Status Updates From Ethical Machines: Your Conc...

Note (Chapters 1-3):

Ethics aren’t subjective. Ethical beliefs are subjective. What’s right or wrong doesn’t change, what you believe is right or wrong can change.

This discussion about the ethics committee reminds me of how accessibility is or is not reviewed

I like this point about AI being good at scale which turns a micro problem into a macro one

I feel like the author has a chip in his shoulder sounds Facebook. Which like valid

“Using a discriminatory AI will force the user to discriminate when they use that AI, even if they’re intending no harm.”

Ugh I want to learn more want the way business models influence privacy levels

The book also doesn’t talk about how slow it is to pass legislation, so even if the law exists, it might be too little too late

It’s very interesting that a lot of the lessons are around “bring in experts.” Is that because Blackman is an expert and thus is keeping his job by providing the advice to hire those like him? I say this partly in jest but also partly in sincerity because what happens when you automate your job away?

Showing the machine explanation if someone else makes the judgement call. Hm feels like something good to keep in mind for my job

The discussion about including people from marginalized communities in the development process makes me think about the massive pressure that comes from being that person in that room. If you are THE representative from X community purposefully there to add diverse perspectives, isn’t that just tokenization in a different font

I’m struggling a bit with the book because in purposefully being short and succinct, it misses the nuance

As someone who does user research, it’s interesting to think about turning to consumers for their opinion on AI ethics. One thing I’ve been taught is to take all user proposed feedback (such as the answer to “how would you like this to work”) with a grain of salt because self reporting is a bit biased compared to actions

I also found out that those “select all x images” do, in fact, train the AI

I’m already intrigued by differentiating AI for good and AI for not bad