News Reports from a number of sources indicate that many believe the FAA went too far in delegating certification tasks to Boeing during the development of the 737 MAX. That level of delegation, particularly with respect to the MCAS system, resulted in problems not being detected during certification, and now being correct after a fix that comes too late for 346 people no longer with us. Analysis The FAA, as one would expect, is always one step behind the innovations being placed in airplanes. As new technologies emerge, the agency has the responsibility for evaluating those technologies and ensuring they are safe. That is easy for physical components and materials science. But the difficult tasks come in evaluating the effectiveness of software, which is increasingly utilized to control aircraft. It is difficult for regulators to vet millions of lines of computer code and to anticipate every circumstance that could occur, and how the software will react. The time required to analyze software would end up being nearly as long as it took to develop it. Looking over the requirements definition and testing protocols is about the best that one can do in this area, and even that has come up short in the case of the 737 MAX and MCAS. In that case, the systems reliance on one sensor, the resulting installation allowing stabilizer travel four times the planned distance, and the system re-booting itself multiple times could have been potentially vetted and flagged, had regulators not delegated the tasks to Boeing personnel. Insight The problem with evaluation of software is only going to get worse, and become more difficult as Artificial Intelligence is utilized first in emerging air taxi and military drone operations, and later integrated into mainstream aircraft. Where will the FAA obtain the expertise to evaluate systems that are built to “learn” and “adapt” to conditions? The FAA does not have a cadre of silicon valley AI experts and programmers who have the expertise to evaluate the systems that will within a generation begin controlling our aircraft. But somebody needs to ensure that the systems are safer. Solving the problem may only be possible through privatization of the FAA, so that regulators can hire the right level of experts in AI to take on the task of regulating autonomous vehicles. But hiring those folks will be difficult in an environment in which the private sector can pay 3-4 times as much for their services. A privatized FAA, funded by the tasks it performs by those using them, could help provide the regulatory scrutiny and independence that is needed.
