The problem with AI doing computer programming: or Self Programming by AI

 The problem with having AI do it's own computer programming is that as programs get more complicated and advanced it might not be possible for a human being or even 100 human beings to basically check the computer program for errors and the AI might not find the same errors a person would because an AI is not human with human considerations and did not grow up here on earth in a human body here on earth but is a program instead usually in a computer or drone or robot here on earth or in Space.

For example, I use what is called "Compose" which is an automatic programming in HTML function here so I don't have to program HTML myself. Where it comes in the most handy is basically "All the time". However, often when I try to convert something quoted on the Internet giving credit where credit is due the Auto-Coder here at blogger.com disagrees with the Auto-coders? or human programmers at what I'm quoting so there is no resolution between the two of them often and because of this there is a lot I cannot share if I want to.

So, I foresee the same problem even with very complicated programs self programmed by AI in various places disagreeing with other programs programmed with different AI with different priorities with all kinds of problems as a result.

The easiest example of a problem a human being had with an AI that caused the human to die was in a Tesla Model S early on in Florida. The "Driver?" was sitting in the driver's seat but had engaged the "Autopilot" feature on the Tesla Model S in Florida. The problem is that a sign painted on a Semi Truck advertising the goods or service drove across the freeway that the Tesla model S was driving down. The problem became that the Autopilot thought that the semi was a roadside Sign advertising something instead of the Semi-truck that was crossing in front of it and this beheaded the Driver of the car who was watching a Harry Potter movie on the TV screen in the center of the console.

These types of problems you are going to see where programs and equipment misidentify what they are looking at and more and more people are going to die in various unusual ways.

You could theoretically plan for all the possible bad things a computer could do which could accidentally cause the end or maiming of a human life. In fact, if you asked an AI to see what those types of deaths or maimings might look like it might be able to tell you what likely could or would occur.

Doing this might be very useful to save 1000s of lives in the U.S. and around the world in the next few hundred years because AI likely could ferret out these types of crazy life threatening problems that could be created accidentally by software programmed by AI in different forms.