Human-Compatible Artificial Intelligence

N/A

Human-Compatible Artificial Intelligence Stuart Russell Computer Science Division, University of California, Berkeley 1 1 Human-Compatible Artificial Intelligence 1.1 Introduction Artificial intelligence (AI) has as its aim the creation of intelligent machines. An entity is considered to be intelligent, roughly speaking, if it chooses actions that are expected to achieve its objectives, given what it has perceived.1 Applying this definition to machines, one can deduce that AI aims to create machines that choose actions…

related reading

• The Future Worth Building Is Human - Thinking Machines Labthinkingmachines.ai
• Why Tool AIs Want to Be Agent AIs · Gwern.netgwern.net
• What failure looks like — LessWronglesswrong.com
• Concerns of an Artificial Intelligence Pioneer | Quanta Magazinequantamagazine.org
• What Does It Mean to Align AI With Human Values? | Quanta Magazinequantamagazine.org
• The hot mess theory of AI misalignment: More intelligent agents behave less coherently | Jascha’s blogsohl-dickstein.github.io
• Four Background Claims - Machine Intelligence Research Instituteintelligence.org
• TechnicalAgenda.pdfintelligence.org
• The Problem — LessWronglesswrong.com
• Optimality is the tiger, and agents are its teeth — LessWronglesswrong.com
• What failure looks like — AI Alignment Forumalignmentforum.org
• Dreams of Friendliness — LessWronglesswrong.com