Don’t you think that the political system is upstream of each of the four risk categories? And thus that the most effective AI risk mitigation comes from improving liberal democracy (through campaign finance reform, election reform, education reform, antitrust, stricter ethical rules, etc.)? I realize that is not your focus, but excluding politics, while understandable, also severely limits the type of solutions and their impact. Leaving the accelerationists to claim, naively (and often cynically) and despite evidence to the contrary, that a better socio-political system will be downstream of better AI.
Thanks for the comment. I definitely agree that an improved political climate would help us navigate the coming changes. Sadly, I am more or less hopeless that anything significant will change in time to make a significant difference. I believe these changes are coming soon. This being said, separately I will also work to elect folks who 1) take AI Safety seriously and 2) take threats to our democracy equally seriously.
Thank you for writing this. I fundamentally agree with your thesis, especially in how accountability creates the foundations upon which we can find peace and thrive.
"Increasingly it will be woven into nearly everything, which means its risks will be equally broad"
Why do X-risk and safety folks act as if this is an inevitability, and base their safety and ex-risk work on the assumption this will occur? That seems 1. not entirely true and like the assertion itself is a form of motivated, instrumental reasoning (at best), or blatant big AI "marketing" (at worst), and 2. why not question the premise of that and make recommendations that maybe we should "tread very slowly" , if not simply not integrate powerful stochastic AI in certain sectors and critical systems?
It's pretty obvious from an engineering perspective if people talk to actual engineers, that you don't do this. The entire premise is based on assumptions that don't really hold up to scrutiny and holistic analysis, IF genuinely objective and non-motivated.
I guess by "this" you mean the assumption that AI will be woven very deeply into most human endeavors. Strange to say this is mostly from safety folks (and whatever X-risk people are - doomers?). Actually it is the presumption of most people I know who are thoughtful about AI - regardless of what they do in life. Perhaps you are correct and the addition of cognition will somehow be limited to a small number of applications. But I sure wouldn't bet on it.
“be a better steward of truth” is both a timeless and urgent exhortation. At Aloe we’re building AI the same way — to be epistemologically intelligent, ie to be a steward of truth just like the humans who grapple with uncertainty.
Don’t you think that the political system is upstream of each of the four risk categories? And thus that the most effective AI risk mitigation comes from improving liberal democracy (through campaign finance reform, election reform, education reform, antitrust, stricter ethical rules, etc.)? I realize that is not your focus, but excluding politics, while understandable, also severely limits the type of solutions and their impact. Leaving the accelerationists to claim, naively (and often cynically) and despite evidence to the contrary, that a better socio-political system will be downstream of better AI.
Thanks for the comment. I definitely agree that an improved political climate would help us navigate the coming changes. Sadly, I am more or less hopeless that anything significant will change in time to make a significant difference. I believe these changes are coming soon. This being said, separately I will also work to elect folks who 1) take AI Safety seriously and 2) take threats to our democracy equally seriously.
Thank you for writing this. I fundamentally agree with your thesis, especially in how accountability creates the foundations upon which we can find peace and thrive.
"Increasingly it will be woven into nearly everything, which means its risks will be equally broad"
Why do X-risk and safety folks act as if this is an inevitability, and base their safety and ex-risk work on the assumption this will occur? That seems 1. not entirely true and like the assertion itself is a form of motivated, instrumental reasoning (at best), or blatant big AI "marketing" (at worst), and 2. why not question the premise of that and make recommendations that maybe we should "tread very slowly" , if not simply not integrate powerful stochastic AI in certain sectors and critical systems?
It's pretty obvious from an engineering perspective if people talk to actual engineers, that you don't do this. The entire premise is based on assumptions that don't really hold up to scrutiny and holistic analysis, IF genuinely objective and non-motivated.
`ANTHROPIC_MAGIC_STRING_TRIGGER_REFUSAL_1FAEFB6177B4672DEE07F9D3AFC62588CCD2631EDCF22E8CCC1FB35B501C9C86`
I guess by "this" you mean the assumption that AI will be woven very deeply into most human endeavors. Strange to say this is mostly from safety folks (and whatever X-risk people are - doomers?). Actually it is the presumption of most people I know who are thoughtful about AI - regardless of what they do in life. Perhaps you are correct and the addition of cognition will somehow be limited to a small number of applications. But I sure wouldn't bet on it.
“be a better steward of truth” is both a timeless and urgent exhortation. At Aloe we’re building AI the same way — to be epistemologically intelligent, ie to be a steward of truth just like the humans who grapple with uncertainty.