Robots will surely get better. Google’s self-driving cars, for example, are impressive, but in heavy rain or snow, a human had better take the wheel. And robots with more limited ambitions are already taking over backbreaking work on factory floors and assisting surgeons for greater precision and control in operating rooms.

But the greatest progress has been in software, which is rapidly moving into the mainstream of the economy. So far, the largest commercial use of learning software has been in marketing, where it improves the odds of making a sale — tailored marketing, targeted advertising and personalized product recommendations.

Big companies and start-ups are beginning to use learning software in higher-stakes decisions like medical diagnosis, crime prevention, hiring selections and loan approvals.

The idea is that an A.I. turbocharger can be applied to all kinds of decisions, making them smarter, fairer and less prone to human whim and bias. The goal could be saving money or saving lives.

Still, even enthusiasts have qualms.

Take consumer lending, a market where several start-ups are using big data and algorithms to assess the credit risk of borrowers. It’s a digital-age twist on the most basic tenet of banking: Know your customer. By harvesting data from many sources, including social network connections, even observing how an applicant fills out online forms, lenders say algorithms can more accurately predict whether a candidate will repay than by simply looking at a person’s credit history.

The promise is more efficient loan underwriting and pricing, saving consumers billions of dollars. But the new A.I. lending essentially amounts to a digital black box that pores over mountains of data. “A decision is made about you, and you have no idea why it was done,” said Rajeev Date, a former deputy director of the Consumer Financial Protection Bureau. “That is disquieting.”

Dr. Herbert Chase, a professor at Columbia’s College of Physicians and Surgeons, was asked as an unpaid researcher to try out IBM’s Watson software when the company’s scientists were adapting the technology for medicine. To try to stump Watson, Dr. Chase recalled a case decades earlier, when he made a correct diagnosis of adult rickets for a young woman, but only after extensive tests and months of being baffled. He fed Watson a few symptoms and the program, which ranks diagnoses by probability, swiftly replied and ranked adult rickets second.