A great paper for the Labor Day weekend, if you have the time and stomach for it, is MIT economist David Autor’s latest work on machines’ ability to take our jobs.

Called “Polanyi’s Paradox and the Shape of Employment Growth,” the paper starts with the quote from the philosopher Michael Polanyi who famously observed that “we can know more than we can tell.”

It’s this idea, that we can do complex jobs that we really don’t fully understand, that lies at the heart of the argument that computers and robots will never successfully replace human labor.

What’s interesting is that the quote also refers to the irreplaceable art of driving a car -- we are good at it even though we may not understand the physics or engineering of our vehicle. A machine-driven car in 1966 seemed beyond the wildest forecast of machine capability, of course, but thanks to Google we know it’s coming, likely in most of our lifetimes.

Autor, however, is mostly an optimist about the ability of humans to hang in there in the labor market.  The computers haven’t been able to do the jobs that require non-routine work, at both ends of the income spectrum. As he put it, “the challenges to substituting machines for workers in tasks requiring adaptability, common sense and creativity remain immense.”

In one great example, he writes it is relatively easy to come up with the machine that can recognize shapes and patterns, but a machine isn’t going to as easily understand what an object is for. Meanwhile, a person looking for a chair will quickly reject the toilet and a traffic cone that, to a computer, might both look suitable for sitting.

His implied advice is that people looking for skills and good employment prospects need to think of jobs that are complemented by computers, and he's not just writing about jobs in management or the top professions.

While many middle-skill jobs do have tasks that can be easily automated, to do them well a person needs a broad range of skills. It’s not that easy to unbundle a job and assign to a computer just the specific, repetitive tasks that lend themselves to automation.

An example he uses is calling software tech support. A computer might search the databases for known issues that seem to be the customer’s problem while the technician politely chats up the customer on the phone. When a solution is located, the technician can read back what the computer found.

Yet that is not a very productive form of labor organization, he wrote, making it possible for jobs to persist with routine and nonroutine aspects are complementary. Back to the example of tech support, the technical expertise has the most value when it comes from a person who has the judgment to help the customer figure out what’s wrong and can talk her through the best solution.

All of which is to suggest, he concluded, that the pessimists about the future of human work are overstating their case. It also means, thankfully, that we can look forward to many more Labor Day weekends.