Please enable JavaScript.
Coggle requires JavaScript to display documents.
The Future of Artificial Intelligence (AI safety ("Prosaic AI"…
The Future of Artificial Intelligence
AI safety
Near-term
"Concrete Problems in AI Safety"
Adversarial examples
Attacks with traffic sign adversarial examples
https://arxiv.org/pdf/1707.08945.pdf
"Synthesising Robust Adversarial Examples"
Turtle robustly identified as a rifle from all angles by image recognition AI.
https://arxiv.org/pdf/1707.07397.pdf
AI security risks
Automated spear phising
Hacking of AI command systems
"Prosaic AI"
OpenAI
Paul Christiano and Dario Amodei, 80k Interviews
DeepMind
Pushmeet Kohli, 80k Interview
Iterated Distillation and Amplification
Paul Christiano
"AI Safety via Debate"
https://debate-game.openai.com/
MIRI approach
Eliezer Yudkowsky
Pure mathematics, decision theory and philosophy
Nick Bostrom's "Superintellience"
Predictions
AI Impacts
Katja Grace, 80k Interview
"S-risks"
Fundamental Research Institute (FRI)
Brian Tomasik