This article considers the question of how we may trust automatically generated program code. The code walkthroughs and inspections of software engineering mimic the ways that mathematicians go about assuring themselves that a mathematical proof is true. Mathematicians have difficulty accepting a computer generated proof because they cannot go through the social processes of trusting its construction. Similarly, those involved in accepting a proof of a computer system or computer generated code cannot go through their traditional processes of trust. The process of software verification is bound up in software quality assurance procedures, which are themselves subject to commercial pressures. Quality standards, including military standards, have procedures for human trust designed into them. An action research case study of an avionics system within a military aircraft company illustrates these points, where the software quality assurance (SQA) procedures were incommensurable with the use of automatically generated code.
Verifying a logic synthesis tool in Nuprl: A case study in software verification
