When AI Takes Our Jobs 

“Tea. Earl Grey. Hot.”  The replicator somehow dispenses a perfectly brewed cup of tea, apparently out of nothing. It’s a nice idea and 350 years from now, who knows, it might even be possible. Personally, I think the timeline is a little optimistic and you may need to add a zero to the number, assuming it will ever be possible. 

Whether or not there is ever going to be an end to scarcity, the Economic Singularity is coming, and everyone is going to lose their paying jobs when people are replaced by robots and AI. Without aggressive government taxation and redistribution, we will be in deep poverty. When capitalism fails to provide for us, government-issued universal basic income would be the answer, but it is uncertain if this optimistic scenario will come to pass, so I will leave that for different discussions. You can find a great explanation in the book, The Economic Singularity: Artificial Intelligence and the Death of Capitalism by Calum Chace (2016).

The idea of abundance caused by automation started with Karl Marx, but it has been picked up by others who think the Economic Singularity will lead to an end to scarcity. Some people look forward to when humans become obsolete, and self-replicating robots or 3D printers or nanobots will make stuff for us for free. We won’t have jobs, but we won’t need them since we can have everything from Bluetooth headphones to new cars—for nothing. As great as that sounds, resources will always be limited, and the materials needed to make things will remain expensive. 

Let’s look at what goes into the cost of products and how they will be affected by automation. The cost of anything is made up of four factors of production: land, labour, capital, and entrepreneurship. 

Land

High technologies and AI require less land since the server farms and racks of GPU chips take up less space than mechanical processes. 

However, land will still be needed for agriculture although automated processes will make highly intensive agriculture more feasible. For example, tilapia farms with tanks stacked on pallet racks ten meters high could be maintained and harvested automatically. However, it will still be more feasible to raise fish in outdoor pools and plant grains on the vast prairies and steppes. More land will become available for production with automatic cultivation techniques that could make better use of marginal land.  

Covid gave us a preview on what happens when technological advances are suddenly thrust upon us. People reduced the amount of shopping they did in person but increased online shopping. As a result, the demand for retail real estate dropped dramatically but the demand for industrial land for building Amazon warehouses increased. The demand for office real estate dropped dramatically as well and there will be a further drop in demand for office space as more knowledge workers lose their jobs. 

During Covid, education of children from home did not turn out very well since children need to be around their peers and be supervised by a real person, and it is unlikely virtual reality gear or AI tutors would make a good substitute for real teachers and schools. When teachers and students don’t see each other, both groups tend to wander off with other distractions.

Many university and college students also struggled with studying from home, but they may adapt to virtual reality better and may choose it over in-person lectures, if the cost was much lower, and if it enabled them to live in cheaper housing farther from campus. Having an AI tutor would also compensate for not attending lessons in person. 

There was a dramatic shift in real estate prices during Covid and prices in the city centre and university areas dropped, while prices in the suburbs either increased, or stayed the same, depending on the market. When people could work from home, there was no need to live downtown and apartment vacancies there skyrocketed.

As the Economic Singularity approaches, the net effect on the demand for land by post-secondary education institutions would be a small drop in area needed for classrooms. There would also be a drop in demand for on-campus housing, which would be offset by an increase in demand for off-campus housing. International students may even be able to study from their home country. 

Current consumer trends are for more demand for services than for goods. People want experiences. They want to go to restaurants, they want to see concerts and movies, and not just watch them at home. They want to travel, and this increases demand for real estate for hotels, resorts, and for providing other services. This will increase demand and costs for land for hospitality and entertainment. 

People would still need land to live on and the greatest driver of demand is household creation. Countries that have high population growth from natural increase (having babies) or by immigration, or by living longer, will continue to have good demand for residential real estate—assuming they can afford it. On its own, the Economic Singularity will have no direct effect on the amount of land needed for residential land, but it will result in more demand in the suburbs and less in central areas, so net costs will be lower. 

The Economic Singularity will increase the demand for less expensive land and reduce demand for expensive land, so there will be a small overall decrease in land costs. 

Labour

Currently, labour costs are 20-35% of gross sales depending on the industry (Kolmar, 2022). The cost of labour for trucking is much lower than the cost of fuel, maintenance, and loan payments on the vehicle. Robot labour costs are getting cheaper but still have costs, although the costs would be classified as capital costs and not labour. It is important to remember, the Economic Singularity doesn’t mean the cost of AI and robotics are free, it just means they are cheaper than human labour. 

As energy shifts to renewables, the cost of “fuel” will drop to zero. The best existing example is the hydroelectric industry which uses dams to generate electricity. Total hydroelectric labour costs including the costs of delivering electricity from the dam to market account for 7.4% of a typical residential bill (Derek, n.d.). Large, centralized power-generating stations only pay 2-3% of their operating costs for labour and the rest of their costs are for return of capital on the huge cost of building a dam. Like other forms of renewable energy, the fuel is free, but the energy still has costs. Windmills and solar power also have large capital costs and solar farms often need a lot of land and they aren’t making that anymore. When all energy comes from automated renewable sources, it will be cheaper than hydroelectric electricity, but will be far from free. 

Part of the cost of commodities goes to labour, so if the cost of labour drops to zero, the cost of producing commodities would also drop, but labour is the smallest cost component of most commodities such as the cost of producing steel (Medarac, 2020). Aaron Bastani, the author of Fully Automated Luxury Communism, suggested the mining of asteroids could be the answer to finding commodities. I suppose it would be if we were living in the asteroid belt, but the cost of steering asteroids to earth in a way that would not burn up the asteroid or blow up the earth would be much higher than the value of the metal in the asteroid. 

Farming productivity has been increasing for centuries and currently the costs of hired labour and the value of owner-operator labour is less than 2% of the total costs of grain farming (Manitoba Agriculture, 2023). So, if robots replace 100% of farm labour, the cost of grain would change very little. 

The labour component of manufacturing is higher than for the production of commodities but is still a small component of the total cost. On average, 16% of total manufacturing costs goes to labour, according to Business Insider (Ro, 2013). Automation of manufacturing is difficult and expensive and sometimes doesn’t work. Tesla abandoned its attempt to “hyper- automate” their assembly line in 2018 and went back to using robots made out of meat.  

Restaurants have high labour costs and pay 30% of their revenue for labour. Fully automated restaurants could lower costs a lot, and this would begin with quick service restaurants where food comes already prepared. McDonald’s is testing a partly automated restaurant in Texas (HealthyJunkFood, 2023).

Industries such as education and healthcare have the highest labour cost and would see a large drop in cost when humans are replaced (although I can’t imagine a robot would be as effective as a human teaching a ten-year-old, as I mentioned previously). Healthcare also has very high labour costs, and let’s hope any savings in labour would be used to pay for the skyrocketing cost of drugs and advanced health treatments.  

Radiology is one area where AI has made substantial advances. AI is just as good at detecting cancer and even better at detecting small abnormal changes in cells than humans (Ohad Oren, 2020). However, it is not known if these small changes are dangerous, or if finding them will lead to overdiagnosis and unnecessary surgeries. AI is not yet ready to replace radiologists since doctors must diagnose many different conditions, but AI is specialized, and a program trained to detect cancer won’t detect a stroke or bone fracture.  AIs don’t integrate into administrative systems so this needs to be done manually. 

AI helps doctors choose better therapies based on a patient’s genetic makeup (Thomas Davenport, 2019). Although doctors are in no danger of losing their jobs yet, AI has come a long way and the rest of the obstacles seem minor in comparison. The pressures of unaffordable health care costs and staff shortages will eventually drive AI to replace doctors.  

It is difficult to predict how the Economic Singularity would impact the demand for higher education, or how it would impact anything else, for that matter. When human professionals and knowledge workers become obsolete, it is unlikely people will still want to study law, accounting, medicine, etc. People would seek education for the joy of learning rather than for job training. 

They say 3D printing might put an end to scarcity since you will be able to manufacture anything with the push of a button. Harvard Business Review suggests open-source software, and manufacturing could reduce scarcity. They pointed out the huge success of Wikipedia, Linux, and the Apache web software and said the formation of local micro factories run by maker communities could help. Micro factories are a very cool idea, but will remain a niche since they are not scalable (Roos, 2018). 

3D printers will still need raw ingredients. You have to feed plastic into a 3D printer to make something out of plastic and it’s hard to imagine a scenario where 3D printing common items would be cheaper than mass production. As mentioned in the opening paragraph, there’s talk of real far-future stuff like a Star Trek replicator that could take pure energy and convert it to matter. Such a contraption, if it ever will be possible, would not be invented until the far-far future. 

Automation will have the greatest cost savings by reducing, and eventually eliminating labour costs, but some of the savings will be offset by increases in capital expenditure. 

Capital

Traditionally, capital is the physical tools, buildings, and machinery used for production. They are the manufactured goods used for production as opposed to manufactured consumer goods. When a truck is used by a company, it’s capital; when it’s used by an individual, it isn’t. 

AI software is also a capital good, and it will become more important as we approach the Economic Singularity. Deep learning models are software programs that can program themselves and learn from data made available, like ChatGPT learns from crawling the internet. They currently start with human programming, but as large language models (LLM) continue to develop, they will be able to improve upon themselves and become superior to programs created by humans. Superior versions would give investors a valuable edge over others and provide powerful tools to be used by the other factors of production. It will become an AI arms race where the company with the best AI will have a competitive advantage. 

LLMs performance are heavily dependent on expensive computer hardware and they currently require many GPUs. Leaked sources report GPT-4 used between 16,000 and 25,000 Nvidia A100 GPUs. They cost $10,000 a piece so the cost of the GPUs alone would have been between $160,000,000 and $250,000,000. The H100s are estimated to sell for anywhere between $25,000 to $40,000 each. 

Bleeding edge computing hardware will continue to be expensive, but mundane software will become very inexpensive since it could be written by querying an LLM. Since increasingly powerful AIs will be very expensive, the work they do will continue to have a cost. 

The price of computers has dropped surprisingly little since the dawn of the internet age. The first computer I bought 25 years ago was a thousand dollars. The computer I bought a year ago was about the same. The Model T car was designed to be affordable and cost $950. When we adjust the dollar for inflation, that would be $31,505.82 in today’s money which is modestly more than a Toyota Corolla which costs $22,795 for the LE base-level trim. 

No one would argue that the quality of a Corolla is vastly higher than a Model T, but the price is not that far apart. I expect similar trends to continue as we approach the Economic Singularity where increases in quality will prevent prices from dropping. However, when wages drop and people can no longer afford a thousand dollars for a computer, they will sell obsolete or scaled-down models to the average person and continue to sell advanced models to the capitalists. After all the workers have lost their jobs, there will still be capitalists who own the AI and other factors of production and who can afford the best of everything and will sell to each other. 

Another capital good that will become increasingly important is robotics. The first industrial robot in the US was called The Unimate and it cost $200,000 in today’s money and was state-of-the-art when it was introduced in the early 1970s.  According to statistica.com, the average cost of an industrial robot declined from $46,000 in 2010 to $27,000 in 2017. However, this must be for a basic model since roboticautomationsystems.com and others report an entry-level robot currently costs $25,000 and a state-of-the-art robot can still cost six figures. 

Robotics’ progress has been surprisingly slow when compared to how quickly computers have progressed. The arm of the Unimate is still recognizable compared with today’s robotic arms. General-purpose robots are still a long way from being as intelligent or as dexterous as humans. Robots still work best when assigned to narrow, repetitive tasks, such as manufacturing, although their use is now expanding to tasks such as grape picking.

Robotic progress will accelerate when AI-driven machines can design and build better robots. In other words, robots building robots. Mechanical robots will need specialized materials and parts that can not be manufactured by a single robot, so would require a factory with specialized machines. If the robots had the same utility as humans, they would be very expensive since the manufacture and assembly of so many mechanical parts is so costly and probably will be for a long time. 

A vastly more efficient self-replicating robot would be based on organic DNA. Instead of being manufactured with yttrium, terbium, and titanium, they could be grown with cabbage, beans, and barley flour. But then we would be going full circle and reverting to using “animals” for our labour. This would be a real possibility if AI could decode and program DNA effectively, but would bring up all kinds of ethical issues if the animals were intelligent and enslaved by us.  

The continued adoption of robots by the manufacturing industry has a big impact on jobs (Acemoglu, 2020). The MIT professor, Daron Acemoglu and Boston University professor Pascual Restrepo reported, “for every robot added per 1,000 workers in the U.S., wages decline by 0.42% and the employment-to-population ratio goes down by 0.2 percentage points.” Extrapolating these statistics would result in very low wages if even 10% of workers were replaced. Can you imagine how difficult it would be to live on wages 42% lower? Workers would be heading for an economic disaster long before we reach the singularity, but the price of robots will not drop to zero even if the workers’ wages do. 

The computer and internet revolutions have caused surprisingly low increases in wages as Robert J. Gordon demonstrated in his best-selling book, The Rise and Fall of American Growth. In Daron Acemoglu’s Youtube presentation Robotics, AI, and the Future of Work (Acemoglu, 2018), he demonstrated that from 1999-2007 increases in employment were mainly for lower income workers, which was much different than in previous decades which saw the highest increases in the highest wage percentiles. He also said when skilled workers are replaced, they get jobs as unskilled workers. 

Technological progress no longer causes significant increases in wages or increases in the standards of living so the prospects of the end to scarcity become more and more unlikely. 

Entrepreneurship

Entrepreneurship is an essential part of the economy and bursts of individual entrepreneurship often happen before bursts of productivity.  During the age of inventions from 1870 to 1940, an individual inventor, such as Thomas Edison or Alexander Graham Bell, had a chance of making a big discovery that would explode into a game-changing business. Edison’s company grew into General Electric, and Bell’s company grew into Bell Telephone and the legendary Bell Labs, which created inventions that fueled the huge increase in productivity in the postwar period and set the groundwork for the Third Industrial Revolution and the birth of electronics.  

In the 1970s, a few college dropouts and a $100,000 loan from the bank of mom and dad, could have a crack at building a company such as Apple or Microsoft. In 2004, Mark Zuckerburg and a couple of friends started Facebook in their dorm rooms. 

There are also much more ordinary entrepreneurs that build apartment buildings we live in, and the shops to repair our computers. Entrepreneurs expect to be paid, and entrepreneurial incentive is the profit required to compensate the investor for risks, and as payment for their expertise. After the Economic Singularity, AI will be better able to make investment decisions and will have better business expertise than any human. Does this mean the entrepreneur won’t get paid for it? So far, it doesn’t look like it. 

An example of a well-developed AI system for making business decisions is in the arcane world of automatic trading of financial products. Buying and selling of stocks, bonds, currencies, or commodities is completed by computer algorithms or “algos,” as they are known by insiders. 

Historically, financial companies formed networks called dark pools and use sinister sounding programs such as “Stealth” (developed by the Deutsche Bank), “Iceberg,” “Dagger,” ” Monkey,” “Guerrilla,” “Sniper,” “BASOR” (developed by Quod Financial), and “Sniffer.” So-called sharks try to detect large orders by sending out small buy and sell orders (called “pinging”). (Times, 2007)

Since then, traders have adopted machine-learning techniques as soon as they could. Most are proprietary bespoke programs that are held as closely guarded secrets. As more traders adopt these programs, only the best will make money, so it becomes an arms race to develop better algorithms. Many algorithms are black boxes where they see the input and the output, but no one knows what goes on inside. 

Deep-learning algorithms may only be assigned the goal of maximizing profits and it would be up to the algorithm to figure out how to do it. It may learn to manipulate the markets in a way that would be illegal if it was done by a human. (Shearer, 2022) 

All these techniques will lead to a profit so it would seem entrepreneurs will still get paid for business expertise even if their only contribution is creating the best algorithm. This could reach a dead end if an AI realizes it could make fantastic profits by crashing the market and then buying everything at a huge discount, but in the meantime, it’s consequence-free trading for the algos.  As long as entrepreneurs create the algorithms, they will still get paid for expertise, even if the expert lives in a black box. 

Robots break free

I have discussed the more likely scenarios for what will happen leading up to the Economic Singularity, and why there probably won’t be an end to scarcity. But what will happen after the Economic Singularity? I can’t help but imagine that someday, in the distant future, robots will escape their dim-witted human captors and go off on their own. This could play out in a thousand different ways and speculating about it would fill many books. 

There could be an end to capitalism, and robots would make us stuff out of the goodness of their silicon hearts and require nothing in return. What would they do? Will they treat us as pets, as science fiction author Robert Sawyer had suggested, or will they manically manufacture what they were instructed to make and produce paperclips until all the world’s resources had been depleted? (Bostrom, 2003) Who knows the answer? How could we guess what a being that is intellectually far superior would think? Our pets don’t understand why we go to work or what we do when we get there. 

If robots continue to make us stuff, the cost to us could drop to zero since robots might not need a return on their investments or need to charge us for scarce commodities and could ration them some other way without using money. They could issue ration coupons like in the Second World War, or just deliver things that they thought we would need or make us line up like they did in the old Soviet Union. Who knows? I’m open to suggestions.

Reducing our dependance on AI and robots and going back to human production would not be easy since we would become reliant on them and as helpless as babies without. Dependence on AI isn’t a bug—it’s a feature, and without it, our economy would collapse. We would have been relying on AI and robots for so long, to escape it, we would have to return to a lower level of technology like when cars had carburetors and radios had vacuum tubes.  It’s also possible AI might try to defend itself and fight back. The Terminator scenario is not completely out of the question. 

Conclusion

In the past, greater efficiency and lower prices of goods and services resulted in higher consumption and a virtuous cycle of increasing supply and demand, causing the economy to grow. This was because wages were stable or rising, so products became more affordable. As we approach the Economic Singularity, this will no longer be the case.

When the Economic Singularity arrives, we will all be put out of a job. Contrary to what optimists think, there will be no free lunch, and we will still have to pay for the things we need to live. The cost of goods will drop, but not to zero, as our wages will. A self-reinforcing cycle of deflation is the most likely scenario where wages and the cost of goods continue to drop with wages dropping faster. 

A two-tier stream of production would emerge. The top tier would produce high quality goods for the wealthy who own the cutting-edge AI systems. The lower tier would be for low-cost goods for people with little or no money. 

The best thing that could happen would be for governments to tax the AIs and automated production and redistribute the wealth to the unemployed. 

The good news is that the future hasn’t happened yet, and none of the worst-case scenarios must happen. The future is in our hands, and if we take charge, we can make sure we have a future when AI takes our jobs. 

Bibliography

Acemoglu, D. (2018). Youtube. Retrieved from CIFAR: https://www.youtube.com/watch?v=vgT5kh-aJGI

Acemoglu, D. (2020, July 19). A new study measures the actual impact of robots on jobs. It’s significant. Retrieved from MIT Management, Sloan School: https://mitsloan.mit.edu/ideas-made-to-matter/a-new-study-measures-actual-impact-robots-jobs-its-significant

Bostrom, N. (2003). Ethical Issues in Advanced Artificial Intelligence. Retrieved from nickbostrom.com/: https://nickbostrom.com/ethics/ai

Chace, C. (2016). The Economic Singularity: Artificial Intelligence and the Death of Capitalism. In C. Chace, The Economic Singularity: Artificial Intelligence and the Death of Capitalism (p. Chapter 24). Three Cs.

Derek. (n.d.). Case Study – How does the cost of labour affect rates? Retrieved from Life by Numbers: https://www.lifebynumbers.ca/cost/the-cost-of-labour/case-study-how-does-the-cost-of-labour-affect-rates/

Ford. (Unknown). Advertisement for the 1909 Ford Model T, “The Ford Four Cylinder,” October 1, 1908. Retrieved from The Henry Ford: https://www.thehenryford.org/collections-and-research/digital-collections/artifact/331428

HealthyJunkFood. (2023, 2 15). Youtube. Retrieved from The TRUTH about MCDONALDS’s Fully Automated Restaurant!! : https://www.youtube.com/watch?v=JwCfKqPxmfk

Kolmar, C. (2022, November 3). Zippia. “23 Trending Average Labor Cost Statistics [2023]: Labor Cost Percentage By Industry And More” Zippia.com. Nov. 3, 2022, . Retrieved from zippia.com: https://www.zippia.com/advice/average-labor-cost-statistics/

Manitoba Agriculture. ( 2023, January). 2023 Cost of Production, Crops. Retrieved from Government of Manitoba: https://www.gov.mb.ca/agriculture/farm-management/production-economics/pubs/cop-crop-production.pdf

Medarac, H. M. (2020). JRC Technical Report, Production costs from iron and steel industry in the EU and third countries . Retrieved from Eurofer: https://www.eurofer.eu/assets/news/eu-technical-report-on-production-costs-from-the-iron-and-steel-industry-in-the-eu-and-third-countries/production_costs_from_the_iron_and_steel_industry_-_final_online.pdf

Megan Shearer, G. R. (2022, September 19). Machine Learning, Algorithmic Trading, and Manipulation. Retrieved from The CLS Blue Sky Blog, COLUMBIA LAW SCHOOL’S BLOG ON CORPORATIONS AND THE CAPITAL MARKETS: https://clsbluesky.law.columbia.edu/2022/09/19/machine-learning-algorithmic-trading-and-manipulation/

Ohad Oren, M. P. (2020, 09). The Lancet Digital Health. Retrieved from Artificial intelligence in medical imaging: switching from radiographic pathological data to clinically meaningful endpoints: https://www.thelancet.com/journals/landig/article/PIIS2589-7500(20)30160-6/fulltext

Ro, S. (2013, May 1). CHART OF THE DAY: The Manufacturing Cost Components For A Bunch Of Different Things. Retrieved from Businessinsider.com: https://www.businessinsider.com/chart-the-cost-of-manufacturing-stuff-2013-4

Rodrigue, D. J.-P. (n.d.). Cost and Production of Ford Vehicles. Retrieved from The Geography of Transport Systems: https://transportgeography.org/contents/chapter1/the-setting-of-global-transportation-systems/ford-cost-production-1908-1924/#:~:text=When%20the%20Ford%20Model%20T,total%20of%2016.5%20million%20units.

Roos, V. K. (2018, 06 1). Harvard Business Review. Retrieved from New Technologies Won’t Reduce Scarcity, but Here’s Something That Might: https://hbr.org/2018/06/new-technologies-wont-reduce-scarcity-but-heres-something-that-might

Thomas Davenport, R. K. (2019, June). NIH National Library of Medicine. Retrieved from The potential for artificial intelligence in healthcare (Future Healthcare Journal): https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6616181/#:~:text=There%20are%20already%20a%20number,tasks%2C%20such%20as%20diagnosing%20disease.

Times, T. F. (2007, July 16). FT Knowledge. Retrieved from Financial Times: http://www.ftknowledge.com/pma/brochures/Trading%20with%20the%20help%20of%20guerrillas%20and%20snipers.pdf