Thank you for your research!

Setting aside general arguments about companies' conflicts of interest regarding AI projections, I want to note that the revenue projections of these companies do not assume straight lines over trends. 

Different sources suggest OpenAI does not expect to be profitable until 2029, and its revenue projection for 2029 is around $100-120 billion. Similarly, Anthropic expects $34.5 billion revenue in 2027. These are very significant numbers, but for comparison Microsoft has an annual revenue of $250 billion. When I see the headlines "AGI by 2027", I expect something far scarier than $34.5 billion annual revenue. Of course one can argue that business deployment of AI takes time, companies can't capture all the value they produce and so on. Nonetheless, I think these numbers are helpful to keep things in perspective.

Thanks a lot for your comments Alex. I really appreciate it as I want to develop my thinking on topic. Thanks a lot for the suggestions as well.

There are some forms of agreements you can make with animals and there are some forms you cannot. I don't see why they can't intentionally coordinate based on understanding of our choices. A cow or a crow might move closer to someone giving them food and act kindly towards them later on, but they will refuse to move closer and cooperate if they realise that person has a history of deception.

There are also possible worlds in which animals' intelligence can be enhanced even further. It could even happen during our lifetimes given a technology explosion. In those possible worlds animals will be able to meet any threshold you want them to pass.

I really struggle to see a consistent way to be respectful towards people in coma or babies without also respecting the animals. You need a very specific argument on why both of these are true:

1. Being uncooperative to animals is fine even though they might become agents(according to your threshold) with some additional technology.
2. Being uncooperative to babies is not fine because many of them will become agents in future.

I believe the only consistent way to disregard animal interests is to deny that animals have interests at all as Yudkowsky does. As long as animals have interests it's very difficult to explain why screwing them over won't send a signal of "I might screw over others if I can get away with it".

Many advocates try to practice non-violent communication by merely criticising actions instead of individuals. But non-violent communication is stricter than that. Here's one example from "Non Violent Communication: A Language of Life". The anecdote starts with a cab driver making an antisemitic remark:

MBR: “You know, when you first started to talk, I felt a lot of anger, a lot of frustration, sadness and discouragement, because I’ve had very different experiences with Jews than you’ve had, and I was wanting you to have much more the kind of experiences I’ve had. Can you tell me what you heard me say?”
Man: “Oh, I’m not saying they’re all . . . ”
MBR: “Excuse me, hold on, hold it. Can you tell me what you heard me say?

Man: “What are you talking about?”
MBR: “Let me repeat what I’m trying to say. I really want you to just hear the pain I felt when I heard your words. It’s really important to me that you hear that. I was saying I felt a real sense of sadness because my experiences with Jewish people have been very different. I was just wishing that you had had some experiences that were different from the ones you were describing. Can you tell me what you heard me say?”

Man: “You’re saying I have no right to talk the way I did.” 

MBR: “No, I would like you to hear me differently. I really don’t want to blame you. I have no desire to blame you.”

In this section many advocates would not be happy with correcting the expression "You’re saying I have no right to talk the way I did.".

In general, I think the language used for rights based theories is in continuity with religious ethics in which actions are divided into "positive/neutral/negative" categories. When you do negative actions you incur some kind of debt and that debt should and will be repaid in terms of punishment in hell. Forbidden actions are forbidden because God backs them up with authority. He exercises punishment when people disregard his authority.

I think given the history and culture it's very difficult to divorce "You have no right to do that" from "It would be good if you were punished" and "Hey, people around, punish that guy and be happy when this guy gets punished".

To be fair, non-violent communication is pretty much against all morality statements. But I think welfarist language(makes happier, gives suffering, better things to do, worse things to do, results in worse/better/best/worst state of affairs) is less continuous with the religious tradition around debts and punishment.

You say:

"But this assumes the other party can engage in the kind of reciprocal decision-making that grounds such rights. Most animals' decision processes don't mirror ours in the way needed for this kind of relationship - they can't make or honor agreements, or intentionally retaliate based on understanding our choices. The question returns to welfare considerations: whether their lives are net positive."

I'm generally confused about this argument. If someone says "you can break your promises and violate trust if your counterparty is never able to find out or retaliate" they will get the retort "no no no decision theory doesn't work that way". If that person says "don't worry, I really picked out those that really are unable to retaliate. I will be super careful with folks who are able to retaliate.", they will get the response "nope, that doesn't work". 

When you hurt those who might have interests but are in no position to exercise agency, don't you send a similar signal that "if you're unable to understand what I'm doing, then I might hurt you". You might say "don't worry, I really picked out those that lack agency from birth, it will be fine" but I'm struggling to see a principled distinction.

I also think many animals have some capacity to communicate, understand promises, feel anger and spite over betrayal and hold grudges. It's possible to lie to many mammals and it's also possible to trade with them.

What is your bar for funding for some of the most common welfare interventions? On the margin, how many animals or animal-years should be affected per dollar for the following welfare improvements:

a. Cage-free transition for egg-laying hens

b. Stunning before slaughter for farmed sea bass and sea bream

c. Transition to ECC/BCC standards

Lewis Bollard:

"I agree with Ellen that legislation / corporate standards are more promising. I've asked if the breeders would accept $ to select on welfare, & the answer was no b/c it's inversely correlated w/ productivity & they can only select on ~2 traits/generation."

Thank you so much for this research. Is there a more intuitive way to interpret SMD values? For example, how many standard deviations is an average vegetarian away from the average person in the general population?