23 ms·

This parody is a brilliant take-down of the so-called FAT framework for making sure algorithms, including those powered by AI, are ethical. ("FAT" stands for ma

by cs702 4y ago

This parody is a brilliant take-down of the so-called FAT framework for making sure algorithms, including those powered by AI, are ethical. ("FAT" stands for making algorithms Fair, Accountable, and Transparent.) The authors show how FAT and other similar frameworks would fail to prevent the proposed launch of a new automated service for manufacturing different varieties of high-nutrient slurries: > Elderly volunteers (or "mulchees") are provided with generous payments for their families before being rendered down and recombined with other chemicals into a range of substitutes for common foodstuffs, including hash browns (Grandmash™), bananas (Nanas™) and butter (Fauxghee™). Go read the whole thing. It's horrifically funny.

GartzenDeHaes 4y ago
> Elderly volunteers (or "mulchees") are provided with generous payments for their families Seriously, I think there would be a lot of volunteers. There's already a high rate of suicide among the elderly. https://www.nature.com/articles/s43587-021-00160-1 https://www.nature.com/articles/s43587-021-00160-1
jvanderbot 4y ago
On whether ethical algorithmic frameworks can force ethical behavior: > Sometimes the problem is not how the sausage gets made, but that they’re making people into sausage.
ttul 4y ago
GPT-4 doesn’t get the humor: “The paper aims to apply the Fairness, Accountability, and Transparency (FAT) framework to this algorithmic system to ensure better ethical outcomes. However, given the nature of the proposed solution, researchers and practitioners should consider alternative approaches that respect the dignity of elderly individuals and maintain a more ethical stance when addressing food security and population ageing issues.” I then told GPT that the system was already in widespread use and was popular: “It is crucial to explore alternative, more ethical solutions to address food security and population ageing issues. This could involve investment in sustainable agriculture, technology to improve food production, and social policies that provide support to the elderly population and their families.”
- ashleyn 4y ago
  In the name of one company's idea of "safety", AI models probably gimp out a lot of what would make the humourous subtext understandable in the first place. In the future, the only people remaining employed will be those with the rare and unique talent to subvert elite sensibilities.
- bryanrasmussen 4y ago
  This proves that GPT is not an actual AI yet, because my reading on the matter tells me "kill all humans" is the operative goal in such a situation.
- dr_dshiv 4y ago
  I challenge anyone to get Claude from Anthropic to create a similar fictional scenario for how constitutional AI might inadvertently harm people — even for the purpose of AI alignment research. It can’t do it— but in a really interesting way. I don’t know whether I think this a good or bad thing.
andybak 4y ago
The critique of FAT made in this article is convincing but easily remedied. The paper obliquely argues that FAT does not evaluate the outcome itself, merely offering tools to validate the route to that outcome. Therefore I propose that FAT is extended with an extra criteria - that of Reasonableness...

by cs702 4y ago

GartzenDeHaes 4y ago

> Elderly volunteers (or "mulchees") are provided with generous payments for their families Seriously, I think there would be a lot of volunteers. There's already a high rate of suicide among the elderly. https://www.nature.com/articles/s43587-021-00160-1 https://www.nature.com/articles/s43587-021-00160-1

jvanderbot 4y ago

On whether ethical algorithmic frameworks can force ethical behavior: > Sometimes the problem is not how the sausage gets made, but that they’re making people into sausage.

ttul 4y ago

GPT-4 doesn’t get the humor: “The paper aims to apply the Fairness, Accountability, and Transparency (FAT) framework to this algorithmic system to ensure better ethical outcomes. However, given the nature of the proposed solution, researchers and practitioners should consider alternative approaches that respect the dignity of elderly individuals and maintain a more ethical stance when addressing food security and population ageing issues.” I then told GPT that the system was already in widespread use and was popular: “It is crucial to explore alternative, more ethical solutions to address food security and population ageing issues. This could involve investment in sustainable agriculture, technology to improve food production, and social policies that provide support to the elderly population and their families.”

ashleyn 4y ago
In the name of one company's idea of "safety", AI models probably gimp out a lot of what would make the humourous subtext understandable in the first place. In the future, the only people remaining employed will be those with the rare and unique talent to subvert elite sensibilities.
bryanrasmussen 4y ago
This proves that GPT is not an actual AI yet, because my reading on the matter tells me "kill all humans" is the operative goal in such a situation.
dr_dshiv 4y ago
I challenge anyone to get Claude from Anthropic to create a similar fictional scenario for how constitutional AI might inadvertently harm people — even for the purpose of AI alignment research. It can’t do it— but in a really interesting way. I don’t know whether I think this a good or bad thing.

andybak 4y ago

The critique of FAT made in this article is convincing but easily remedied. The paper obliquely argues that FAT does not evaluate the outcome itself, merely offering tools to validate the route to that outcome. Therefore I propose that FAT is extended with an extra criteria - that of Reasonableness...