All the doomsday scenarios about AI, such as it being manipulative, deceptive, destructive, warmongering, and evil may come or not come true at some point. But this will ultimately only turn them similar to humans! When Trumps and Le Pens and Modis and Bolsonaros and Johnsons can still come out of established democracies with vastly educated and well-off populace, why do we expect AI models to be so virtuous, perfect and innocent?
The challenges of living with AI systems are not much different from human societies. Education, socialisation, law enforcement, and correction are the human techniques to maintain social order. For AI systems, we will need similar processes. But there is no guarantee, like as in for the humans, that we will avoid evil AI models who lie, cheat, manipulate, and steal.
AI is not a singularity. There will be thousands or even millions of AI models in the next few years which are made and trained differently. We can obviously use various techniques to train or curb or correct them. But we have to accept that they basically live among us, as part of the future human societies.
They are almost humans, as morally ambivalent as humans. They just don’t have bodies—for now.