Cool, but I question the association with nutrition labels. Those have a specific, regulated purpose and this AI checklist doesn't. Is the nutrition label association supposed to subtly associate to some authenticity?
Yes, IMO. It’s supposed to be aspirational, I think — perhaps prompting the thought “if the government regulates what I eat, and I like that, maybe the government should regulate the software that controls my life…”
But the government does not really regulate what you eat. They require a nutritional label that has objectively true information on it. It's possible for others to check whether the information on the label is true.
Even then it's still a pinky-promise system. I had some food recently where the macros added up to 30% more calories than were listed on the label.
Truth is completely relative for the people doing the labeling.
The main things to remember is that a horrible ingredient may go by any of a dozen different names, and the ingredients too horrible to even mention can easily be swept under the rug with catch-alls such as "spices" and "artificial flavors".
In fact, that scary label on sea salt, This salt does not supply iodide, a necessary nutrient. is objectively false, because sea salt usually contains plenty, but if they don't artificially add it, they're required to lie.
This is a nice concise format to display model cards (https://arxiv.org/abs/1810.03993). I like the style and the "standard". But, I do agree with other commenters that the 'nutrition' term is probably inappropriate. We get the style from the card - you don't need to carry the analogy any further.
"The goal of the FactSheet project is to foster trust in AI by increasing transparency an increased understanding of how AI was created and deployed and enabling governance the ability to control how AI is created and deployed. Increased transparency provides information for AI consumers to better understand how the AI model a program component that is generated by learning patterns in training data to make predictions on new data, such as a loan application. or service an executable program, deployed behind an API, that allows it to respond to program requests from other programs or services* was created. This allows a consumer of the model to determine if it is appropriate for their situation."
That's all well and good, but it's still just taking the company's word for it. For something like this to be really useful, the claims at least need to be validated by some trustworthy group.
I like this! I think I’ll use it for my AI product. It actually really quickly answers a lot of customer questions that we have to address via FAQs currently.
Is this like Receiptify - presenting data in a kitchey format? I don't understand what it's parodying. Or is it just meant to be a standalone format for comparing AI privacy? Surely spreadsheets have already solved this more effectively.
As a software developer who interfaces with Twilio API from time to time, sure, okay I get it. It makes the pitch easier to whoever needs to sign off on the project that requires an AI service. I do feel like the nutrition label metaphor is a bit _much_, however.
"As a software developer who interfaces with Twilio API from time to time ..."
As a developer - and sysadmin - who interfaces with the Twilio APIs hundreds of times per day for both personal and business use I wonder why they can't just be a boring telco infra company and continue to improve upon the simple building blocks that aren't spam tools.
Not if the veracity and rules for creating these labels is held to a strict regulatory standard. We can certainly argue if the current food-based "nutrition facts" really provides all the information it should, but knowingly printing false information has clearly defined punishments.
