I think you underestimate how much MIRI agrees with FLI.
Agreed—or, at least, he underestimates how much FLI agrees with MIRI. This is pretty obvious e.g. in the references section of the technical agenda that was attached to FLI’s open letter. Out of a total of 95 references:
Six are MIRI’s technical reports that’ve only been published on their website: Vingean Reflection, Realistic World-Models, Value Learning, Aligning Superintelligence, Reasoning Under Logical Uncertainty, Toward Idealized Decision Theory
Five are written by MIRI’s staff or Research Associates: Avoiding Unintended AI Behaviors, Ethical Artificial Intelligence, Self-Modeling Agents and Reward Generator Corruption, Problem Equilibirum in the Prisoner’s Dilemma, Corrigibility,
Eight are ones that tend to agree with MIRI’s stances and which have been cited in MIRI’s work: Superintelligence, Superintelligent Will, Singularity A Philosophical Analysis, Speculations concerning the first ultraintelligent machine, The nature of self-improving AI, Space-Time Embedded Intelligence, FAI: the Physics Challenge, The Coming Technological Singularity
That’s 19⁄95 (20%) references produced either directly by MIRI or people closely associated with them, or that have MIRI-compatible premises.
I think you and Vaniver both misunderstood my endorsement of FLI. I endorse them not because of their views on AI risk, which are in line MIRI’s and entirely misguided in my opinion. But the important question is not what you believe, but what you do about it. Despite those views they are still willing to fund practical, evidence-based research into artificial intelligence, engaging with the existing community rather than needlessly trying to reinvent the field.
Agreed—or, at least, he underestimates how much FLI agrees with MIRI. This is pretty obvious e.g. in the references section of the technical agenda that was attached to FLI’s open letter. Out of a total of 95 references:
Six are MIRI’s technical reports that’ve only been published on their website: Vingean Reflection, Realistic World-Models, Value Learning, Aligning Superintelligence, Reasoning Under Logical Uncertainty, Toward Idealized Decision Theory
Five are written by MIRI’s staff or Research Associates: Avoiding Unintended AI Behaviors, Ethical Artificial Intelligence, Self-Modeling Agents and Reward Generator Corruption, Problem Equilibirum in the Prisoner’s Dilemma, Corrigibility,
Eight are ones that tend to agree with MIRI’s stances and which have been cited in MIRI’s work: Superintelligence, Superintelligent Will, Singularity A Philosophical Analysis, Speculations concerning the first ultraintelligent machine, The nature of self-improving AI, Space-Time Embedded Intelligence, FAI: the Physics Challenge, The Coming Technological Singularity
That’s 19⁄95 (20%) references produced either directly by MIRI or people closely associated with them, or that have MIRI-compatible premises.
I think you and Vaniver both misunderstood my endorsement of FLI. I endorse them not because of their views on AI risk, which are in line MIRI’s and entirely misguided in my opinion. But the important question is not what you believe, but what you do about it. Despite those views they are still willing to fund practical, evidence-based research into artificial intelligence, engaging with the existing community rather than needlessly trying to reinvent the field.