What are the most common jobs? How can Machine Learning be useful?

@gbonaert Gregory Bonaert bonaert

I’m currently studying achine learning (among other things) and after listening the Hamming’s great talk You and Your Research I had to ask myself “What’s are the most important problems in ML?”

A more conventional way to approach this problem is to review the current literature and find out which problems are being solved and how popular they are.

This post looks at the problem in a different way, from a first-principles approach. For me, the two main goals of ML are to understand how we can learn in all contexts efficiently (knowledge gain) and to improve people’s lives by saving them from dangerous, repetitive, annoying or wasted work.

Ideally, ML should help as many people as possible. We should the order jobs by how much ML can be useful. Different sorting criteria are possible, such as how dangerous the job is or how much ML might be able to help in a 10 year time frame.

The approach we’ll take here is to figure out what jobs are the most common, because for those jobs, even a small improvement can help the lives of millions of people.

The US Department of Labor sponsors the CareerOneStop website, which has a list of most common jobs, which we copy below.

Occupation2018 Employment
Retail salespersons4,510,900
Combined food preparation and serving workers, including fast food3,704,200
Cashiers3,648,500
Office clerks, general3,158,500
Registered nurses3,059,800
Customer service representatives2,972,600
Laborers and freight, stock, and material movers, hand2,953,800
Waiters and waitresses2,634,600
Personal care aides2,421,200
Janitors and cleaners, except maids and housekeeping cleaners2,404,400
Secretaries and administrative assistants, except legal, medical, and executive2,382,500
General and operations managers2,376,400
Stock clerks and order fillers2,056,600
Heavy and tractor-trailer truck drivers1,958,800
Bookkeeping, accounting, and auditing clerks1,707,700
First-line supervisors of office and administrative support workers1,557,500
First-line supervisors of retail sales workers1,548,300
Nursing assistants1,513,200
Maids and housekeeping cleaners1,494,400
Maintenance and repair workers, general1,488,000
Elementary school teachers, except special education1,434,400
Accountants and auditors1,424,000
Sales representatives, wholesale and manufacturing, except technical and scientific products1,406,400
Construction laborers1,405,000
Teacher assistants1,380,300

Retail salespersons – 4,510,900
How ML can help:

Combined food preparation and serving workers, including fast food – 3,704,200
How ML can help: this job, according to what I understood, means handling customer orders, receiving payment, preparing simple food items like sandwiches, salads, pizza, french fries, hamburgers, coffee and doing some cleaning. They are not waiters or chefs / cooks.

The customer handling part is very similar to cashiers, except natural language understanding may be even more important. For food preparation, ML doesn’t seem absolutely necessary (classical engineering might be enough). It may however be very useful to improve object handling and manipulation. Most food preparation could be done automatically, and some order handling too.

Currently, this is very rare, except for order handling with screens. McDonalds still employs people to handle orders and make food.

Skills: object recognition, person recognition, speech recognition, question answering, speech synthesis, object manipulation

Cashiers – 3,648,500
How ML can affect it: it could entirely automate this job, or at least considerably reduce the amount of cashiers that are needed. See Amazon Go. A person would probably still needed to ensure the system is operational and help customers that are stuck, but more than that isn’t strictly required. However, some people prefer a human touch, so this may be preserved in more luxurious stores.

Skills: object recognition, person recognition, speech recognition, question answering, speech synthesis

Office clerks, general – 3,158,500
How ML can help: Office clerks do many things:

  • Answer phone calls
  • Welcome and direct visitors
  • Sort and distribute mail
  • Do data entry
  • Handle schedules and agendas
  • Word processing tasks: design and draft reports, minutes, forms, charts
  • Process some invoices

Lots of data entry could be automated, by having a system that sees how a human does it and then repeat that process X times until all data is entered. Some human intervention might be needed: check the answers and edge cases, answer questions when the system isn’t confident, input images of the data if the information is stored physically.

Answering phone calls is difficult to automate (see Customer service representative). Greeting people and directing them could be automated, but visitors may dislike that. The clerk may also be smart and help their boss a lot by redirecting visitors or avoiding that their boss has to deal with them.

Sorting mail could be automated, but distributing it is probably something that’s not worth being automated (unless there’s huge amounts of mail). Handling schedule, agendas and doing word processing drafts requires knowing what their boss wants and so may be very difficult to automate (guessing intent and preferences is a very hard task).

Skillls: OCR, structure recognition in documents, learning what clicks / actions to do on a computer automatically (if data should be taken from a program and not physically), speech recognition, question answering, speech synthesis

Registered nurses – 3,059,800
How ML can help: we should be very careful about adding ML to the daily work life of nurses, since they perform such an important job. At least initially, the goal should be to augment their job, possibly by producing helpful suggestion given the patient information or monitoring patients and risks more intelligently so that nurses can intervene in time and have all the information they need.

The human component is very important in this job, and it should be preserved. People want to be heard and cared for by other humans; they might reveal more about their ills, symptoms and psychological state. A human nurse will also make them less lonely and possibly lead to a better medical result.

ML could potentially be useful to avoid human errors, such as forgetting things or not noticing problems. Again, this is there to complement the nurse’s work.

Skills: action recognition, question answering, recommended action suggestion, verification of work, key information highlighting

Customer service representatives – 2,972,600
How ML can help: sufficiently advanced ML could automatically answer most questions at any time of the day, in any language. If the system is unable to do it or has missing information, a human operator could take over to help the customer and the system would automatically learn how to handle that.

People really dislike talking to bots, because they’re mechanic and generally stupid / very limited. If the system passes the Turing test, then this could automate a large part of the job.

Huge amounts of work are still needed though, because understanding speech, find an appropriate answer / response and reacting with a human-like voice are still far from being solved problems.

Skills: speech recognition, human language understanding, question answering, human-like reaction and speech synthesis, reaction, automated human intervention system

Laborers and freight, stock, and material movers, hand – 2,953,800
This job includes pickers in warehouses, who may need pack and wrap items and load / unload them from a truck. They may pack groceries for customers or shipping materials for transport, labeling and recording what they packed. Garbage men, car and equipment washer and machine feeders are also included. These jobs often require repetitive movement, heavy lifting and a fast work rate.

How ML can help: lately, technology tended to be against the workers, by either automating their job or ensuring high work rates are respected, which stresses the workers. These jobs can be quite hard, so making sure humans don’t have to do them might be positive.

If we could develop robots able to lift, move, wrap, unwrap and manipulate objects, then most of the work could be automated. They would need to understand the objects (could be helped through barcodes or similar tech) and what task to do. Washers and especially garbage men would be quite tricky to help though.

Skills: robotics, object recognition, object physical manipulation

Waiters and waitresses – 2,634,600
How ML can help:

Personal care aides – 2,421,200
How ML can help: this job, like nurses or teachers, should not be automated. Again, the human component is essential and produces tons of value in the aided person’s life. ML probably isn’t that useful for this job, except for health monitoring or alerting for when the aide isn’t there (if the person requires help, because they fell or need something urgently). There may be many more exception, but I have very little experience in this field so I only listed these 2 which I’m confident about.

Skills: anomaly detection

Janitors and cleaners, except maids and housekeeping cleaners – 2,404,400
How ML can help: this job isn’t very fun and ideally people wouldn’t have to do it: it would be really nice if things were clean without any human work. Cleaning, manual labors and material movers require the same type of abilities, but janitor and cleaner work is probably much harder to implement, because the environment isn’t controlled and the things to clean can be of any shape, size or material.

Skills: advanced robotics, object recognition, advanced object physical manipulation

Secretaries and administrative assistants, except legal, medical, and executive – 2,382,500
How ML can help: see office clerk

General and operations managers – 2,376,400
Excludes First-Line Supervisors. Supervisors tend to directly manage people and ensure they’re doing their work correctly and report to general managers. The general managers tend to manage an entire team or department, have the power to hire and fire, and choose the policy and direction the team / department will follow (to some extent, managers have bosses too). They assign duties, set goals and check financial data or performance reports to see how the team is doing.

How ML can help: this is a very general job, which requires social skills, high level thinking and creativity. It will be very hard for machine learning to contribute significantly to this job. ML may help dealing with all the financial and performance data automatically, and answer questions the manager might like to ask to the data. It may also help with budgets (though that is very political, so managers would probably not like to delegate this task).

Stock clerks and order fillers – 2,056,600
They receive merchandise, unpack it, and move product from the warehouse to the shelves. They track the movement of items and check for damaged good. Order fillers retrieve customer orders and prepare them to be shipped.

How ML can help: see Laborers and freight, stock, and material movers

Heavy and tractor-trailer truck drivers – 1,958,800
How ML can help: I expect cars and trucks to have better self-driving abilities over time, up to a point where a human driver wouldn’t be needed for most cases. Given their weight, safety is extremely important, because a truck crash will probably lead to the death of the passengers of the other vehicle.

In the long term, I believe this job will be automated, since it’s commercial in nature and cost is the most important factor. Starsky Robotics is pursuing an interesting approach: the truck self drives on the highway, but a human start driving from a distant office when the truck leaves the highway.

Tractor driving is less risky to human safety and so may be deployed much earlier. It drives over irregular terrain and may get stuck, so there might need to be occasional human intervention. Avoiding getting stuck is probably one of the main tractor-specific problems.

Skills: self driving (computer vision, planning)

Bookkeeping, accounting, and auditing clerks – 1,707,700
How ML can help:

First-line supervisors of office and administrative support workers – 1,557,500
How ML can help:

First-line supervisors of retail sales workers – 1,548,300
How ML can help:

Nursing assistants – 1,513,200
How ML can help: see nurses

Maids and housekeeping cleaners – 1,494,400
How ML can help: see janitors and cleaners

Maintenance and repair workers, general – 1,488,000
How ML can help:

Elementary school teachers, except special education – 1,434,400
How ML can help: ML will probably not have a huge impact on teachers in general, since it’s a job where the human social component matters a lot and can’t be substituted. Almost everyone would be better taught by a person than by a machine.

The main advantage could be to add personnalisation and tailor exercises to each student. This would take too much time for a teacher to do, but a good system could do it for them.

Skills: exercise synthesis, automated evaluation of skills

Accountants and auditors – 1,424,000
How ML can help:

Sales representatives, wholesale and manufacturing, except technical and scientific products – 1,406,400
How ML can help:

Construction laborers – 1,405,000
How ML can help: this job is going to be very hard to automate. It’s a hard, precise but varied, technical job in the physical world that isn’t done in a factory or controlled environment. Each of these makes it difficult to create a robot that could do it.

Better tooling would certainly be useful, but it’s unclear how ML would be needed here; it’s seems like we’re applying a solution to a problem that doesn’t need it.

Teacher assistants – 1,380,300
How ML can help: see elementary school teachers

Tags:

Leave a Reply