I appreciate y’all continuing to think about this, and this is a solid draft as it got my brain thinking which is what good drafts do.
some thoughts:
A) “Science and technology could accelerate farmed animal welfare work, especially through alternative proteins.” There seems to be implication here that powerful AI equates to a greater likelihood of the end of factory farming. Perhaps, but I think there is also an argument to make that conventional factory farming could benefit from powerful AI as much or more so (it doesn’t suffer from a number of limitations that alt-proteins face). To have more faith in powerful AI helping alt-proteins net more than factory farming, I would like to see a robust attempt at making the case that it will help factory farming more.
B) “As AI’s capabilities in causal inference and behavioral prediction grow, all kinds of high-quality, granular data (for example, on consumer behavior, psychological drivers, policy responses, and market trends) will be incredibly valuable. Rather than only measuring direct impacts, campaigns can be designed to capture rich behavioral data and test messaging at scale, feeding into AI models that refine strategies and uncover new leverage points. Done well, today’s direct work could reduce short-term animal suffering while also serving as the training ground for smarter, faster advocacy in the years ahead.”
I would love to see this expanded on as well because to me feels like these arguments at face value are getting into powerful AI as magic territory.
“[...] campaigns can be designed to capture rich behavioral data and test messaging at scale.” This implies there is rich-enough data that could be captured today. That seems far from obvious to me. Some of the reason the social media well-being research is so constrained is because of the privacy concerns that would come with getting rich enough data to answer some of the well being questions with confidence. And social media or digital environments generally are an ideal environments because there is so much data, but (A) it is not clear how useful all that data is at changing behavior, especially a complex behavior like voting or diet, and (B) a lot of the behavior the animal welfare movement is interested happens offline (we would need to make the offline word much more data rich in types and invade privacy even more) and (C) of the rich data that currently exists offline and online exists in various silos and combining them isn’t necessarily technological issues, but a political and social one in some aspects. In addition, it is also a problem within the animal welfare movement because it would take really good project and data management, which are skill sets that hard to come by.
But even if you had (A) incredibly rich behavioral data that was practically achievable to get and (B) at a cost our movement could afford it and had the operational capacity, it is not obvious the epistemic metaphysics of social science would allow for anything close to a social science Laplace’s demon, which using data to change voting and diet seems close to IMO. There might be hard limits to what is possible to know in the social sciences, and therefore ability to influence behavior, whether we are in a world where AGI exists or doesn’t.
An example of possible limitations would be how RCTs are required for studying a social science intervention in a statistically robust way, but also might be crippling said intervention’s ability from having even medium-sized impact in the world at scale, which is the scale we are interested in. See: Cause, Effect, and The Structure Of The Social World by Megan T. Stevenson: (paper) H/T Seth Green. I’m not saying that is the definitely the case but I don’t think it is safe to assume it isn’t, and is worthy of exploring before saying AGI would be incredibly valuable for the animal welfare movement’s ability to understand and influence behavior.
I appreciate y’all continuing to think about this, and this is a solid draft as it got my brain thinking which is what good drafts do.
some thoughts:
A) “Science and technology could accelerate farmed animal welfare work, especially through alternative proteins.” There seems to be implication here that powerful AI equates to a greater likelihood of the end of factory farming. Perhaps, but I think there is also an argument to make that conventional factory farming could benefit from powerful AI as much or more so (it doesn’t suffer from a number of limitations that alt-proteins face). To have more faith in powerful AI helping alt-proteins net more than factory farming, I would like to see a robust attempt at making the case that it will help factory farming more.
B) “As AI’s capabilities in causal inference and behavioral prediction grow, all kinds of high-quality, granular data (for example, on consumer behavior, psychological drivers, policy responses, and market trends) will be incredibly valuable. Rather than only measuring direct impacts, campaigns can be designed to capture rich behavioral data and test messaging at scale, feeding into AI models that refine strategies and uncover new leverage points. Done well, today’s direct work could reduce short-term animal suffering while also serving as the training ground for smarter, faster advocacy in the years ahead.”
I would love to see this expanded on as well because to me feels like these arguments at face value are getting into powerful AI as magic territory.
“[...] campaigns can be designed to capture rich behavioral data and test messaging at scale.” This implies there is rich-enough data that could be captured today. That seems far from obvious to me. Some of the reason the social media well-being research is so constrained is because of the privacy concerns that would come with getting rich enough data to answer some of the well being questions with confidence. And social media or digital environments generally are an ideal environments because there is so much data, but (A) it is not clear how useful all that data is at changing behavior, especially a complex behavior like voting or diet, and (B) a lot of the behavior the animal welfare movement is interested happens offline (we would need to make the offline word much more data rich in types and invade privacy even more) and (C) of the rich data that currently exists offline and online exists in various silos and combining them isn’t necessarily technological issues, but a political and social one in some aspects. In addition, it is also a problem within the animal welfare movement because it would take really good project and data management, which are skill sets that hard to come by.
But even if you had (A) incredibly rich behavioral data that was practically achievable to get and (B) at a cost our movement could afford it and had the operational capacity, it is not obvious the epistemic metaphysics of social science would allow for anything close to a social science Laplace’s demon, which using data to change voting and diet seems close to IMO. There might be hard limits to what is possible to know in the social sciences, and therefore ability to influence behavior, whether we are in a world where AGI exists or doesn’t.
An example of possible limitations would be how RCTs are required for studying a social science intervention in a statistically robust way, but also might be crippling said intervention’s ability from having even medium-sized impact in the world at scale, which is the scale we are interested in. See: Cause, Effect, and The Structure Of The Social World by Megan T. Stevenson: (paper) H/T Seth Green. I’m not saying that is the definitely the case but I don’t think it is safe to assume it isn’t, and is worthy of exploring before saying AGI would be incredibly valuable for the animal welfare movement’s ability to understand and influence behavior.