Self-flight is already running on all commercial airlines, but only at the cruising altitude. When you flight from London to New York, most of the time the pilots are monitoring the autopilot systems. Running AI on cruising altitude is arguably a simper problem than self-driving cars. The system has been in-place for years.
The real danger of airlines is taking off/landing. As you can imagine, the costs for a bad judgement could be quite serious. Anything could happen while the plane is still under the clouds. While our AI technology is improving, it will never replace human intervention, especially when the costs of collision is very high. If you don't believe me, google for how Facebook is hiring humans for reading "fake news". Facebook has publicity acknowledged AI will never be sufficient for them, then why would AI good enough for replacing well trained pilots?
AI on a commercial airplane is already taking place, but there is no possibility ever it will take over the world. We will always need pilots who knows how to work with the system, making important decisions.
Trained pilots + good AI assistance will make the world a better place.