Superintelligence (Part 2)

This is the second part in a series of essays on superintelligence. Here is a link to the first part.

One methodology that can help us analyze the impact AI has on work is to realize that a job is a bundle of tasks.

Let’s take a look at the job of being a host at a restaurant. The job is composed of several but not infinite tasks. They used to take reservations over the phone but much but has gone away with the adoption of OpenTable. Even without that one task, the job still includes showing people to their table, potentially handling takeout orders and providing a positive first experience when someone enters the restaurant. The jobs still exist but the list of tasks has changed slightly.

A different job but one of which hasn’t fared so well was that of a typist. A typist’s jobs really included just one task - type. A lot. Larger companies had typing pools. A typing pool was a centralized group of typists who handled all the document production for an organization. Executives and managers would either dictate letters and memos into dictation machines or write them out longhand, then send the work to the pool. A supervisor would distribute assignments, and typists would produce the final documents on typewriters.They were a common feature of corporate offices from roughly the early 1900s through the 1970s-80s.

The model made sense in its era for a few reasons: typewriters were considered “expensive capital equipment”, typing was a specialized skill requiring significant training, and there were real economies of scale in centralizing the work. A good typist could produce 60-80+ words per minute with high accuracy, and organizations needed that throughput for correspondence, reports, contracts, and internal memos.

The pools started declining in the 1970s-80s as word processors and then PCs arrived. Once corrections became trivial and the equipment cost dropped, it made more sense for individuals to type their own documents. The transition wasn’t instant though—plenty of executives resisted typing their own work well into the 1990s, viewing it as secretarial work beneath their role.

So the next question is… Are other jobs more like that of a restaurant host or a typist?

Or… What are some jobs that are more at risk in the next 3 years?

1) Driving

We touched on this a bit in a previous article: “Look! No Hands!”. Now we’re going to look a bit closer and talk about driving as a paid job.

By now, many of us have taken a ride in an Uber or Lyft. These services are an improvement over the traditional taxi experience. The two leading services are Uber and Lyft. Together they have millions of drivers in the US but less than a million drive as their primary job. These services offered 3.5 billion rides last year in the US.

Waymo and Tesla offer a similar service but without a human driver. The hope of these services over time is to offer a better experience for less money. The early leader has been Waymo and in 2025, they offered 14 million rides or less than a third of one percent of that of the ride services that had a driver. But the growth of Waymo and Tesla moving forward could be shockingly fast. It would not surprise me if 50% of rides were autonomous in 3 years. In that case, we’re talking about potentially a million job losses or more.

If people start using autonomous taxis at scale in the next 3 years, it’s hard to believe that humans will be driving 100% of the long haul semi trucks in 5 years.

There are 3 million truck drivers in the US. About half of those are long haul heavy truck drivers. These are people that drive heavy and long trucks more than 250 miles a day. On average, these truck drivers make $80,000 a year but they don’t stay at the same company for long. In many trucking companies the annual turn over is 90%.

It is actually easier for a computer to drive a vehicle on the highway than on city streets. Highways are much more controlled environments for safety reasons. The increase of speed would dramatically increase the danger without fencing, signage and other simplifications.

Wages are 40% of the cost a trucking company needs to charge for their service. So trucks that are autonomous can offer a significantly cheaper product than those with drivers. Additionally, autonomous trucks can potentially drive more hours in the day. The average truck is in use 11 hours a day. With no need for a driver, trucks could potentially double the amount of time they are being used.

Is truck driving a single task job? In some cases maybe but in a lot of cases probably not. You need someone to sign for a delivery or to help load or unload the truck. But at 40% of the cost of the service there will be a lot of pressure to rethink how things work to eliminate the driver.

I think adoption of autonomous trucks will be slower than cars. It could easily take double the amount of time, 6 years, for the first 50% of trucking to be automated. That’s still hundreds of thousands of jobs even before you consider the doubling of the amount of time a truck can be on the road. While the rate of initial adoption seems hard to predict, the longer range seems easier. It seems unlikely that the vast majority (90%) of trucking won’t have automated driving in 10 years.

Before forming a higher level opinion we’re going to explore more jobs. In upcoming essays, we’ll look at software engineers, junior lawyers and doctors.

More soon!