AI apocalypse has been the favourite subject of many science fiction movies so far. However, when respected scientists and entrepreneurs vocalize similar concerns, we should be concerned. Being a student in fields close to "AI", I would like to believe that AI is going to be mostly harmless or at least claim that it depends on the user.
There are various reasons why I believe that robots with human like intelligence not going to be much much superior than the entire mankind and even if they are, why the apocalypse like situation is unlikely.
Inherent Compromises
Human like intelligence, or AI is inherently different from computers like intelligence. Humans learn by experience, i.e. by trial and error which is true for modern AI or machine learning algorithms. Hence, making mistakes is not only our weakness, but also our strength. Any algorithm that is not deterministic is bound make wrong or suboptimal choices at least some of the time.Another thing to understand about the AI, that is relevant to the argument, is "overfitting". It is often believed that an AI machine can grow infinitely wiser because of the tremendous amount of data we can throw at it. However, the more data will mean that the machine grows wiser but more and more rigid in terms of adaptability. Think of how old people find it difficult to adjust to changing surroundings and find it difficult to learn new things. This is, in my understanding, equivalent to a machine learning term called "overfitting". In a way, I am claiming that experience and adaptability are mutually incompatible and each AI machine designed will have to strike a balance among the two. Making a machine that "understands" everything and is still open to learning new things is as challenging as designing a car that is both most powerful and has best fuel economy.
Evolution is better
Another things that makes me undermine AI is my respect towards evolution. Evolution is like a constant search for a better organism with sole aim of survival. The algorithm has been running constantly for 3.6 billion years now. I am not saying that we are the optimal point in the search, but AI will have start all over again to make something as spontaneously survivable as human beings with entirely different ingredients.Peace is better than war
This brings me to the second part of the argument that is even if AI becomes super intelligent than humans, it is not going to take over us by a war. There is evidence to support that world is becoming less and less violent. And I believe that this is not because we have increased sense of morality but because we understand that war is suboptimal for both sides. If we accidentally build an AI machine, that is far more intelligent than us, probably it is going to understand the importance of peace better than us.Also, I believe that the difference between human like robots and robots assisted humans is going to decrease over time. By the time we will have intelligent robots, we will have implants and assistive devices that will make human decisions far more logical, fast and data driven. This will also make it difficult for the evil AI machine to decide who is a machine and who is a human.
Super intelligent will become Politicians or CEOs
We should ask ourselves, do the super intelligent rule the world as it is now? Yes they do. Most intelligent people in the world realize that morality is for show off, not for practice. Hence they become politicians or CEOs. I would like to speculate that super intelligent AI machines that can lie better than humans will indeed take over the world, but the transition will be too smooth to be noticed. It is more likely that we will be fighting for their "human like" rights.In the end, I would just add my political advice to a sci-fi article: our best bets against the current intelligent and future super intelligent are in keeping the power over natural resources decentralized.
