Why Artificial Intelligence can be dangerous for the future of humanity?

2023-05-04 07:59:00, Tech Paolo Mauri

Why Artificial Intelligence can be dangerous for the future of humanity?

Illustrative photo

Nik Bostrom is a Swedish professor of philosophy. He directs the Institute for the Future of Humanity at Britain's University of Oxford. Bostrom's field of study is that of transhumanism, and includes areas such as cloning, Artificial Intelligence (AI), super-intelligence, the possibility of transfer of consciousness to technological support, nano-technologies and theses on simulated reality.

As for AI and its future developments, the professor is pessimistic, and says they pose a danger to all of humanity. This risk can appear over a very long period of time, but according to him it can also appear much faster. To understand what the risks really are, but above all how AI works in view of the possible arrival of super-intelligence, we interviewed him.

Professor, you are very critical of the development of AI, with a very pessimistic long-term perspective regarding the dangers it may pose to all of humanity. So I ask you: Is it possible to teach a machine ethics, sensitivity and compassion, exclusively human values? How is this done?

I have both a sympathetic and a skeptical approach to Artificial Intelligence. I am very optimistic about the potential progress if we get things right. But on the other hand I think there are serious risks associated with the development of super-intelligence machines. It is not yet known exactly what we need to drive advanced AI systems, and this is an ongoing area of ??research. Many of the smartest people I know now work in this very field.

If Artificial Intelligence can be taught human values, knowing that man is not only characterized by positive values, but also evil, vile and violent, will it not be dangerous that the machine independently, and by knew us well, would copy these too, and behave aggressively towards us?

Yes, she can even become aggressive towards us even without those negative values. Can Isaac Asimov's Laws of Robotics Apply to AI?

Isaac Asimov's approach was very simplistic.

Do you believe that controlling the abilities of an Artificial Intelligence can only be a temporary measure? Wouldn't it be enough to introduce another type of algorithm that automatically curbs any aggressive behavior? Is it possible to define such an algorithm?

Isn't it aggressive behavior when we crush an ant colony to build a parking lot?

Knowing that Artificial Intelligence works based on the data set we give it, wouldn't it be enough to choose the data set wisely to avoid bad surprises?

Intelligent AIs can generalize beyond the data they operate on. They can build models of the world and use them to understand new situations and make plans.

Artificial Intelligence machines are faster than humans in learning concepts and reasoning, but are not yet capable of performing automatic human actions, that is, those based on instinct. Will this void ever be filled?

I think AIs are currently slower than humans at learning concepts. For example, ChatGPT-4 is trained on almost the entire Internet. But people reach a similar conceptual understanding based on much less data./ Adapted from CNA.al