What Is Friendly AI?
The Artificial Intelligence refers to AI systems which are helpful to humanity rather than harmful or neutral is called Friendly AI. This term is used as these systems are friendly because they want to be and not because they are forced to do so. Apart from referring to completed systems, Friendly AI also refers to the name of theoretical and engineering discipline.
Eliezer Yudkowsky of Singularity Institute for Artificial Intelligence originated the term “Friendly AI”. His goal was to create a Friendly AI smart enough which is smart enough to improve itself without the intervention of the programmer. In 2001, he published a book named “Creating Friendly AI” which was the first rigorous treatment on the topic. Yudkowsky invokes arguments for cognitive sciences and evolutionary psychology to support his approach.
As the past attempts to solve the problem aroused by simple AI involves strict programmatic constraints and were bound to collapse under alternative interpretations, so friendly AI is considered important as an academic discipline. Another problem with AI is anthropomorphism. Many thinkers assume that AI should be similar to evolution of organisms that tend to be selfish and it should gain it immediately or after becoming enough smart.
There is no other way than self centered goal systems that are involved in evolution of organism. Millions of years can be taken by Altruistic goal systems to evolve and under the conditions which members of the community gain by helping each other rather than neglecting. So we can build a system or mind from scratch which is not a self centered goal system. This does not mean that the AI would be restricted but creation of AI which is unselfish by nature.
Although the above observation contributes to the field of friendly AI but it is extremely new and still much work is to be done. There is an argument among researchers that until we don’t have smarter AI systems to experiment on, we cannot determine necessary design features that can be used to implement Friendly AI. Some researchers argue that theoretical approach is ethically necessary before an attempt on general AI is started.
Category: Technology