There is no simple way of discussing the economic impact of artificial intelligence. As someone who has been warning about the impacts of automation for a decade now, and trying to get people to understand that automation has already been impacting us for decades, I can tell you that this nuanced conversation basically gets boiled down to "Technology is going to end all jobs" versus "Everything will be fine." Both of those are wrong. The truth as usual, exists somewhere in the middle, and comes with a lot of "it depends" asterisks. In this post, my goal is to help people understand that although AI will get rid of a lot of jobs, that doesn't mean everyone will be jobless and forever unemployable. And that despite a market for human labor continuing to exist with AI shouldering a lot of work, the wave of job disruption is not going to be pleasant, and the path we take depends greatly on the adoption of an unconditional, universal basic income as a steadily rising AI dividend. The future is not out of our hands. It is entirely within our hands, and up to our collective decision-making to benefit or hurt us.
First, to get everyone up to speed, this is what I've previously written to explain the realities of automation. Here's a key paragraph to save you the time of reading the whole thing, although I still do recommend reading the whole thing anyway.
Decade after decade, medium-skill manufacturing/office jobs have been disappearing, and in response, the unemployed have found new employment in new low-skill service jobs. People unemployed by machines still require income, so they end up finding what they can get. At the same time, they are competing against others doing the same thing and thus people are bidding down their own wages and taking any jobs they can get in a race against the machines. This also serves to make investments in automation less attractive. As an added bonus, the jobs that are being automated are more productive jobs than most of the jobs being newly created. Cheaper human labor and an increasing number of low productivity jobs together then result in a “paradoxical” deceleration of productivity growth. Long story short, the middle of the labor market is disappearing. That’s the reality, and it’s been happening for decades.
This general trend is what seems most likely to continue, albeit potentially more focused on the upper-middle instead of the lower-middle based on early analysis of the capabilities of GPT-4 and other generative AI.
The above table is from a working paper published in collaboration with OpenAI on March 27, 2023. The paper titled "GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models" attempts to estimate how GPT-4 will impact jobs. The researchers make three different estimates to create a lower and upper bound (and even have GPT-4 itself make estimates too) and find that the most impacted will be workers with Bachelor's degrees. Those with Master's degrees will be impacted too, but not as much, and those with high school degrees or GEDs will be impacted even less than either more-educated groups.
From the abstract of the paper:
Our findings reveal that around 80% of the U.S. workforce could have at least 10% of their work tasks affected by the introduction of [Large Language Models], while approximately 19% of workers may see at least 50% of their tasks impacted... our analysis suggests that, with access to an LLM, about 15% of all worker tasks in the US could be completed significantly faster at the same level of quality. When incorporating software and tooling built on top of LLMs, this share increases to between 47 and 56% of all tasks.
Now, regarding use of the word "impact," it's important to understand that this doesn't mean that the work will be performed by AI entirely on its own, but that those jobs are most likely to use AI to be more productive. For some jobs, that's just going to mean a lot more productivity with more work done in less time, but for other jobs, that can mean fewer of those jobs.
Here's an example that I think illustrates this well: technical support jobs. Imagine a technical support call center with 100 employees. Now imagine that AI usage makes everyone more productive to the point that average wait times fall from 30 minutes to 3 minutes. For everyone involved, this is a great outcome. Customers are happy waiting a tenth of the time they usually would, all 100 workers are happy they get to keep their jobs, and the business itself is happy. In fact, the business gains market share from similar less productive call centers that haven't adopted AI yet. This even means the business actually expands and hires more workers.
Now consider what happens if the average wait times drop from 10 minutes to 0 minutes. Hitting zero means that a lot of the workers are just sitting around waiting for another call. They're so much more productive that fewer of them are needed. If the business can find more customers, it can make sense to keep all 100 workers, but if they can't, it makes sense to cut enough workers to bring average wait times up to one to five minutes. This could mean letting go of 50 workers, or it could just mean not hiring to fill positions when workers leave. As should be apparent, one of the big asterisks is demand. If demand is sufficient, certain jobs will continue to exist, whereas if it isn't sufficient, they won't.
This isn't just theoretical. The first study of real-world use of LLMs like ChatGPT has been published where a Fortune 500 company used the technology for a year. On average, customer support agents provided the AI became 14% more productive, with the biggest impact observed among the least-skilled workers who became 35% more productive. That's with AI less capable than GPT-4 and without the connection to tools, which OpenAI believes will triple productivity boosts observed from just LLMs on their own.
Now let's zero in on the example of one person in particular. Let's call him Jim, and let's say Jim's company has incorporated AI to become 25% more productive. In Jim's case, there is not sufficient consumer demand for it to make sense to keep Jim, so Jim's employer lets him go to save labor costs and increase their profits.
Jim is now unemployed. He applies for unemployment insurance. If he lived in Florida, he would have three months to find a new job, and would get a maximum of $1,100 a month during that time. If he lived in Washington, he would have 6 months to find a new job, and would get a maximum of $3,376 a month during that time. If he was a freelancer, he'd get no time and no money as 60% of the unemployed get from unemployment insurance. Let's say he gets four months and $1,500 a month. That's how much time he has before his income falls to zero.
There are mostly two ways Jim's future can go. He will learn how to use AI and find a job that pays as much or more than he was previously, or he will find whatever he can get which ends up being a job that pays less than he was earning previously. Given existing trends, and just how much more productive AI can make most workers, the result with the most likelihood is less income for Jim.
It is also not at all a sure thing that Jim will find a job within four months. There's a good chance so long as a lot of employers are looking to fill positions as there are right now, but given the Fed's desire to increase unemployment, and given the ability of AI to enable ten people to do the work of fifteen, Jim may have a harder and harder time finding the next job, as employers decide they just don't need as many people as they once did without AI.
If Jim cannot find a new job within four months, he falls into poverty. If he chooses to then seek government assistance, he needs to apply for what he's eligible for, which is likely to just be SNAP (aka food stamps). That's about $250 a month for a maximum of three months. So that's three more months free of hunger, and most likely Jim will need to rely on his local food bank too.
If Jim can find a job within three more months, he'll be okay, but again, he will mostly likely find a job that pays less money than he was before, even though it's also possible he'll find an even higher paying job. But what if he doesn't find any job? He's out of luck and joins the 13 million Americans living in poverty who get nothing from the federal government to help them.
Jim isn't about to starve though, so he begins to steal in order to survive. He steals food and steals stuff to sell for food. Taking this route, a few more possibilities open up. He either eventually finds a job (which is even harder now than ever for him), or he gets arrested and imprisoned where he receives free room and board at a public cost of about $5,000 a month, or he ends up dead thanks to poverty now being the fourth largest cause of death in the US among those age 15 and older.
That's how far Jim can fall. He can fall to his death. It's not the most likely scenario, but it is entirely possible. It's also entirely possible he can end up upskilled with even more income with a job he enjoys more. Random chance will play a large role in Jim's future, whether he or anyone else recognizes it.
Now let's redo this entire scenario in another universe where universal basic income exists as a floor underneath Jim.
Jim-2 works for a company in a parallel universe that has incorporated AI to become 25% more productive, just as in Jim-1's universe. But, because UBI also exists, and the bottom 90% of the country has a disposable income boost, Jim-2's company sees a sufficient amount of consumer demand to not let him go. And that's the story of Jim-2.
Just kidding, that's a pretty short story. So let's assume that even with the higher amount of demand that comes from UBI to create and fuel jobs through people spending their UBI, that Jim-2 still gets let go.
Jim-2 is now unemployed, but because of UBI, his income falls to the lowest possible limit of $1,200 a month. If he lived in Florida, he'd have three months to find a new job and would get a maximum of $1,100 a month during that time, bringing his total income to $2,300 with UBI. If he lived in Washington, he'd have 6 months to find a new job, and would get a maximum of $3,376 a month during that time, bringing his total income to $4,576 a month with UBI. If he was a freelancer, he'd have only the UBI of $1,200 a month, and an unlimited time to find a new job without ever losing that income floor. Let's say he gets four months and $1,500 a month, the same as Jim-1, for a total of $2,700 a month with UBI. He has four months with $2,700 a month before falling to $1,200 a month.
There are still mostly two ways Jim-2's future can go, although a couple other options also become more possible with UBI. He will either learn how to use AI and find a job that pays as much or more than he was previously, or he will find whatever he can get which ends up being a job that pays less than he was earning previously. Given existing trends, and just how much more productive AI can make most workers, the result with the most likelihood is still less income for Jim-2, albeit with his UBI serving to further boost his income. However, two other realistic options include starting his own business or pursuing unpaid work, perhaps as a volunteer in his community planting trees to fight climate change.
It's still not at all a sure thing that Jim-2 will find a job within four months, but thanks to UBI, he has more time to upskill, or relocate, or go back to school or do whatever else he needs to find his next job. There's a good chance so long as a lot of employers are looking to fill positions as there are right now, but given the Fed's desire to increase unemployment, and given the ability of AI to enable ten people to do the work of fifteen, Jim-2 may have a harder and harder time finding the next job, as employers decide they don't need as many people as they once did without AI. But also thanks to UBI, there are more job openings than there would otherwise be thanks to higher demand for goods and services than there would be without UBI.
Now, if Jim-2 cannot find a new job within four months, he falls to $1,200 a month. That is the absolute minimum. If he chooses to then seek extra government assistance, he needs to apply for what he's eligible for, which is likely to just be SNAP (aka food stamps). That's about $250 a month for a maximum of three months. So that's three months of $1,450 a month instead of $1,200 which Jim-2 decides isn't worth the trouble. He's fine without the extra money for food, and is also fine without the assistance of food banks. He declines both options.
If Jim can find a job, he'll be back to having plenty more than $1,200 a month, with his new paychecks being additional to his UBI that's always there. But what if he doesn't find a job? He will never have less than $1,200 a month. Unlike Jim-1, he never experiences poverty as defined by the current federal poverty guidelines.
Also unlike Jim-1, Jim-2 never needs to steal in order to survive. He never ends up arrested, or imprisoned, or dead of poverty. Instead, he eventually finds another job, and eventually brings his income back up to $70,000, but he pays $7,400 less in taxes than he would without UBI in place, because despite the additional $7,000 in taxes as a result of a $14,400 UBI, the remaining $7,400 functions as a tax credit. In Jim-2's universe, the bottom 90% have a lower tax burden than in Jim-1's. Only the top 10% - the owners and primary beneficiaries of AI - see an increased tax burden.
Clearly, Jim-2 is far better off than Jim-1. Poverty is not one of the possibilities for Jim-2. He is protected from that. It's not allowed to happen. Death therefore is also far less likely to happen. Jim-2 is more likely to end up upskilled than Jim-1. He's more likely to get a new job, especially if his universe makes the active attempt to make it more possible through programs like free skills training, or free education, or investing in the creation of green jobs. Chance will still play a large role in Jim-2's future, but the odds are far more in Jim-2's favor than Jim-1's.
Please note that Jim-2's universe is not all sunshine and rainbows. UBI does not automatically make a world where everything is awesome. It just clearly creates a better world where poverty doesn't exist, and people have a cushion under them at all times that makes it easier for them to find the next thing. It's a world with less stress, more security, less crime, better health, more entrepreneurs, and more freedom for people to determine for themselves what work it is they wish to be doing, and at what price they will do it for.
Now if it were up to me, another detail of Jim-2's UBI would be that it grows as productivity grows, or as inflation grows, whichever is higher. As AI makes us all more productive, that increase in national productivity should flow equally to all of us as a dividend. Over time, and as AI improves and becomes more widely adopted, $1,200 a month in UBI should become $1,500 (in 2023 dollars), and then $2,000, and then $3,000, etc. And taxes should be utilized as a tool of inflation management and inequality reduction, so that over time, inequality continues to decrease and prices remain stable year over year.
Without UBI, and without more taxes at the top, inequality will grow as AI advances. It's not that everyone will end up without a job, but it's entirely possible that perhaps 120 million people are employed instead of 165 million, and that it creates a deflationary environment where less income means less spending which means fewer jobs which means less income, which means less spending, in a vicious cycle. That unemployment outcome also depends on other things besides UBI, like how long the workweek is. Going to 4-day weeks can mean more people working shorter weeks, which can mean 165 million people employed, just as now, but with more leisure time. With UBI, that can mean working part-time jobs with full-time income.
These are the kinds of choices we need to make that lie before us.
There is no future where there are zero jobs. There is no "end of work." As long as people have money to spend, there will be people willing to do something for someone else for money. If money concentrates into fewer and fewer hands, that's how people lose the ability to earn income. To explain why, let's now talk brunch.
The Jazz Brunch Economy
A big fear that exists regardless of the existence of universal basic income is the belief that AI and robots will just be so good at doing work, that there's just nothing to do, let alone anything to do for money that technology can do better. This brings me to what I think of as the Jazz Brunch Economy, because I'm from New Orleans, and it provides a unique example.
Let's pretend that AI does indeed eliminate half of all jobs, and instead of lowering the work week to 20 hours from 40 so that everyone can still find a job, we instead have half of workers working 40 hours and the other half unable to find jobs. Not the brightest idea, but let's say it happens. Does this mean there's no way of earning money for half the population? Are they out of luck?
In New Orleans, there's something called a "jazz brunch." It's where you can go to eat brunch somewhere, and there's a live band playing. We've all eaten in restaurants that lack any music, or have recorded music playing. Recorded music is a form of automation. So if the technology exists to listen to recorded music at a restaurant, why ever hire a band to play? Because people enjoy it. It's a special experience to go to a jazz brunch. Live music will always be something we enjoy despite the ability to listen to recorded music. When it comes to jazz brunches, it's something people specifically look for, because it's not the norm. They choose which restaurant to eat at based on whether or not a band will be playing there. It's how some restaurants are able to bring in customers that would otherwise choose to eat elsewhere. It's an experience that even draws tourists to New Orleans in the first place. That's what humans will always be able to provide even when intelligent machines can do our work for us - the unique experience of being human.
Another example from New Orleans is something called "house floats." This was born during the pandemic when we skipped having a Mardi Gras celebration in 2021. A lot of work goes into building the parade floats every year, which requires a lot of workers. The lack of Mardi Gras that year left a lot of people unemployed. A response to that was the creation of the house float, where people paid artists to decorate their homes like floats. That employed a lot of artists and the demand for it came out of nowhere. One day house floats weren't a thing, and the next day they were. People just decided they wanted to pay someone to creatively decorate the outside of their homes. It was a new job that was born in 2021 - house float artist.
If suddenly everyone, everywhere across the country, wanted the outside of their homes decorated by local artists, there would be a lot more people getting paid to do that suddenly much more common job. It's all about customer demand.
I think these examples illustrate both the unique value-add that humans have despite technology, and how new jobs can indeed be created, just by some people wanting to pay other people to do something both the payer and the payee enjoy. This all depends of course on people having money to spend.
It's possible that a kind of "Made in USA" label will come into existence for human-based goods and services. Something like "Made by Humans" or "Human-Made" will become something that consumers specifically look for despite cheaper automated options existing, just as people shop for stuff made domestically instead of internationally, or shop organic even when it costs more to make those choices. Because that demand seems likely to exist, those jobs seem likely to exist, because consumer demand is what ultimately creates jobs.
It is possible, and quite likely even, that many jobs will stop existing to the degree they currently do. They'll go away, or there will be fewer of them, or they will require new skills. Some of the people who had those jobs, if they don't find newly created jobs, if they have basic income, will decide to pursue something very human, like playing live music. Venues that don't currently offer live music will start offering live music, and people who enjoy live music will be willing to pay a bit extra to enjoy the experience of that live music. Those jobs will exist because of consumer demand. Something like house floats will pop up out of nowhere too, where something we never or rarely paid people to do before will be something we start paying people to do, just because we want to pay someone to do it, and because someone wants to be paid to do it. We can't really imagine what those things will be yet, but I'm sure you can think of some of your own examples if you really stop and think about it. They happen all the time.
People really don't enjoy doing nothing, and people will always enjoy paying others to do something they can't do as well, or don't want to do themselves, and that will remain true to some degree even when technology can do so much more for us. Even right now, there are already new examples, of paying someone else to use AI to accomplish something, because some people are already better at using AI than others, and because time will always be limited. Because time is limited, there will always be a reason to just pay someone to do something for us, even if the tools exist for us to do it ourselves using AI, as long as people have money to spend.
That's the one neat trick. We have to make sure everyone always has money to spend if we want to make sure that people can always earn additional money doing some kind of work, or afford to do some kind of work they enjoy for free. Besides volunteering, it's also a lot easier to start a business to create new jobs if you have an income underneath you to support you, and the people around you also have an income to be your customers.
Our collective economic future is up to us. We can choose the path where we all equally share in the massive wealth generation that AI makes possible (that we all made possible together with our collective data) by greatly increasing our collective productivity, and where we all always have the money to spend to pursue new interests, paid or unpaid. Or we can choose the path we're currently still on, where AI makes life harder for a large percentage of the population while massively enriching a relative few, and where jobs really do become harder to come by because spending power has been concentrated in too few hands to provide sufficient overall demand to fuel our economy.
Which path do you prefer?
Do you want more content like this? Please click the subscribe button and also consider making a monthly pledge in support of my perpetual UBI advocacy.
Special thanks to my monthly supporters on Patreon: Gisele Huff, Haroon Mokhtarzada, Steven Grimm, Matthew Cheney, Camilo Riviere, Katie Moussouris, Tricia Garrett, Zack Sargent, David Ruark, Larry Cohen, Liya Brook, Frederick Weber, John Steinberger, Laurel Gillespie, Dylan J Hirsch-Shell, Tom Cooper, Michael Tinker, Judith Bliss, Robert Collins, Daryl Smith, Joanna Zarach, ace bailey, Daragh Ward, Albert Wenger, Bridget I Flynn, Peter T Knight, David Ihnen, Max Henrion, Elizabeth Corker, Gray Scott, Gerald Huff, Albert Daniel Brockman, Michael Honey, Natalie Foster, Joe Ballou, Chris Rauchle, Arjun , chris heinz, Pavel S, Zachary Weaver, Herb, Justin Seifert, Jodi Sarda, Rosa Tran, Deanna McHugh, Ryan Ash-Miller, miki, Adam Parsons, bradzone, Lee Lor, Akber Khan, John Sullivan, Team TJ, Yang Deng, Yan Xie, Marie janicke, engageSimply - Judy Shapiro, Phuong Truong, Garry Turner, Tim, Warren J Polk, Timothy P O'Connor, Jeffrey Emmett, Stephen Castro-Starkey, Oliver Bestwalter, Kev Roberts, Walter Schaerer, Loren Sickles, anti666, Eric Skiff, Thomas Welsh, Kai Wong, and Laura Ashby. Become a monthly patron to see your name here too.
UBI Guide Newsletter
Join the newsletter to receive the latest updates in your inbox.