and dangerous. The least skilled worker assigns a high value to equality, because it means he may get to eat.

[@f__kyoudegenerates]

@@davidwuhrer6704 income equality means stealing from other people to make people who didn't work for what they have equal. literally that is all it means. I know poor people who pulled themselves out of extreme poverty. It's bs to say there is inequality. equality of outcome how ever is socialist tyranny.

[@bramvanduijn8086]

AI is good at achieving goals, not so much at coming up with them. Also, we may not like the goals they come up with.

[@davidwuhrer6704]

@@bramvanduijn8086 AI is, as the same says, artificial. That means that an AI's goals are exactly what we make them. But we can't change their goals afterwards.

[@davidwuhrer6704]

@@f__kyoudegenerates _> income equality means stealing from other people to make people who didn't work for what they have equal._ How is creating a wider gap between the rich and the poor making people equal?

[@ChaseAerospace]

WAY oversimplified the economy, but we have to start somewhere! Can't wait to see what this can do with a better economy simulation to play around in. I'm thinking dynamic consumer demand, simulated supply chain errors, differences in product quality for similar products across many industries instead of just building houses, the AI actually negotiating prices, etc. And does the AI take paying off national debt into consideration at all? Good thing quantum computers are on the way! lol, maybe life is a simulation someone put together to discover some "optimal" tax policy because that's honestly the level of complexity an "accurate" economy simulation would need.

[@fm56001]

yes

[@jcasp0947]

this the most amazed and engaged ive prob ever been watching a youtube vid, i wanna do this kinda stuff when im older

[@cybertrk]

Equality of opportunity or equality of outcome? A very important distinction...

[@lastinspace2103]

I think outcome because all the wealth is pretty distributed

[@user-my9fe2nv4m]

Yea, I'm worried about this too. If the Ai is trying to maximize equality on all fronts, it could end horrendously.

[@LifeInspector]

If productivity was also higher than the other two models, it shouldn't matter; it's better for everyone on the whole, as a society.

[@supaflylob]

an increase in equality of opportunity will also increase equality of outcomes. they are not conflicting goals even though people like to make it seem like they are

[@farmingice8715]

Judging by the results, it looks like equality of outcome.

[@chraman169]

The authors assumed that equality of outcome is something desirable.

[@sebastiansandoval4861]

Well for a lot of people it is, for politicians i'm not so sure

[@rinrin4711]

@@sebastiansandoval4861, it is not desirable for anyone. Equality of opportunity, however, is.

[@bramvanduijn8086]

They assumed a non-zero equality is desirable. I'm sad that they didn't run the communism tax model where all productivity is taxed 100% at all times and then distributed evenly. I'm curious what the agents would do in that situation. On one hand, for each coin of productivity they delivery they get a quarter coin, so they should always be improving their productivity, on the other hand, defection is less punished in such a system, if you stop working you would gain 100% of your input energy, but lose coins in proportion to the proportion you produced before. So if you produced 5% of the coins, you'd lose 1/20th of your income if you stop working, but if you produced 60% of the coins, then you'd lose 60% of your income. The question is: would they stop working? It all depends on how the agents value their energy.

[@HaloDude557]

@@bramvanduijn8086 the model is pseudoscience garbage and wouldn't teach us anything

[@germangarcia6118]

@@rinrin4711 Then we need to remove inheritances from the system so everyone has equal opportunity

[@KipringPayne]

There is a serious flaw - motivation. If you take away from a productive AI - it will continue to work. Do that for a person - their productivity will wane. Free Market beats out everything - every time.

[@Capybarrrraaaa]

It really doesn't, the free market, at least without massive limitation, is a horrible idea. The majority of people don't understand what they buy, and that can end up causing all sorts of issues from battery-farming, dangerous products, sufficiency-not-quality. And even on the economic side of it; free markets tend only to push money towards those who already have money. Without a "good faith" competitor, there's just no reason to 'compete'; cooperative price-fixing wins. And that's assuming that you're right with that "productivity will wane". When you're losing money you don't need, not much happens, but incentivising people who would otherwise not have money they need creates a lot of motivation.

[@hamoonist]

The parameter was equality x productivity. It assumes equality is a good things. I’d like to see one where it only cares about the well being of the least advantageous. In other words, in which model the bottom 10% will have a better life. I’ll be really surprised if I see the suggested model will be significantly different than a free market capitalist system.

[@keksentdecker]

wow, this sounds awesome! There are so many exciting papers out there but without channels like you I would probably not find them so easily

[@saldownik]

1985, here we come.

[@cadespaulding3837]

We still need to decide how important equality and productivity are in relation to each other.

[@Muskar2]

Equality is just the oversimplified measure for utilitarian happiness, right?

[@ToyKeeper]

Yeah... It's probably a good idea to make it attempt to maximize quality of life first, and only factor in productivity far enough to make sure it's still positive. Like, ensure that scientific and technological progress are being made, at least a little, but otherwise focus on maxing out equality and sustainability. Moving forward slowly is arguably better for humans anyway, since people don't cope well with rapid change... so the model shouldn't try to optimize for faster productivity. Mostly just make sure everyone gets what they need, and make sure we don't burn the planet in the process.

[@cadespaulding3837]

We also need to factor in the fact that the same amount of money goes farther in more productive societies.

[@MrTomyCJ]

@@Muskar2 I don't think so, and it should't be that, because that isn't the definition of equality. Neither I think that using equality as a primary parameter is a good idea. The primary should be quality of life or something similar. You can have more equality in a poor country where someone has $10 and the other $15, than in a rich country where someone has $100 and the other $200.

[@MrTomyCJ]

The animations ilustrating a "balance" between equality and production is misleading and not impartial. Fist, equality != well being, and second, it implies that one is in detriment of the other, when in reality a normal system should be so that one increases the other (productivity -> well being).

[@bootlegcaesar7481]

I feel like there should have been a fairness factor since equality isn't always fair

[@NietzscheanMan]

A model produces what its creators want it to produce.

[@skellurip]

ai taxation > free market taxation (at least in this experimentation) ai powered centralized distribution > free market distribution (look at amazon distribution model it's centralized and ai powered) mutual and index fund > free market stocks exchange (index fund is safer and usually perform much better than individually traded stocks)

[@DanatronOne]

TECHNOCRACY!

[@z-beeblebrox]

As it happens, "Technocracy" is already claimed by "government run by technical experts". Government run by AI goes by the far less sexy "Cyberocracy", which doesn't roll off the tongue at all, and sounds like it was coined by an 80s comic book writer, but that's what we're stuck with.

[@DanatronOne]

@@z-beeblebrox kinda like how "astrology" was already claimed so people who study space have to use "astronomy"

[@sageinit]

@@z-beeblebrox I'd prefer recursive metacyberocracy over cyberocracy, maybe. Gotta keep track of those higher order dynamics.

[@456MrPeople]

Shit, I'd take that over what we have right now.

[@filipwolffs]

@@z-beeblebrox Could we maybe call it a synthocracy?

[@user-wr9hc6gf2n]

Politicians denounce science all the time, they would never allow for an AI to show the way.

[@noikristjansson6038]

But this is the way

[@sofia.eris.bauhaus]

maybe we shouldn't leave the law to the government, then 🏴.

[@aicat253]

@@sofia.eris.bauhaus based and breadpilled.

[@saldownik]

Fool, politicians will use AI for sure to maximize effectiveness of their policies soon if they aren't doing it today. They will have different goals in mind though. Not that you will ever learn the parameters they will feed to AI.

[@user-wr9hc6gf2n]

@@nagualdesign One of the basic requirements for a functioning democracy is an educated electorate. The Electorate can be ill-educated, but it is not always so. A well educated electorate is less emotional and reactionary as well. Corrupt democracies prefer them to be ill-educated so they can be easier manipulated through social media and other propaganda.

[@axmoylotl]

i've had this idea for a while and its really cool to see someone capable actually create it. I honestly think more advanced versions of these games/ai will be how countries are run in the future, which will either mean really good or humanity killing AI

[@adrianrg75]

Human nature is inherently creative and imperfect. You cannot program human action.

[@finn8518]

not that of single human beings, but as a population we‘re actually pretty predictable, sort of like atoms

[@superresistant8041]

You don’t have to program individuals. You program crowds.

[@kovecsroli]

"What a time to be alive" :O - completely agree

[@odw32]

It will be pretty interesting if "democracy" starts taking on a new meaning -- one where we don't vote on people who promise outcomes, but on the outcomes themselves. There's always choice. Things like which comforts & liberties we want, or the minimum boundary for poverty we consider humane, or the importance of the environment or our own health over wealth; Those are not a set of optimums which can be calculated by an AI (yet?), not without a general AI which understands things like "happiness" and "ethics". But if an AI can calculate policy from parameters, we could just vote on how the outcome should be balanced -- You'd eliminate the whole issue of honesty in politics, the need for politicians would even disappear as the budgets and department focus automatically sorts itself out after every election.

[@Xcess007]

*mind blown*

[@triffid0hunter]

That's called direct democracy, as opposed to representative democracy - see en.wikipedia.org/wiki/Direct_democracy

[@yourikhan4425]

Interesting. I wonder what effect it would have on the motivation for the workers to be as efficient as possible.

[@NorthEevee]

This actually seems like a fun simulation game, where you have to micro manage a small town. I'd love to play it as it has some interesting mechanics, with it being AI driven and all.

[@Alexis-kg1sm]

Remembers me of Majesty: The Fantasy Kingdom Sim

[@casualgamingroom1472]

@@Alexis-kg1sm Thanks for the tip, I found it on Steam! Here I go!!

[@user-xl5kd6il6c]

Equality of Outcome isn't a valid metric, which throws the whole paper out the window. Also funny how even their own paper proves Free Market is the better system and they didn't even notice

[@wilburdemitel8468]

based

[@hedgie9823]

How so

[@user-xl5kd6il6c]

@@hedgie9823 In simple terms, because communism is the only system that achieves it, by pulling everyone down. Exceeding in a field takes a lot of effort and dedication, effort that no one will dedicate their life for if there's no reward for it. (no matter how much communists lie about it). I can explain it more in depth if you want, tell me if you disagree with any point more specifically

[@elmegagamer1]

@@user-xl5kd6il6c dude scientists dont get shit and they make almost everything, then someone just comes and makes profit off it

[@elmegagamer1]

Please tell me what was the economic incentive for universities creating the internet

[@BigDaddyWes]

"They look at inequality and wealth distribution." Well that's the big difference there.

[@anteeko]

This, what is the point of maximizing equality? I don’t understand.

[@jaideepshekhar4621]

@@anteeko You're okay with one person having 90% of the wealth, and the rest having 10% of the wealth? Place yourself in their position bud.

[@GMMDMMG]

If this paper were to consider the ethical components of the system it is studying, it would come to the realization that the only difference between the free market model and the other two is that in the other two there basically is legalized robbery (tax), because the economic agents do not have the choice to participate or not in the tax scheme, while in a good simulation of a free market model people would actually be able to voluntarily associate themselves with certain central/government-like (but not coercive/tyranical) services (because nothing in the free market makes it impossible for people to develop artificial intelligences and use them to navigate economically). That means, fundamentally, that you could have a free market model that scores the same as the A.I. Economist model in the "productivity x equality" valuation (which is a proportionally inverse relation, just like "rational x coercive") because the A.I. Economist model would be voluntarily used. However, who would want to produce things to others solely because the others did not? I'm sure there are people that would like it, since there are people that agree with tyranny and would like to have their production redistributed (and as long as that is not forced upon individuals, it is just like regularized donation), but that is something people should decide and not something decided upon poeple. A fair community is the one that sees as real it's constituents.

[@LongRest]

This is great field of work. I hope that better and more complex simulations will follow. Right now i have a concern about a little details, that people often use to reduce their expenses. For example, they can move to a different tax zone or use non-official employees (this is how it works in my country). It's a far way to go for AI solution of tax problem. I hope we will get there sometime.

[@Lumcoin]

Optimizing for sustainability -> price externalities, use the same system and it will work

[@tanishqpandya3700]

Perceptilabs, bruh i needed something like this but the everywhere the costs were super high

[@TwoMinutePapers]

👌

[@sageinit]

@@TwoMinutePapers also check out machinations.io

[@OrbitalLizardStudios]

The attention to detail with the visuals is amazing, it kind of looks like the Link's Awakening remake,

[@andregusmao3196]

The sad part is the 3:45 constraints, it is impossible to simulate the complex and caos environment of the real world but I believe as we refine the representative models we can have a fair approximation of the reality in the future.

[@AricGardnerMontreal]

The most skilled AI who build all the houses is probably not as motivated as the model would suggest Also where is the agent that does nothing all day?

[@dianagama3390]

And how efficient are subsidies in improving low-skilled workers' productivity? Or are they there just for the sake of redistribution?

[@Tvde1]

@@dianagama3390 exactly. It's a funny animation, but useless

[@Earth1960]

"Also where is the agent that does nothing all day?" Exactly where he has always been, in the imagination of Conservatives.

[@NotFine]

Never would i have thought i would see these arguments boil up on two minute papers

[@robertwyatt3912]

Hey man, just because a topic is controversial, doesn’t mean that we shouldn’t have a rational discussion about it.

[@NotFine]

@@robertwyatt3912 I agree

[@GAPIntoTheGame]

This is definitely amazing. But as with any model of reality, in order to determine it’s accuracy it must be empirically tested until exhaustion. If it is accurate enough, that means that any assumptions about the agents is a sufficiently good proxy for humans, kind of scary to think we’re so easily reduced to basic elements without losing accuracy

[@thailux6494]

As an economist, this is so cool! Tools like this will increase welfare and accelerate the rate at which we discover more things about how the economy and economic agents work!

[@seamusoblainn4603]

The question isn't can AI help, but would government, corporations, and oligarchs listen.

[@cybertrk]

By and for the people... citizens vote two ways... with their ballot and with their wallet.

[@BagerAkbay]

we might also try to model an AI to teach us how to make them listen

[@RialuCaos]

I think we all should already know the answer to that question.

[@davidwuhrer6704]

AI is certainly helping corporations and plutocrats.

[@Rojk]

They already use simulations?

[@rolandsbreza7448]

This reminds me of a channel Primer and I love it

[@Foxsloth]

Wealth inequality is not inherently unfair. Highly productive people should hold a majority of the wealth, since in a free market they can only attain wealth they’ve created.

[@Foxsloth]

In contrast, wealth redistribution IS inherently theft.

[@Foxsloth]

It also doesn’t make sense to multiply “equality” and productivity. Wealth equality isn’t the same as purchasing power. Meaning that absolute $ income may increase, but taxation also increases costs of goods and services. ie doubling your income doesn’t mean much when prices are also doubled.

[@Foxsloth]

Using this same multiplier as a score of economic success would result in high scores for places like the USSR. Where most ppl were equally poor, yet all were forced into the workforce. The designer of the sim had an outcome in mind.

[@mkaeterna9161]

Probably optimal in theory, but a soulless neolibral hellscape in practice

[@Onihikage]

Rather than equality, I'd like to see this simulation assess the happiness of the actors in the system, using the same factors that are known to affect human levels of happiness (such as having agency, meeting one's needs, natural vs artificial environment, feeling productive, etc.). Combine that with Dr. Károly's suggestion of environmental sustainability, and AI like this could be an extremely useful tool for assessing strategies to solve some of the biggest problems facing humankind. I can't wait to see the next paper!

[@bonafernando]

If the AI makes everyone equally poorer, it accomplishes its goal. I agree that happiness is more important than equality. If I have a super rich neighbor that owns the moon, but I'm happy. I'd be ok with that.

[@Shefetoful]

I think the productivity equality graph is highly misleading even if it's only a simplification, it's nowhere near like that.

[@saldownik]

Good catch, seems bogus like hell. Mapping the curve with AI assistance would be informative.

[@snowballeffect7812]

How is it then? Seems pretty accurate at least for the one based on real-life.

[@dustinmorrison6315]

Cool stuff. It gets a lot more difficult when you consider captive states and bilateral tax treaties.

[@petersmythe6462]

My thoughts: 1. Why stop at tax policy? Simulate an entire virtual economy. Or network of economies, militaries, and more, using agent models. Afterall, we should be concerned if our policy produces an incentive to invade someone. 2. Non-cooperation is always a concern for any system in which the state does not take the money out before private individuals see a dime. I'd love to see simulations where non-cooperation is possible and law enforcement is also simulated. 3. Modern day economies are one of two things, globalized or North Korean. Simulating the tax policy of North Korea is not very interesting because they just increase the price of products to pay for the government in a sort of ultra-direct hidden sales tax on non-necessities. They don't use a conventional income tax or private sales tax. This means that if you are NOT simulating North Korea you need to consider globalization as a factor. I.E. There will be foreign companies operating in a country, there will be domestic companies operating in foreign markets, there will be tax havens, there will be migrant labor, etc. Companies may relocate all or part of their operation based on incentives. 4. Economic agent models would have a lot of trouble simulating the labor system of North Korea because it is heavily reliant on two things that require a bit more brainpower than just looking at very easily modeled cost-benefit and chasing quarterly profits. Namely, voting on management and central planning. This means that the AI of every single citizen would need the skill of not only recognizing which agents are doing a bad job at managing them and deciding when it's appropriate to recall them, but also at deciding which other agents will be suitable delegates and managers. Chasing quarterly profits is a poor way to model a capitalist structure but it's at least sort of basically emulating the structure. For a socialist one it won't work whatsoever. 5. Central planning is a famously complex endeavor. It requires gathering a lot of reasonably accurate information on supply and demand, i.e. How many products are consumed and how many are manufactured. It also requires some way to estimate how much utility this consumption is providing. The flow of correct information is vital to this type of economy and information flow and the possibility of lying need to be modeled for this type of economy to work. It also requires that your agents are competent enough to design five year plans for a large and complex economy.

[@i.p.7687]

An AI can do anything of its trained for long enough in a realistic simulation, so the hard part is waiting for the AI to finish. Even if it's being told to run a society.

[@fakedungeonmaster5740]

how do we know that the simulation actually reflects reality? how do we know that the results of the simulation are applicable to reality?

[@matthewbennett7435]

At this point it does not reflect reality. In the video they state at about 3:45 "within the constraints of this simulation". For now this is an oversimplified model. The hope is this project can get further developed until it becomes a decent model of an actual economy. At which point it could be a valuable tool for deciding tax policy.

[@sssss-zk9oo]

@@matthewbennett7435 the problem isn't that humans are incapable of figuring out the optimal system. its that data collecting especially real-time data collecting is insanely difficult to do. this AI is kinda useless from the get-go to apply it to real life, but it can be useful tool for economic theories.

[@samuelwillcox5920]

So many things get left out in these simulations they can’t possible be accurate enough to be useful on a larger scale. For instance some governments look great on paper but fail to account for human error or other issues. What’s to say maximizing productivity isn’t better than equality? For instance if there is enough competition the prices of goods could be lowered to an extent that the lives of the poorest people would be just as good as the middle class today. This simulation seems blindsighted to the fact that there are positive outcomes to productivity that benefit the whole and you lose those benefits as soon as redistribution is set in place.

[@masterchief7301]

Except we have longitudinal studies that demonstrate that things like trickle down economics, which maximize for productivity, does not work.

[@samuelwillcox5920]

@@masterchief7301 “During the Reagan administration, real GDP growth averaged 3.5%, compared to 2.9% during the preceding eight years. The annual average unemployment rate declined by 1.7 percentage points, from 7.2% in 1980 to 5.5% in 1988, after it had increased by 1.6 percentage points over the preceding eight years.”

[@fuckjoebiden]

@@samuelwillcox5920 that doesnt mean the wealth trickled down to the general population. under the trump administration, the rich gained a sick amount of wealth while the poor are poorer than ever before. job creation can mean high paying tech jobs, or it can mean sub-living wage jobs whose workers still can't afford to move out of their parents' house

[@danwbeeston6146]

@@samuelwillcox5920 Top marginal tax rate in Reagan era was 50%. Today it is 37% - we're more trickle-down now and as a result growth is lower

[@danwbeeston6146]

@@samuelwillcox5920 Prior to Reagan in the "golden years" of capitalism - top marginal tax rate was 70%.

[@besknighter]

Interesting! What actually is the tax policy that it came up with? It would also be interesting to see how it would design the system if it took into account corruption by policies enforcers and citizens and their motivations behind such subverted acts. BTW, I've seen (but not read yet) a few articles that argues that optimizing productivity x equality is suboptimal, arguing that "poverty is the problem, not necessarily inequality". I'm not going to argue about this statement because I know very little about sociology and economics, but I'd like to see how would an AI design a tax policy if that was the case. Would it be different? How so? Maybe this great optimization of productivity x equality that it arrived at is already a really amazing step in the direction where it's solving poverty instead of aiming purely at inequality. It seems like it. So many questions! Can't wait to see more about this!

[@Refresh5406]

*codes "taxation is theft" into the algorithm*

[@facundogoiriz7323]

Hahaha

[@filipwolffs]

"Mr Peralta, please stop breaking into our computer labs."

[@zuluactual839]

Ahahaha n o

[@fischX]

I would be interested in how this tax scheme works and how complex / simple it is.

[@Biped]

You should read the thing. It's basically just a graph of the tax rate in one axis and the income on the other. It taxes very low incomes lower, very high incomes higher and the middle class pretty evenly. Quite simple. Of course it's only a simplified model.

[@crimsama2451]

Id imagine "free market" productivity would drop off a cliff once inequality hits a certain point.

[@stephennewcomb4277]

thing is at the upper echelon of tax brackets it is competitive businesses and millionaires are willing to move at will and a 1% difference could pay for the new house or skyscraper they relocate to in the first year. competition in a tax bracket which you don't control the other tax bracket is one thing that would also need to be taken into account. this would help simulate california to texas businesses and other countries competing against each other. that is where free market also shines in it increases the number of businesses to tax from. If you tax 10% from 100 people or 9% from 125 companies the 9% brings in more revenue. This model while is great does only suffice under the premise of fixed number and location.

[@theamazingplayer5225]

Reminds me of a book called Scythe.

[@jessebusma5339]

This simulation makes the assumption that all the agents under a state will intelligently seek the best way to optimize their earnings potential, instead of just looking for someone to blame for their current predicament.

[@Alex.In_Wonderland]

which company created this??! this is insane! i would love to see how it performs and what the output would look like if there were not JUST more, but also more complex parameters! This is soo cool!

[@matthewjoynt7170]

This is a very interesting simulation, but I think it leaves out two key components to the tax environment. Firstly, at certain level of taxing on the rich they are going either hire people to find loopholes to abuse, which is very likely if you look at the older tax policies if 70% marginal income tax and how they used loopholes to pay nowhere near that, and if it is not possible they will move there wealth to a country that will not tax them as much. It also assumes that when you significantly subsidize the lower class that they will continue to be productive. As someone who works in small business that pays minimum wage, because that is all they can afford, my personal experience with this is that they while Covid policies for unemployment did not force you to be looking for work it was near impossible to find people to fill positions. These two factors really play a key role in how strong a tax system is going to be. To be fair the video did not state the actual tax laws being implemented and maybe these have been accounted for, and to that I say we should not be striving for equality of outcome, but equality of opportunity.

[@SimplestUsername]

3:08 So basically the US Federal model just makes makes the rich poorer with no benifit to anyone else. That's a choice.

[@evilotto9200]

ai simulation confirms biases - 😊 ai wrekts personal worldview - 😠 gawds, we're an enlightened audience

[@artjonos1743]

Confirms?

[@artjonos1743]

@@JoshuaValvi1 Oblivious to context are we?

[@seanhardy_]

@@artjonos1743 They were merely being respectful for the sake of the non-native English speakers who also use this platform. I hope you don't go around insulting people because of minor miscommunications, it's not the best way to spend your time

[@artjonos1743]

@@seanhardy_ Not sure that definition is helpful to the non-native speaker, considering the video and top comment. Sounded bit condescending, mb

[@harrisonward6258]

Almost all of the advances in behavioral economics suggests that simulating tax policy with hyper rational agents working on a constraint optimization problem is actually a step backwards for our understanding of human responses to policy, not a step forward, this per David Laibson a Harvard econ professor. Sure this is fun and an interesting exercise but with out vastly more organic agents this branch of research is unlikely to do much to advance our understanding of real world optimal policies

[@petersmythe6462]

Another thing that would be nice to simulate is the human capital model. People who have less of their needs met are less productive and people who have access to resources and training *and* have their needs met are often more productive and also work in different industries. A model where the agents have a hierarchy of needs and also need to go to schools to learn skills (untrained workers would be highly unproductive) would probably more reflect a real society. So the super high efficiency of free market models would be at least somewhat reduced if they are in an environment where profit is maximized but their needs are not. An economy built on the back of illiterate farmworkers who have a nutritionally incomplete diet would for example be highly unequal, highly profitable, and relatively unproductive. Investing in your population is smarter than simply extracting the maximum amount possible from them if the goal is overall productivity in the long term.

[@Nehji_Hann]

We absolutely need this kind of change. Some official person: "It's not so simple" So you're saying you won't even try? Who cares if it's not simple? If it's possible at all we have to make it happen.

[@Ennocb]

Interesting idea and definitely worth exploring but the model is lacking many real-life parameters.

[@rickymort135]

It's a proof of concept for the RL approach but yes they need to scale up to the complexity of the simulation now

[@nzuckman]

bro do you even macroeconomics

[@forloop7713]

The AI would quickly determine that the lower the taxes the better and at this point people would reject the AI and say it's unscientific

[@robvadeberg]

If you paid attention to the video, yes, that is exactly what the AI determined, as long as you're aiming solely to optimize productivity. However, the AI also concludes that it comes at the cost of equality, and therefore it came up with another solution that offers a nice compromise between the two, that does involve taxes.

[@youtb210]

It'd be cool for our creation to take upon the role of helping us evolve. It'd be kind of poetic. Like a son raised by a loving family, who in turn takes care of them when they're old and ill, or a son who gets abused and in turn becomes resentful towards his family.... Which one will we be? I hope it's the former.

[@exedeath]

This paper already started wrong by caring about equality, what matter is how good is the quality of life of those with the lowest quality of life is, and not how their quality of life is compared to the ones with most quality of life.

[@Unethical.Dodgson]

I agree with this to an extent but that's where the balancing of equality:productivity comes in. Because the aim isn't to achieve 100% equality -- the most successful will still live relatively like kings but the overall standard of living for the middle and lower classes would be raised well beyond what most models have achieved. Which is the goal.

[@exedeath]

@@Unethical.Dodgson "Because the aim isn't to achieve 100% equality" the aim shouldnt care about equality just about improviing quality of life of those with lowest quality of life.

[@jasertio]

What do they mean by optimizing for equality?

[@bramvanduijn8086]

quote from the linked paper: "Here, equality is defined as 1 - Gini index and computed on incomes (after taxes and subsidies) at the end of an episode "

[@martinnel1347]

Can you do one comparing altruism and selfishness? For instance making two worlds, one with AI more likely to give, and one with those more likely to take in a business environment, those who share, and those who keep to themselves... and whether one or the other will end up more wealthy or equal

[@Elenthiriel]

there is a lot of parameters that we have to take in to account. For example, although the AI have found a "equilíbrium" between productivity, it does not mean that it would work in the real world, because she has to take into account for example, the level of abstraction needed to execute a job, and as we know, the higher the level the abstraction the higher might be the outcome because the higher is the investments for acquiring the skill, that ultimately will provide some level of value, so... i'm that the AI did not take it in to account

[@Chaisz3r0]

I see an important flaw in the setup. At least according to the explanation in the video, the policy maker is not also part of one of the worker groups. This means it is unaffected by the policies it creates and is not trying to maximise its own well-being. This allows it to make decisions that would otherwise go against its best interest. This limits real-life applicability.

[@francois-rozet]

Indeed, it probably "should" be like that, but we all know it isn't...

[@callumfinlayson-palmer8393]

That's why the wealthy keeps getting wealthier and the poor stay downtrodden. It's rare to see improvements and even then they are usually marginal. Getting more Apolitical people on board with socialism or just with some of our policy will go along way with pressuring and challagening the power system.

[@NicOesterby]

Good point, Chais. But on the other hand, if the tax system was actually designed by an AI, then the AI *is* unaffected by its only policies, since its an AI not a person.

[@Omegatcu]

It also doesn't solve any other metric that would be important... low vs high productive workers, competing systems that will attract workers or the very simple problem that some workers will work more hours than others...

[@callumfinlayson-palmer8393]

@@NicOesterby the can programme the AI to want the same goal as the people who it affects.