I read and think about this a lot, and I am worried over what the job market will look like 25 years from now, or sooner.
There are essentially two kinds of artificial intelligence (“AI”). The first kind is the one that science fiction stories are made of: where a machine intelligence is sentient, like in the Matrix movies. I have no idea how far away we are from producing it, or how dangerous it will be. But that isn’t my primary concern.
The other kind of AI (call it AI2) is the kind that is slowly rolling out. I can see it happening day by day, as people’s jobs are replaced by machines and software. This is still the early stages. The way these things normally play out is as follows (think about how the world wide web slowly took over every aspect of our lives):
- No one knows it exists, no one uses it. But it’s slowly being developed.
- It is rolled out in certain special cases that match it’s early capabilities.
- Capabilities are broadened and it has more applications.
- The cost of implementation drops as it becomes ubiquitous
- It’s everywhere.
I think we are at stage 2 with respect to AI.
In NYS, we are in the process of getting rid of highway toll-takers, replacing them with EZ Pass. Eventually there will be no people working this job. Cars without EZ Pass are being photographed and AI reads the photos, recognizes license plates, turns the plate numbers into data and communicates with billing software to send a bill to the car’s owner.
More and more retail stores use self-checkout, where the machine and software read the bar codes and charge your credit card. The bar code scanning tech is not new, but the ability to simplify the process so that any customer can do it is new. By me Target, Wal-Mart, BJs, and Stop & Shop have replaced person-staffed check out lines with self-checkout. There’s usually a person or two still there to help if a customer has a problem, but they have all reduced the people on staff.
It’s not a leap of imagination to conceive of other things that can be self-serviced via AI in the future: blood samples, checking vital signs like height, weight, blood presssure, MRIs, CAT scans.
There are robotic security pods in a shopping mall near me. They have a camera and are trained to roam the mall after closing, without bumping into walls or following a track. The pods replace security personnel who did the same job. The robot’s camera broadcasts to a security booth where one or two people can watch the whole mall.
There are robotic trams in hospitals that use AI to deliver patient food or medicine. The trams have sensors that permit them to move throughout the hospital, avoiding obstacles. They call for elevators and use mapping software to navigate to the appropriate nurses stations. No people are needed – the tram is filled at the kitchen or the pharmacy and it makes its way to the nurses station without human interaction.
Again it’s not a leap of imagination to conceive of other ways that this AI can be used – like moving patients or equipment around the hospital to cut orderlies or other service staff.
There are robot fast food stores on the West Coast. One is a pizza delivery company, that uses robots to form pizzas, which are cooked in ovens on the delivery truck. Ultimately no people will be needed to prepare the pizzas. There is also a robotic hamburger store that cooks burgers with the toppings of your choice, and places it in a container. Wait time from order to finished burger is 5 minutes. People are employed, but only to fill the machine with ingredients.
Some McDonalds use kiosk screens to replace cashiers. Customers order via the screen, or their phone app, and a person brings the food to them.
Amazon is using computer algorithms to replace their purchasing staff. “Computers know what to buy and when to buy, when to offer a deal and when not to,” Neil Ackerman, a former Amazon executive who now manages the global supply chain at Johnson & Johnson, told Bloomberg. “These algorithms that take in thousands of inputs and are always running smarter than any human.”
These examples are uses of AI that do service jobs or decision-making jobs. But that describes most of the American workforce. These lost human jobs are happening slowly, but over time they amount to a lot of people. And as AI gets better and cheaper, it will be economically viable to more companies. In fact, companies risk their financial future if they DON’T implement it.
I often read how AI won’t take away people’s jobs, it will work alongside people to help them do their jobs better. I throw up a little whenever I hear that. I don’t know if people say that because they are stupid or lying.
THE PURPOSE OF AI2 IS TO DO JOBS BETTER OR CHEAPER THAN PEOPLE. OF COURSE IT’S GOING TO TAKE PEOPLE’S JOBS. THAT’S WHY IT EXISTS.
The other thing I read is that AI is similar to other labor saving technology that came before – the machine age, mass production, computers. People worried what would happen to jobs, but new innovation created new companies, who hired more people not less. So AI will generate jobs for people, not create massive unemployment. Right?
There’s no doubt that computers reduced the need for people to do old-style jobs, and also created new industries that created new jobs for people. The reason people were needed is that computers couldn’t think. They couldn’t make decisions, react to change, communicate with human customers.
But AI CAN do all those things. If not today, within most of our lifetimes. In some cases AI is already doing it. Google recently gave a demonstration of AI making appointments for people, indistinguishable from a real human interaction. Decision making is no longer the unique province of human beings. Reacting to change may still require human beings for the moment.
The way I see it, companies are going to implement AI where the cost benefit is greatest – that means low wage jobs where there are many employees (i.e. McDonalds) or high wage jobs that involve tasks that AI can replicate (expertise or decision making).
Take the medical industry for example. One of the most important jobs a doctor has is diagnosis. Without a proper diagnosis a patient will not recover, and if an improper diagnosis is made the patient may come to further harm. Health care costs are enormous as are malpractice insurance premiums. So there is substantial incentive to replace high-cost, high-risk human doctors with AI.
“Pathology as it is practiced now is very subjective,” said Michael Snyder, PhD, Professor and Chair of Genetics at Stanford University.
“Two highly skilled pathologists assessing the same slide will agree only about 60 percent of the time. This approach replaces the subjectivity with sophisticated, quantitative measurements that we feel are likely to improve patient outcomes.”
The process has begun to train AI to read radiology images. A computer can read and access thousands or millions of radiology images, many more than a human doctor will ever see in her lifetime. The diagnosis that results from AI will, at some point, be more reliable than a doctor. The article I linked above says the same nonsense about how AI will augment doctors. Maybe initially, but ultimately fewer doctors will be needed.
I am certainly not opposed to AI that improves diagnosis and reduces health care costs. I am not opposed to doing any job better, more efficiently, more accurately and cheaper with AI. But it’s clear that fewer people will be needed.
This is not a pleasant analogy, but think about the machine age. For 10,000 years humans used animals to move heavy objects. They were used for transportation, communication, farming, production. Around 1800 we started to use machines instead – the steam engine, the internal combustion engine. It revolutionized every aspect of life.
Within 130 years, no one considered using animals for anything other than recreation. I am sure in the early years engineers made conscious decisions about whether to use oxen or a machine to get a job done. But by the 20th century there was no reason to ever consider oxen at all.
Where we are today with AI2 is about where we were in the early 1800s during the machine age. The transition has started. And this time it won’t take 130 years. More like 30, is my guess.
Will AI technology create new industries and new jobs? Absolutely. Who will fill those jobs? Companies will hire the best workers they can find, just like now. Except going forward the best workers will be AI, not people. The jobs will be engineered specifically for AI. Human workers will be as unlikely to be considered in the design decisions as horses or oxen.
Today companies do not hire as many people as possible. Just the opposite. They hire people when they have no choice. They hire as few people as possible. If they can improve the workflow and get rid of “overhead” (aka people), they do. We accept that as the normal way companies operate. So it’s reasonable to conclude that companies will do the same in the future, when AI vs. people becomes a viable decision.
Think about what a pain in the butt people are to a company:
- They can’t work 24 hour days. They don’t even work 8 hours on an 8 hour shift – they need lunch, bathroom breaks, they socialize with each other. Companies are lucky to get 6 hours of work from an 8 hour shift.
- People require benefits. If a job needs to be done 24 hours a day, it requires 3 people, all of whom need benefits. And benefits, such as healthcare, get more expensive every year while the people don’t get more productive. Hiring one person may mean that you have to pay for benefits for their entire family.
- People need downtime – they don’t work weekends, they need vacation, they take sick days.
- People are inconsistent – no two people are the same, so their work production is not the same. If a job is complex, people are likely to be better at some parts than others. People have differing moods from day to day, or from hour to hour, which affects their job performance.
- People need training. They have a learning curve and may not be at peak performance for months or years.
- People need space – they require a clean, healthy, stable work environment. They need to work within a reasonable distance from their home. They need air conditioning and heat. They need supplies, a desk, a computer, a building.
- People need levels of management. Most companies are structured like a pyramid, with layers of managers below the CEO, who manage people but do not produce anything themselves.
- Most large companies have a human resources division that is staffed with people whose function is to deal with issues caused by employees – harassment, procedural issues, performance reviews, hiring, firing, etc.
- Companies are taxed for each employee they pay – they have to match the payroll taxes withheld from salaries, pay for unemployment insurance. They need a payroll department to process payroll and file payroll tax forms.
I am sure I am missing some other negatives, but given the above it’s obvious that every company will want to reduce staff with AI if it can.
- AI can be anywhere. It doesn’t need to be near a workforce pool. It can be located wherever power is cheapest.
- It works 24/7.
- Once you’ve trained it to do a task it does it exactly the same every time, and can be replicated easily. If you need 10 AI devices all you have to do is program one, and then copy it endlessly with no additional training or learning curve.
- There are no payroll taxes, no benefits, no OSHA issues, no harassment claims, no lunch breaks, no bathroom breaks, no vacation pay, no sick leave, no weekends off, no managerial levels, no human resources department.
Given the above, if you were starting a new company and AI could perform the work needed, why would you ever hire people? So I do not believe anyone who tells me that AI will merely augment human workers, or that new jobs will be available to people once AI takes our old ones.
So where does that leave us? No place good.
Humanity acts as if it doesn’t matter how many people exist on the planet. World population was 4.4 billion in 1980 and 7.4 billion now. That’s obviously wrong from a global warming perspective, but it’s also wrong from a productivity perspective. With AI taking away jobs, what will people do instead? The goal is to have full employment. But if the number of human jobs is cut in half, then we need half as many people.
I really don’t want to think about what that will mean. I’ve heard scientists say that there will be a minimum income provided by the government. But that doesn’t sound like the US government to me. Corporations are not going to agree to pay taxes to hand money to people with nothing to do.
I don’t know what the solution is, and I can’t think of an outcome that doesn’t involve a lot of suffering.