3 ms·

exactly. AI won't change the underlying incentives. We can't solve the problem of "alignment" between humans, so how can we ever solve it for AI? > Scientific

by lantry 1y ago

exactly. AI won't change the underlying incentives. We can't solve the problem of "alignment" between humans, so how can we ever solve it for AI? > Scientific progress is the biggest driver of overall progress; it’s hugely exciting to think about how much more we could have. Does he mean "how much more we could have if our federal government wasn't actively destroying our science institutions"?

__MatrixMan__ 1y ago
One could imagine scientific institutions that are less vulnerable to the whims of government. As loathe as I am to defend an Altman point, AI could play a role in making them possible.
- lantry 1y ago
  There are lots of things AI could do. But the things AI will do are governed by the incentives of the underlying system (i.e. society).
  - __MatrixMan__ 1y ago
    Those incentives are themselves the output of a variety of heuristics that we rely on only due to a lack of viable alternatives. I expect that eventually AI will make some alternatives viable because the bottleneck is so often an information processing one. Three examples: - Loans get issued when they're likely to make a banker rich. We know that's a lousy proxy for assurances that their associated ventures will be beneficial to the public, but it kinda sorta works enough. - Those loans cause money to enter the system and then we all chase it around and we cooperate with people who give us some of it even though we know that having money is a lousy proxy for competence or vision or integrity or any other virtue that might actually warrant such deference, but it kinda sorta works enough. - Research gets done in support of publications, the count of which is used to assess the capability of the researcher, even though we know that having many publications is a lousy proxy for having a made a positive impact, but it kinda sorta works enough. If we can fix these systems that are just limping along... If money can enter the system in support of ventures which are widely helpful, not just profitable... If we can support each other due to traits we find valuable in each other, not just because of some happenstance regarding scarcity. If we can encourage effective research in pursuit of widely sought after goals, rather than just counting publications... well that will be a whole new world with a new incentive structure. AI could make it happen if it manages to dispense with all these lousy proxies.
    - lantry 1y ago
      > AI could make it happen if it manages to dispense with all these lousy proxies. The thing that you, altman, and every other AI booster fail to explain is how AI will do this. Every argument is littered with "if"s and "could"s, and it's just taken for granted that AI will be smart enough to find a solution that solves all our problems. It's also taken for granted that we'll actually implement the solution! If AI said that curing cancer requires raising taxes, cancer would remain uncured.
      - __MatrixMan__ 1y ago
        I've got a few designs in mind for an alternative to scarcity based economics. I don't presume to have it right, but its worth a shot and its a bit more concrete than blind trust in one technology or another. I thought it would take my whole life to just lay the foundation, but with some AI help, maybe not. The trouble with you naysayers is you always take as immutable things that we can change.
        NoGravitas 1y ago
        We could have had meaningful post-scarcity economics in 1968 (read Murray Bookchin). It is not a technological problem. It is a social problem.
        __MatrixMan__ 1y ago
        That's a false dichotomy, technological advancement has been solving and causing social problems for as long as there has been technology.
      - Karrot_Kream 1y ago
        So I think this piece by Altman is mostly hype smoke also but I really dislike your framing here. Somehow the "AI boosters" have to come up with a how but the "AI doomers" can just sit back, criticize, and not solution? Let's be real here, US politics is fucked and none of us knows how to fix it.

by lantry 1y ago

__MatrixMan__ 1y ago

One could imagine scientific institutions that are less vulnerable to the whims of government. As loathe as I am to defend an Altman point, AI could play a role in making them possible.

lantry 1y ago
There are lots of things AI could do. But the things AI will do are governed by the incentives of the underlying system (i.e. society).
- __MatrixMan__ 1y ago
  Those incentives are themselves the output of a variety of heuristics that we rely on only due to a lack of viable alternatives. I expect that eventually AI will make some alternatives viable because the bottleneck is so often an information processing one. Three examples: - Loans get issued when they're likely to make a banker rich. We know that's a lousy proxy for assurances that their associated ventures will be beneficial to the public, but it kinda sorta works enough. - Those loans cause money to enter the system and then we all chase it around and we cooperate with people who give us some of it even though we know that having money is a lousy proxy for competence or vision or integrity or any other virtue that might actually warrant such deference, but it kinda sorta works enough. - Research gets done in support of publications, the count of which is used to assess the capability of the researcher, even though we know that having many publications is a lousy proxy for having a made a positive impact, but it kinda sorta works enough. If we can fix these systems that are just limping along... If money can enter the system in support of ventures which are widely helpful, not just profitable... If we can support each other due to traits we find valuable in each other, not just because of some happenstance regarding scarcity. If we can encourage effective research in pursuit of widely sought after goals, rather than just counting publications... well that will be a whole new world with a new incentive structure. AI could make it happen if it manages to dispense with all these lousy proxies.
  - lantry 1y ago
    > AI could make it happen if it manages to dispense with all these lousy proxies. The thing that you, altman, and every other AI booster fail to explain is how AI will do this. Every argument is littered with "if"s and "could"s, and it's just taken for granted that AI will be smart enough to find a solution that solves all our problems. It's also taken for granted that we'll actually implement the solution! If AI said that curing cancer requires raising taxes, cancer would remain uncured.
    - __MatrixMan__ 1y ago
      I've got a few designs in mind for an alternative to scarcity based economics. I don't presume to have it right, but its worth a shot and its a bit more concrete than blind trust in one technology or another. I thought it would take my whole life to just lay the foundation, but with some AI help, maybe not. The trouble with you naysayers is you always take as immutable things that we can change.
      - NoGravitas 1y ago
        We could have had meaningful post-scarcity economics in 1968 (read Murray Bookchin). It is not a technological problem. It is a social problem.
        __MatrixMan__ 1y ago
        That's a false dichotomy, technological advancement has been solving and causing social problems for as long as there has been technology.
    - Karrot_Kream 1y ago
      So I think this piece by Altman is mostly hype smoke also but I really dislike your framing here. Somehow the "AI boosters" have to come up with a how but the "AI doomers" can just sit back, criticize, and not solution? Let's be real here, US politics is fucked and none of us knows how to fix it.