I encourage you to make your own mini-scenario (a couple pages long) that is basically an alternate version of AI 2027, but with more realism-according-to-you. Like, pretend you wrote a 100-page scenario branching off from AI 2027 at some point, and then actually write the 3-page summary of it. (I’m suggesting 3 pages to make it low-effort for you, the more pages you are willing to write the better)
In reading AI2027 I couldn’t help but to imagine an adjunct, dystopian storyline based on the scenario you provided. I’m a practicing psychotherapist and have already heard rumblings of my job being replaced by virtual therapists in the future. This is a terrifying prospect given the lesson in the swiftness and effectiveness of mind-control in citizens through politics over the course of the last 10 years. There is great good to come of AI in the future but we are on the cusp of releasing technology, akin to the nuclear weaponry of the 40′s, on a planet that is already on the precipice of collapse. So much more to say but I think it can be summed up by saying, “garbage in, garbage out” and we have yet to sort out the differences.
Talking about alternate realistic[1] scenarios, Zvi mentioned in his post that “Nvidia even outright advocates that it should be allowed to sell to China openly, and no one in Washington seems to hold them accountable for this.”
Were Washington to let NVIDIA sell chips to China, the latter would receive far more compute, which would likely end up in DeepCent’s hands. Then the slowdown might cause the aligned AI created in OpenBrain to be weaker than the misaligned AI created in DeepCent. What would the two AIs do?
I think that unrealistic scenarios like destruction of Taiwan and South Korea due to the nuclear war between India and Pakistan in May 2025 can also provide useful insights. For example, if we make the erroneous assumption that the total compute in the USA stops increasing and the compute in China increases linearly, while the AI takeoff potential per compute stays the same, then by May 2030 OpenBrain and DeepCent will have created misaligned AGIs and be unable to slow down and reassess.
I encourage you to make your own mini-scenario (a couple pages long) that is basically an alternate version of AI 2027, but with more realism-according-to-you. Like, pretend you wrote a 100-page scenario branching off from AI 2027 at some point, and then actually write the 3-page summary of it. (I’m suggesting 3 pages to make it low-effort for you, the more pages you are willing to write the better)
In reading AI2027 I couldn’t help but to imagine an adjunct, dystopian storyline based on the scenario you provided. I’m a practicing psychotherapist and have already heard rumblings of my job being replaced by virtual therapists in the future. This is a terrifying prospect given the lesson in the swiftness and effectiveness of mind-control in citizens through politics over the course of the last 10 years. There is great good to come of AI in the future but we are on the cusp of releasing technology, akin to the nuclear weaponry of the 40′s, on a planet that is already on the precipice of collapse. So much more to say but I think it can be summed up by saying, “garbage in, garbage out” and we have yet to sort out the differences.
Talking about alternate realistic[1] scenarios, Zvi mentioned in his post that “Nvidia even outright advocates that it should be allowed to sell to China openly, and no one in Washington seems to hold them accountable for this.”
Were Washington to let NVIDIA sell chips to China, the latter would receive far more compute, which would likely end up in DeepCent’s hands. Then the slowdown might cause the aligned AI created in OpenBrain to be weaker than the misaligned AI created in DeepCent. What would the two AIs do?
I think that unrealistic scenarios like destruction of Taiwan and South Korea due to the nuclear war between India and Pakistan in May 2025 can also provide useful insights. For example, if we make the erroneous assumption that the total compute in the USA stops increasing and the compute in China increases linearly, while the AI takeoff potential per compute stays the same, then by May 2030 OpenBrain and DeepCent will have created misaligned AGIs and be unable to slow down and reassess.