Trending: Taste-testing the Cosmic Crisp, Washington state’s star-studded ‘apple of the future’
A dermatologist uses a dermatoscope, a type of handheld microscope, to look at skin. Computer scientists at Stanford have created an artificially intelligent diagnosis algorithm for skin cancer that matched the performance of board-certified dermatologists. (Stanford Photo / Matt Young)

Computer scientists have created artificial-intelligence algorithms that are at least as good as trained humans at recognizing the signs of skin cancer or malaria, but does that mean your future physician will be a bot?

Two experts on AI explain in the journal Science why the rapid rise of machine learning could be good for well-paid professionals like dermatologists and epidemiologists, no big deal for workers on the low end of the wage spectrum, but big trouble for employees in the middle.

That’s because those middle-spectrum jobs are particularly vulnerable to the machine-learning treatment, MIT’s Erik Brynjolfsson and Carnegie Mellon University’s Tom Mitchell say.

“Although economic effects … are relatively limited today, and we are not facing the imminent ‘end of work’ as is sometimes proclaimed, the implications for the economy and the workforce going forward are profound,” they say in their commentary, published today.

The two researchers are also the co-chairs of a National Academies committee that put out a report on information technology and the future of the U.S. workforce, in which the authors say there’s a strong case that “technological advances have contributed to wage inequality.”

Machine learning is an important swath of the artificial-intelligence field, having to do with the capacity of computer programs to take in huge gulps of data and tease out correlations between input and output on their own. Some of those correlations may be so subtle they go unnoticed by human experts.

The highest-profile examples of machine learning’s power include the game-playing prowess of Google DeepMind’s AlphaGo Zero and Carnegie Mellon’s Libratus poker program. But there are less flashy examples as well, such as Stanford University’s cancer-detecting algorithm and Intellectual Ventures’ malaria-hunting microscope.

An application that spots the signs of skin cancer isn’t likely to render dermatologists obsolete, Mitchell said in a news release.

“I think what’s going to happen to dermatologists is, they will become better dermatologists and will have more time to spend with patients,” he said. “People whose jobs involve human-to-human interaction are going to be more valuable, because they can’t be automated.”

On the other end of the spectrum, machine learning hasn’t had a huge effect on low-skill service jobs such as janitorial services or home health assistance, the researchers note.

But between the highs and the lows, there’s a widening swath of workers whose jobs could be threatened. “Some ‘creative’ tasks that were previously reserved for humans will be increasingly automatable in the coming years,” Mitchell and Brynjolfsson say.

Machines are becoming increasingly adept at answering questions where there’s a clear right vs. wrong answer, or even a better vs. worse answer.

“At the same time, the role of humans in more clearly defining goals will become more important, suggesting an increased role for scientists, entrepreneurs, and those making a contribution by asking the right questions, even if the machines are often better able to find the solutions to those questions once they are clearly defined,” the researchers write.

Here are four questions that make a job more likely to be automated if the answer is “yes”:

  • Does the job involve learning a function that maps well-defined inputs to well-defined outputs?
  • Do large data sets exist that contain input-output pairs, or can those data sets be created?
  • Does the task provide clear feedback with clearly definable goals and metrics?
  • Is there a tolerance for error, with no need for 100 percent accuracy or an absolutely optimal solution?

And here are four questions that make a job less likely to be automated if the answer is “yes”:

  • Does the task involve a long chain of logic or reasoning that depends on diverse background knowledge or common sense?
  • Is there a need for a detailed explanation of how a decision was made?
  • Does the phenomenon or function being learned change rapidly over time?
  • Is specialized dexterity, physical skill or mobility required?

For more about the expected impact of machine learning, including economic factors that are likely to have an impact on AI’s spread, check out the commentary on Science’s website.

Subscribe to GeekWire's Space & Science weekly newsletter


Job Listings on GeekWork

Find more jobs on GeekWork. Employers, post a job here.