Sharing Information About Corner Cases

The definition of what is good enough when it comes to new technology is still evolving. In safety-critical applications, as well as a number of other areas such as drones or domestic robots, people will need to watch over all machines very closely rather than the other way around. While these machines may serve a useful purpose, they also need to be monitored to ensure they don’t go too far astray.

New technology is more like an early adopter program than a final product. And as with all early adopter technology programs there are bugs that need to be identified and refinements that will need to be made. Unfortunately for the auto, medical and aerospace industries, these deficiencies can inflict serious harm on people. Bugs in a home appliance or a computer operating system are simply an annoyance.

There is certainly a good argument to move forward on some of this technology. There were 38,300 people killed on U.S. roads in 2015 and 4.4 million seriously injured, according to a report from the National Safety Council. That number was up 8% year over year. “Americans take their safety on the roadways for granted,” said NSC president and CEO Deborah Hersman in a statement earlier this year. “Driving a car is one of the riskiest activities any of us undertake, in spite of decades of vehicle design improvements and traffic safety advancements.”

In contrast, there has been one highly publicized fatality in a self-driving car, and an unknown number of minor accidents. There are certainly far fewer autonomous vehicles on the road, but the statistics are encouraging enough for car companies to begin introducing more autonomous and semi-autonomous features each year. Time will tell how good of an idea this really is.

What’s arguably more worrisome are areas of new technology where sales volumes are much lower, such as industrial and commercial automation, robotics, and medical devices. It’s not that the manufacturers care any less about safety, but they will have far fewer real-world use cases against which to update their database of acceptable interactions.

As a whole, the tech industry has become quite comfortable with rushing products to market before they are fully baked and field testing them with consumers instead of engineers. From a cynical point of view, this is far better for a systems vendor from a competitive and cost standpoint. But it’s also better from a product development standpoint, because no matter how many corner cases that can be thought about ahead of time, there are always others that no one considered.

This approach works best in markets with more consumers of technology, because more people in more unique situations can identify more corner cases. There are an estimated 60 million cars sold worldwide each year, for example, compared with hundreds or thousands in other industries. But it also works best where there are tightly crafted standards to make sure that issues are identified for an entire industry rather than a single company. At this point, there is no clear methodology or infrastructure anywhere for trading that kind of information because this is all brand new and much of the information is considered proprietary.

That’s a big problem, and it’s one that needs to be solved by every segment of the tech industry, from better verification of hardware to software to trading of information about what problems have turned up in real-world testing. So far, there has been very little discussion about this and almost no action. It’s time that changed.