This market resolves as YES if, five years after the development of weakly general AI, wealth inequality in the U.S. has increased compared to the pre-AI baseline. Otherwise, this market resolves as NO.
Additional context
As of the first quarter of 2023:
The wealthiest 0.1% of Americans own 12.8% of the country's wealth.
The wealthiest 1% of Americans own 31.3% of the country's wealth.
The wealthiest 10% of Americans own 69.0% of the country's wealth.
The middle-class (50th to 90th percentile) own 28.6% of the country's wealth.
The remaining half of the population owns 2.4% of the country's wealth.
The current disparity between the wealthiest 10% and the poorest 50% is (69/2.4) = 28.75; i.e. the wealthiest 3,400,000 people in the U.S. own 28.75 times as much wealth as the poorest 170,000,000 people do.
The development of weakly general AI could shake up the U.S. economy significantly - but whether it will shake up the economy, and whether that shaking up will increase or decrease inequality, remains to be seen.
Fine print
I will measure the disparity between what the wealthiest 10% of Americans own and what the poorest 50% own, using the most up-to-date data I can find, on the day that this Metaculus question resolves. That disparity will set the "pre-AI baseline," and I will update this market to reflect it. Five years later, I will measure the disparity again, once again using the most up-to-date data I can find. If the disparity has increased, this market resolves as YES - otherwise, it resolves as NO (including if the disparity is the exact same, i.e. has not changed.) I will attempt to use the same data source both times, but if that's not possible, I will settle for using a different data source.
This market resolves based on the state of the U.S. economy at the end of that five-year period, not its state at any point in between.
If the U.S. stops existing prior to the development of weakly general AI, this market resolves N/A, even if the U.S. is re-formed at some point afterwards. If, after the development of weakly general AI, the U.S. stops providing accurate reports of its citizens' wealth and interferes with attempts by outside observers to measure that wealth, this market resolves YES because that's a certain sign that things have really gone to hell in re: income inequality and oligarchy/plutocracy. If the U.S. stops providing accurate reports of its citizens' wealth because AI caused the development of Fully Automated Luxury Gay Space Communism and measuring wealth is pointless, this market resolves NO instead.
The preliminary closing date is set to December 31, 2099, but it will be adjusted as necessary. When the Metaculus question linked above resolves, the closing date of this market will be changed to exactly five years after the question was resolved.
1,000I believe there is one main reason that the wealth inequity in United States will increase. First of all, the rapid development of AI like ChatGPT has already replaced some labor posts, which means the capitalists are able to decrease the costs of human resources. It also means that more people will lose their jobs and the same job will be more competitive for people to get enrolled, cause there are less jobs and more people, the requirements of all jobs will be much higher, which means the education degree is being worthless, it indirectly leads to bigger wealth inequity due to the higher education costs.
@JiaqiuHuangfu I think it's a complex issue that can't be reduced to GENERAL AI by isolating the problem, from an economic point of view, the gap between the rich and the poor is bound to get bigger and bigger as time passes (a situation that would have to exclude the government from intervening through taxes or legal policies), this is dictated by our existing economic model, in that case then the answer to the question would definitely be YES, and the answer to the question would be YES.
But in terms of technological development, the increase in productivity brought about by GENERAL AI may also expand the economic pie and create some new jobs, thus offsetting inequality.
On the contrary, the development of general AI may accelerate automation and disadvantage low-skilled workers, and since general AI is able to automate some of the tasks that are currently performed by humans, income may be concentrated on those who own the technology, and thus wealth will be concentrated on a few, however, the ultimate winner lies in government policies, for example, the government can tax and subsidize the producers of AI. general AI producers and subsidize the transition of workers, it can also subsidize major influences that are not related to general AI, such as education and welfare, etc. Considering these factors, I choose to believe that there will be a series of good policies in the future to offset this inequality, so I do not agree that general AI will exacerbate inequality in the U.S.
Source :
A report by the Economic Policy Institute estimates that algorithmic and AI displacement of human jobs could reach 15-25 million jobs by 2030. This highlights the uncertain impact of technological progress based on policy responses. ( https://www.brookings.edu/research/)
A World Economic Forum study on the impact of AI on jobs found that while there are displacement risks, human-AI collaboration could also create 58 million to 133 million new roles. ( WEF https://www.weforum.org/reports/the-future-of-jobs-report-2020)
Something interesting is that a sufficiently intelligent AI could destroy the basis of the most (I think) common argument for inequality. If the point of inequality is to incentivize people to produce lots of value, then a world with AI doing most intellectual work provides much less justification for inequality. That might be more likely to take effect in fifty years than in five, though.
My concern here is that Metaculus could well go through a long phase of lawyering and nitpicking whether a particular system meets those specific requirements, and the question won't resolve until well after any market shocks from AI progress have been processed by the economy.
I don't think it's inconceivable that GPT-8 will be writing most federal regulations in the US while still being awful at Atari games, for example. That market could easily have its resolution delayed by some criteria like that, even while it's intuitively obvious to most people that "weakly general AI" is having drastic effects on the economy.
@NLeseul That's entirely possible, yeah - but I would expect to notice significant user outcry on Metaculus if that ended up happening. In the event that they do take way too long to resolve the market, I might end up holding a poll and asking "which AI model was the first weakly general model?" and using the results from that as a starting point.
@NLeseul I don't think it's inconceivable that GPT-8 will be writing most federal regulations in the US while still being awful at Atari games, for example.
This is one of the most horrifying dystopias I’ve ever heard of.
Haven't thought too hard about this, but a few things that seem relevant:
- Over time, as long as there are not huge destructive wars/pandemics/etc, wealth inequality just tends to steadily increase, in most countries most of the time. (At least if you are measuring inequality BEFORE taxes & transfers... inequality after taxes and transfers actually looks pretty stable/good.) So maybe the baseline assumption should be that this resolves YES, before even considering AI.
- On the other hand, we are in a very hot economy right now (very low unemployment, etc) and that seems like it is helping to reduce inequality, raising wages especially of people at the bottom and probably increasing the overall amount of the economy that goes to wages (vs returns to capital). Hopefully our full-employment economy will be able to continue doing its thing for some years to come (unless we unexpectedly fall into recession).
- Current LLM-style AI seems particularly suited for replacing / complementing the jobs of highly-paid white-collar professionals -- doctors, lawyers, etc. So I would expect the economic rewards of LLMs to flow towards:
1. The bottom 50% of low-wage workers and people who do physical stuff for a living (like construction workers and care providers), who are going to benefit from doctor/lawyer type stuff getting cheaper and from the whole economy speeding up thanks to AI-powered innovation.
2. The top, like, 0.01% of startup founders and capital-owners, who are going to profit directly from the AI apps which just disrupted a bunch of white-collar fields.
A lot of those doctors/lawyers/etc are in your "top 10%" category, so the question will be how those three groups add up -- if the gains to 0.01%-tier founders/investors, minus the losses of 10%-tier white collars workers, exceed the gains to 50%-tier blue-collar workers, then this market goes YES. But I am very uncertain about this; 75% seems too bullish to me.
Question regarding definition of "pre-AI baseline"
Will it be the day that the actual Metaculus question is resolved?
Or
The date to which the Metaculus question is resolved to?
Resolution date will be the first date at which the system (subsequently judged to satisfy the criteria) and its capabilities are publicly described in a talk, press release, paper, or other report available to the general public
Ostensibly, the market will actually resolve some time after the date to which the market resolves to.
==================================
Questions regarding disparity measure:
Will this measure be depended upon a report published by www.statistia.com or some/any other data source?
Explicitly, Will the disparity measure be "a unitless metric derived by taking the ratio of % of wealth held by the top 10% of earners and the % of wealth held by the bottom 50% of earners"?
@ShitakiIntaki The day the market resolves, under the assumption that weakly general AI will only have major economic effects once people are sure it exists.
Ideally, I'll use a data source that I trust to be around five years later, so something long-established like the Census Bureau, World Bank, etc.
That is what the measure will be!