Computer Systems will Fail
A lot of people are putting their trust in the new AI systems. In reality, all that we can be confident in is that eventually these systems will fail. The failure point is not in the systems themselves, but in the people. Users will always push systems past their safety points. People will always misuse tools. Human beings will make systems fail.
At a manufacturing plant, some material got jammed and a machine stopped working. A young guy hops up on the conveyor belt, climbs up and starts unjamming the material. This is a total violation of safety rules and if something had started up again, he would have been killed. But this is typical of how human beings operate.
People push themselves and their machines past the limits. Folks drive on bald tires. Truckers drive past the time limits. People drive cars with broken brakes, front windshields totally covered in ice or broken out, or any number of other things that make the car unsafe. People run through red lights with the confidence that others will yield the right of way.
No matter how good a computer system is, humans will use that computer system in ways that it was not intended for.
People will use software in ways that the designers didn't expect. Using systems in these ways can cause problems when the software is upgraded. When Microsoft releases a new version of their software, they run into situations where people have used bugs as features. Those people then report problems with the new version as the "feature" / bug had been fixed. How they used the system does not work anymore.
Many computer CPU chips have been misused. Every computer CPU chip has instructions that were not "finished", documented, or even intended. On many a CPU, people explored those undocumented instructions, found out how they worked on a certain version of the CPU, wrote programs around those undocumented instructions, and then had problems on the next version of the CPU when that "instruction" didn't work any more or worked in a different way.
A few years ago, Microsoft attempted to release an AI chat bot. It took only a few hours before people had fed it enough harmful material that it was generating similar harmful material for more innocent requests.
An AI system intended to help diagnose hospital patients had not been trained on enough minority patient data and was giving incorrect suggestions when faced with minority patients.
So, we can expect the computer system to fail even when all the hardware is working properly. It will fail because human beings will push the hardware past safe operating parameters, push the software with data that it wasn't designed for, and use the system in places and situations that it wasn't intended for.