Disentangling Fiction from Frameworks: A Critical Review of Instrumental Convergence in AI Ethics Risk

This paper debunks the hypothetical situation presented by instrumental convergence -- the belief that Artificial Intelligence will develop sub-goals which will make it indifferent to its previous programming and unstoppable against humans -- by stating how AI lacks intrinsic desires, self-preservation instincts, and agency unless explicitly programmed with such tendencies. In other words, lacking these characteristics which would motivate it to develop a selfish character is unrealistic and a product of anthropomorphism, which wrongly associates AI with human attributes that make instrumental convergence possible. Having such associations with AI is dangerous as it may lead to its mischaracterization, overestimation, and lack of accountability for the people that developed its bugs and errors. Overall, instrumental convergence is unrealistic and provides additional barriers to AI Alignment with human values and understanding of its capabilities.