“Weak AI” and/or “Narrow AI” is defined in the article linked below as Artificial Intelligence (AI) that may be able to match or even exceed human intelligence in a narrowly specified area, but not in a broader, overall general intelligence. As a result, we end up with tools that can operate faster, better, stronger in a variety of ways than human intelligence but without the wisdom and experience to apply that power in ways that always help and never hurt. The article correctly describes a real threat, but the threat comes not from AI but from the failure of limited human intelligence.
If we deploy a limited or narrow AI system that does not qualify as general intelligence with the ability to make wise decisions and that system is so complex that we can not predict its behavior, then we are responsible for any failure. Garbage in – garbage out. We write the program. We provide and filter the input. We analyze the performance of the program. We deploy it and enable it. Any catastrophic failure at that point is a human failure, not an artificial failure.
At some point, no doubt, we will deploy a fully aware and conscious artificial intelligence that is capable of making wiser decisions than most humans. And at some point, one of these will fail. But overall, the failure rate is likely to be lower than if humans were performing the same task. And in many cases, humans won’t be able to perform the same task.