On the question, “What does morally aligned ai look like?” Some quick thoughts.
It probably depends on what the AI system looks like in general.
If agentic, with coherent, long-term, large-scale goals, the goals should emphasise delivering wellbeing and alleviating suffering across all sentient beings.
If following an ecosystem of services model, and constitutional, the constitution should emphasise not excessively harming sentient beings and commit to evaluating effects of its services on the welfare of important sentient beings.
If following an ecosystem of services model and not constitutional, I can think of two possibilities.
A centralized distributor of services might have a branch focused on setting global priorities and galvanising work in priority areas such as moral alignment. Depending on whether the centre is public or private, this global priorities task force might be embedded in a company or a governance institution like the US government or the UN.
AI systems themselves are trained across the board to take the welfare of all sentient beings into account, similar to current ethical AI efforts.
I'm curious if anyone's thinking more about this carefully!
Exciting! I'd be curious to hear more about your current projects and maybe help if possible! Is there a big-tent slack channel or something for work in this space?
What five (minimally) processed meat products would be the most important to include? [I didn't refer to a list other than the example list they gave, so some of these might not be appropriate] Chicken nuggets Bacon Shelled prawns/shrimp fish slices with a coating raw seasoned chicken portions
If we did not use a list approach, please describe any alternative approaches you would propose to define which minimally processed meat products are included? Prioritise labelling meat products that require a lot of animals per product, e.g many shrimp produce one serving of shrimp, whereas a small amount of a cow produces one serving of beef.
33 a) . Do you agree that method of production labelling should be mandatory? Yes 33 b). Please explain your answer. If you answered no, please detail any alternative approaches that you feel would be effective in delivering informative, consistent and accessible information on method of production to consumers Many food production methods directly cause significant suffering for the animals involved, create environmental harms, and introduce health risks (e.g zoonotic diseases and antibiotic resistance). Given the interest people have in animal welfare (as shown by surveys reporting high willingness-to-pay for better animal welfare), as well as in environment and health, it's important to keep consumers informed about the industry their purchasing decisions support. It is also important to reward producers engaged in better practices.
34 a) . Do you agree that any new mandatory method of production labelling should apply to both domestic and imported products? Yes 34 b) . Please explain your answer Domestic and imported products should face equal requirements when brought to the UK market. Method of production labelling can be disadvantageous to some producers with lower production quality in terms of welfare and environment considerations. It would be unfair to the UK producers with low production quality to compete with imported products for which the same labels are not required, since those imported products may also have similar (or more damaging) production methods. Introducing mandatory method of production labelling also helps establish the UK as a leader in mandatory method of production labelling, and introduces economic incentives for such requirements (and their indubitable benefits) to arise in other parts of the world.
39 a) . How important do you think it is that a method of production label includes processed as well as unprocessed animal products? Important 39 b) . Please explain your answer Eventually, a method of production label should include processed as well as unprocessed animal products. This is two main reasons:
1) It is important not to invisibilise the inclusion of animal products in products that less obviously contain animal products, such as cakes containing eggs and milk.
2) Further processing of food tends to produce less healthy products, and is also a change that larger producers can introduce more readily than small-scale producers. Further processing should also not be an "escape route" for large-scale producers to avoid honestly portraying their production methods and experiencing the market's reaction to greater transparency.
However, I understand the importance of legislation being practical for the food industry, so I consider this an "important" rather than "very important" issue.
41 b) . If you would like to propose an additional priority for labelling, please state below. Shrimp -- including minimally processed: e.g shells removed; breaded. Wild caught fish -- including minimally processed: e.g tinned; breaded; fish fingers; salted; marinated. Farmed fish -- including salmon, tuna, trout; same as above. Chicken -- nuggets.
42. If we did not use a list approach, please describe any alternative approaches you would propose to define which minimally processed meats are included If recognisable as coming from a single unprocessed animal product, the minimally processed meat should be included.
47 a). To what extent do you agree that standards should be based on inputs which are important for welfare, given the lack of examples of labels based on welfare outcomes and the additional supply chain complexity this would involve? Strongly agree 47 b) . Please explain your answer I strongly agree that standards should be based on inputs which are important for welfare. This fills a demand in the market for making ethical choices between food products. Though supply chains are complex and there are few existing examples of labels based on welfare outcomes, the UK can take this opportunity to establish itself as a moral leader in the food industry.
52. If you would like to suggest changes to the levels at which individual standards are set in the draft tiers, available in Annex B, please do so. For laying hens: I find it important that tier metrics should capture the experiences and welfare outcomes of animals. For instance, rather than making resources available to an animal, the animal should be in a condition to use those resource. Rather than prohibiting a painful practice such as beak trimming, the source of excessive stress-induced pecking should be verifiably removed. I give two examples --Outdoor access and beak trimming -- below.
Outdoor access should be replaced with "Outdoor use". Not only should the outdoors be accessible, but the animals and the environment should be in such a condition that the outdoor access should be voluntarily used. The way to verify this is to actively check whether a randomly selected chicken is found to use the outdoor space for at least one several-hour period each week in her life, weather permitting. After all, there is no use having a door to the outside if the outside environment is not suitable for use, or if the animals are not in a physical and mental state to use it. This could be said for use of destructible material and dustbathing area as well.
Similar for Beak Trimming in the "highest" category of Annex B for Laying Hens. There, it is written that beak trimming is not allowed because environmental factors should ensure that hens do not excessively peck each other. There should be a more verifiable metric, such as "pecking does not lead to infection for 95% of the flock."
Similar may be said for the meat chickens and pigs' welfare standards in Annex B.
For meat chickens and pigs: Similar sentiments as question 52 for laying hens.
Outdoor access should be replaced with "Outdoor use". Existence of enrichment materials should be replaced with "Use of Enrichment Materials." Not only should the outdoors be accessible, but the animals and the environment should be in such a condition that the outdoor access should be voluntarily used. The way to verify this is to actively check whether a randomly selected animal is found to use the outdoor space for at least one several-hour period each week in her life, weather permitting; or found to use the enrichment materials regularly. After all, there is no use having a these resources if they are not suitable for use, or if the animals are not in a physical and mental state to use them.
53 . Do you agree with the proposal above detailing the period of life covered by the draft standards for each sector? For laying hens: No, it is too short Please explain your answer I support inclusion of the period of life when hens are being reared as pullets (usually from day-old-chicks up to 16 weeks of age), as well as transport and slaughter, which can be extremely stressful. Regardless, I strongly believe it is better to have labelling encompass the proposed period of life than no labelling at all.
For meat chickens: No, it is too short Please explain your answer: I support inclusion of the period of life including transport and slaughter, which can be extremely stressful. Regardless, I strongly believe it is better for labelling to encompass the proposed period of life than no labelling at all.
For pigs: No, it is too short Please explain your answer: I support inclusion of the period of life including transport and slaughter, which can be extremely stressful. Regardless, I strongly believe it is better for labelling to encompass the proposed period of life than no labelling at all.
57 a) . Do you feel the label should include terminology describing both method of production and level of welfare: Yes, both.
58. Please share any comments on label terminology options based on the draft standards in Annex B. This may include individual terms you feel should, or should not, be used. Please provide supporting evidence where available. Method of production term: At later stages of the labelling effort, I think it is very important to make clear to consumers what these methods of production entail. Consumers' expectations for what "free range" , "barn," and "organic" mean may differ starkly from what they actually mean. I urge welfare labelling to be clearer about the lived experiences of animals.
For instance, for laying hens in the UK, some labels include caged, barn, free range, and organic. When consumers consider "free range", they tend to imagine happy chickens stretching their wings outdoors and scratching in the dirt for almost their entire lives. In reality, according to the RSPCA, "Legal requirements for free-range eggs ensure a minimum amount of space and litter for the hens: no more than nine hens a square metre, 10cm of feeder a bird and one drinker for 10 birds." Guardian (2017). Though this evidence may be outdated, the fact remains that food labelling should be authentic. What consumers think the label means and what it actually means should line up. I urge mandatory labels to account for this in the future, either using more informative labels, or changing legislation itself.
In any case, I am very excited about this step towards mandatory labelling. Though some groups may argue that the rules aren't stringent and pro-welfare enough for the food systems we ultimately desire, I think it's important and worth celebrating that we are starting somewhere.
58 continued. Level of welfare term. Welfare standards should be more accurate. Animals at Standard, Improved, and High levels of welfare standards may still be experiencing debilitating injuries, pain, skin diseases, bullying, limited range of movement, and very limited expression of natural behaviours. It may be misleading to use the "high" welfare term. I recommend more accurate terms.
I also recommend terms that promote positive competition for every-improving welfare levels. Terminology should reflect that the labels are relative to other producers' practices, while also not misleading consumers about the animals' actual wellbeing. For instance, terms like Poorest, Poorer, Standard, Industry Leader and Industry Best may be better.
Welfare standards should be continually updated, where terms become associated with ever-increasing production practices. For instance, a term like "Industry Best" or "Highest" should be granted to no more than the top 10% of food products with the highest welfare standards. This encourages healthy competition between producers to produce and innovate for higher welfare products.
It is important that this update encourages an upward trend in welfare practices. That is, if welfare practices become worse over time for whatever reason, it may happen that the proportion of food labelled "Highest" in its reference class falls from 10% to 5%.
60a) On (d) a picture illustrating method of production: To what extent do you support the inclusion of a picture illustrating the method of production? Support. 60 b) . Please explain your answer. It is very important that the picture accurately represents the animals in the farm. If health problems, e.g broken bones, missing feathers, sores, inability to move, are common, this should be represented by the image. Many consumers are interested in the welfare of the animals they eat -- a picture of a sore, representation of a broken bone, may be unappetising but honest. "Nicer" images may be misleading.
62. Please share any comments you would like to make on the mocked-up example label. I think it's very important that the minimal outline should differ depending on the welfare level of the animal. If the welfare level is "Standard", a more representative illustration might include several chickens packed closely in a cage, rather than the more idyllic image portrayed above.
On the question, “What does morally aligned ai look like?” Some quick thoughts.
It probably depends on what the AI system looks like in general.
If agentic, with coherent, long-term, large-scale goals, the goals should emphasise delivering wellbeing and alleviating suffering across all sentient beings.
If following an ecosystem of services model, and constitutional, the constitution should emphasise not excessively harming sentient beings and commit to evaluating effects of its services on the welfare of important sentient beings.
If following an ecosystem of services model and not constitutional, I can think of two possibilities.
A centralized distributor of services might have a branch focused on setting global priorities and galvanising work in priority areas such as moral alignment. Depending on whether the centre is public or private, this global priorities task force might be embedded in a company or a governance institution like the US government or the UN.
AI systems themselves are trained across the board to take the welfare of all sentient beings into account, similar to current ethical AI efforts.
I'm curious if anyone's thinking more about this carefully!
Exciting! I'd be curious to hear more about your current projects and maybe help if possible! Is there a big-tent slack channel or something for work in this space?
I think you put the wrong link in the link to JamesOz's comment at the start of the post!
This blog post also comments on this question. I think it's got a lot of blind spots, but also a lot of good points! https://www.effectivealtruism.org/articles/the-expected-value-of-extinction-risk-reduction-is-positive#fn-fn-40
Good post! Also made me laugh xD
Good post, thanks for writing! I wonder if you've thought or read much about the ecosystem effects of eliminating the screwworm?
Thanks for writing this! I appreciate the clear thinking and the synthesis of the literature.
My very hastily typed up long-answer responses:
What five (minimally) processed meat products would be the most important to include? [I didn't refer to a list other than the example list they gave, so some of these might not be appropriate]
Chicken nuggets
Bacon
Shelled prawns/shrimp
fish slices with a coating
raw seasoned chicken portions
If we did not use a list approach, please describe any alternative approaches you would propose to define which minimally processed meat products are included?
Prioritise labelling meat products that require a lot of animals per product, e.g many shrimp produce one serving of shrimp, whereas a small amount of a cow produces one serving of beef.
33 a) . Do you agree that method of production labelling should be mandatory? Yes
33 b). Please explain your answer. If you answered no, please detail any alternative approaches that you feel would be effective in delivering informative, consistent and accessible information on method of production to consumers
Many food production methods directly cause significant suffering for the animals involved, create environmental harms, and introduce health risks (e.g zoonotic diseases and antibiotic resistance). Given the interest people have in animal welfare (as shown by surveys reporting high willingness-to-pay for better animal welfare), as well as in environment and health, it's important to keep consumers informed about the industry their purchasing decisions support. It is also important to reward producers engaged in better practices.
34 a) . Do you agree that any new mandatory method of production labelling should apply to both domestic and imported products? Yes
34 b) . Please explain your answer
Domestic and imported products should face equal requirements when brought to the UK market. Method of production labelling can be disadvantageous to some producers with lower production quality in terms of welfare and environment considerations. It would be unfair to the UK producers with low production quality to compete with imported products for which the same labels are not required, since those imported products may also have similar (or more damaging) production methods. Introducing mandatory method of production labelling also helps establish the UK as a leader in mandatory method of production labelling, and introduces economic incentives for such requirements (and their indubitable benefits) to arise in other parts of the world.
39 a) . How important do you think it is that a method of production label includes processed as well as unprocessed animal products? Important
39 b) . Please explain your answer
Eventually, a method of production label should include processed as well as unprocessed animal products. This is two main reasons:
1) It is important not to invisibilise the inclusion of animal products in products that less obviously contain animal products, such as cakes containing eggs and milk.
2) Further processing of food tends to produce less healthy products, and is also a change that larger producers can introduce more readily than small-scale producers. Further processing should also not be an "escape route" for large-scale producers to avoid honestly portraying their production methods and experiencing the market's reaction to greater transparency.
However, I understand the importance of legislation being practical for the food industry, so I consider this an "important" rather than "very important" issue.
41 b) . If you would like to propose an additional priority for labelling, please state below.
Shrimp -- including minimally processed: e.g shells removed; breaded.
Wild caught fish -- including minimally processed: e.g tinned; breaded; fish fingers; salted; marinated.
Farmed fish -- including salmon, tuna, trout; same as above.
Chicken -- nuggets.
42. If we did not use a list approach, please describe any alternative approaches you would propose to define which minimally processed meats are included
If recognisable as coming from a single unprocessed animal product, the minimally processed meat should be included.
47 a). To what extent do you agree that standards should be based on inputs which are important for welfare, given the lack of examples of labels based on welfare outcomes and the additional supply chain complexity this would involve? Strongly agree
47 b) . Please explain your answer
I strongly agree that standards should be based on inputs which are important for welfare. This fills a demand in the market for making ethical choices between food products. Though supply chains are complex and there are few existing examples of labels based on welfare outcomes, the UK can take this opportunity to establish itself as a moral leader in the food industry.
52. If you would like to suggest changes to the levels at which individual standards are set in the draft tiers, available in Annex B, please do so. For laying hens:
I find it important that tier metrics should capture the experiences and welfare outcomes of animals. For instance, rather than making resources available to an animal, the animal should be in a condition to use those resource. Rather than prohibiting a painful practice such as beak trimming, the source of excessive stress-induced pecking should be verifiably removed. I give two examples --Outdoor access and beak trimming -- below.
Outdoor access should be replaced with "Outdoor use". Not only should the outdoors be accessible, but the animals and the environment should be in such a condition that the outdoor access should be voluntarily used. The way to verify this is to actively check whether a randomly selected chicken is found to use the outdoor space for at least one several-hour period each week in her life, weather permitting. After all, there is no use having a door to the outside if the outside environment is not suitable for use, or if the animals are not in a physical and mental state to use it. This could be said for use of destructible material and dustbathing area as well.
Similar for Beak Trimming in the "highest" category of Annex B for Laying Hens. There, it is written that beak trimming is not allowed because environmental factors should ensure that hens do not excessively peck each other. There should be a more verifiable metric, such as "pecking does not lead to infection for 95% of the flock."
Similar may be said for the meat chickens and pigs' welfare standards in Annex B.
For meat chickens and pigs:
Similar sentiments as question 52 for laying hens.
Outdoor access should be replaced with "Outdoor use". Existence of enrichment materials should be replaced with "Use of Enrichment Materials." Not only should the outdoors be accessible, but the animals and the environment should be in such a condition that the outdoor access should be voluntarily used. The way to verify this is to actively check whether a randomly selected animal is found to use the outdoor space for at least one several-hour period each week in her life, weather permitting; or found to use the enrichment materials regularly. After all, there is no use having a these resources if they are not suitable for use, or if the animals are not in a physical and mental state to use them.
53 . Do you agree with the proposal above detailing the period of life covered by the draft standards for each sector? For laying hens: No, it is too short
Please explain your answer
I support inclusion of the period of life when hens are being reared as pullets (usually from day-old-chicks up to 16 weeks of age), as well as transport and slaughter, which can be extremely stressful. Regardless, I strongly believe it is better to have labelling encompass the proposed period of life than no labelling at all.
For meat chickens: No, it is too short
Please explain your answer:
I support inclusion of the period of life including transport and slaughter, which can be extremely stressful. Regardless, I strongly believe it is better for labelling to encompass the proposed period of life than no labelling at all.
For pigs: No, it is too short
Please explain your answer:
I support inclusion of the period of life including transport and slaughter, which can be extremely stressful. Regardless, I strongly believe it is better for labelling to encompass the proposed period of life than no labelling at all.
57 a) . Do you feel the label should include terminology describing both method of production and level of welfare: Yes, both.
58. Please share any comments on label terminology options based on the draft standards in Annex B. This may include individual terms you feel should, or should not, be used. Please provide supporting evidence where available. Method of production term:
At later stages of the labelling effort, I think it is very important to make clear to consumers what these methods of production entail. Consumers' expectations for what "free range" , "barn," and "organic" mean may differ starkly from what they actually mean. I urge welfare labelling to be clearer about the lived experiences of animals.
For instance, for laying hens in the UK, some labels include caged, barn, free range, and organic. When consumers consider "free range", they tend to imagine happy chickens stretching their wings outdoors and scratching in the dirt for almost their entire lives. In reality, according to the RSPCA, "Legal requirements for free-range eggs ensure a minimum amount of space and litter for the hens: no more than nine hens a square metre, 10cm of feeder a bird and one drinker for 10 birds." Guardian (2017). Though this evidence may be outdated, the fact remains that food labelling should be authentic. What consumers think the label means and what it actually means should line up. I urge mandatory labels to account for this in the future, either using more informative labels, or changing legislation itself.
In any case, I am very excited about this step towards mandatory labelling. Though some groups may argue that the rules aren't stringent and pro-welfare enough for the food systems we ultimately desire, I think it's important and worth celebrating that we are starting somewhere.
58 continued. Level of welfare term.
Welfare standards should be more accurate. Animals at Standard, Improved, and High levels of welfare standards may still be experiencing debilitating injuries, pain, skin diseases, bullying, limited range of movement, and very limited expression of natural behaviours. It may be misleading to use the "high" welfare term. I recommend more accurate terms.
I also recommend terms that promote positive competition for every-improving welfare levels. Terminology should reflect that the labels are relative to other producers' practices, while also not misleading consumers about the animals' actual wellbeing. For instance, terms like Poorest, Poorer, Standard, Industry Leader and Industry Best may be better.
Welfare standards should be continually updated, where terms become associated with ever-increasing production practices. For instance, a term like "Industry Best" or "Highest" should be granted to no more than the top 10% of food products with the highest welfare standards. This encourages healthy competition between producers to produce and innovate for higher welfare products.
It is important that this update encourages an upward trend in welfare practices. That is, if welfare practices become worse over time for whatever reason, it may happen that the proportion of food labelled "Highest" in its reference class falls from 10% to 5%.
60a) On (d) a picture illustrating method of production: To what extent do you support the inclusion of a picture illustrating the method of production? Support.
60 b) . Please explain your answer.
It is very important that the picture accurately represents the animals in the farm. If health problems, e.g broken bones, missing feathers, sores, inability to move, are common, this should be represented by the image. Many consumers are interested in the welfare of the animals they eat -- a picture of a sore, representation of a broken bone, may be unappetising but honest. "Nicer" images may be misleading.
62. Please share any comments you would like to make on the mocked-up example label.
I think it's very important that the minimal outline should differ depending on the welfare level of the animal. If the welfare level is "Standard", a more representative illustration might include several chickens packed closely in a cage, rather than the more idyllic image portrayed above.
Thank you both for your responses! Makes sense
Great attempt and wonderfully informative write-up. Thanks Koen and Ren :D