A serious AI program will generate failed experiments. The maturity of the program is measured by how cleanly it ends them.