Good morning friends! Stream is starting in just a bit, come hang out & code with me. 😁💻
because, let's be honest, every single one of us has caused harm to others - knowingly or not. and we cannot allow ourselves to do better if we cannot even conceive of ways in which our actions can be harmful, because we are all well-intentioned, "good" people
I've also been doing some background reading and have found so far:
1) Most work on perception of animacy is done visually (Scholl & Tremoulet '00)
2) A lot is on children, with the assumption that adults already know that computers/robots aren't animate (Mikropoulos et al '03)
Alright y'all, for stream tomorrow we'll be getting back into live coding as I analyze the data from the survey I ran last week about people's tendency to assign agency/animacy to different types of language technology.
I do like the approach of training a detector along with the model given the clear malicious use cases. (And also, it looks like, not actually releasing the model?)
🎹✨ "When trained on speech or music, and without any transcript or annotation, AudioLM generates syntactically and semantically plausible continuations while also maintaining speaker identity and prosody for unseen speakers."
The FTC began the next step in its process of gathering information that will help inform whether to propose new privacy regulations for how companies can collect, use and share consumer data for business purposes.
@firstname.lastname@example.org has the story. https://buff.ly/3Bqtikq
tons of people across the state and across the country have been asking how they can help with the Jackson water crisis.. #jxnwatercrisis
here’s what i’ve compiled and will add to this week: 🧵
I know I'm not the first person to makes these points, and I sure hope I won't be the last, but it's just been weighing on me heavier lately given things like:
1. automated firings (https://www.businessinsider.com/facebook-contract-workers-accenture-austin-lost-jobs-2022-8)
2. this paper on direct algo. discrimination (https://onlinelibrary.wiley.com/doi/epdf/10.1111/1468-2230.12759)
Like, if you transfer this to another field, it's bonkers. Imagine a civil engineer being like "building bridges is my passion! but also it's not fault if this super experimental design crumples the moment someone steps on it. They ~should have read the EULA~🤪🤷♀️"
It really feels like the software & especially ML as a field wants its cake (to build things that change the world! huge paychecks! professional respect!) and eat it to (have no responsibility for the harms of their work! not have to consider social impact! no licensing!).
And it's more than just NLP. Software engineering is one of the highest paid kinds of "engineering" professions https://www.levels.fyi but isn't a profession. https://www.aies-conference.com/2018/contents/papers/main/AIES_2018_paper_43.pdf
A barber has greater personal liability than someone working in tech. http://pages.erau.edu/~kornecka/papers/21SSC_ethics.pdf
I know not everyone feels this way but I do see it a lot. I'm thinking in particular of 1) quotes in this article and 2) that in the metasurvey everyone believes NLP has the potential to be dangerous... but very few people want it to be legislated.
Been thinking a lot about responsibility, power & software engineering/research lately.
Are there other fields with similar levels of privilege/prestige/resources but a similar lack of belief in personal responsibility in the effects/outcomes of the work?
It's quite striking to have hundreds of conversations over the years with people in dozens of cities and counties who each imagine that their jail is the worst or unique. And to have so many people genuinely believing that the problem is somehow a few particular jail policies.
Data nerds of Twitter, I think we can collectively figure this out https://twitter.com/robinhanson/status/1567966707918209026
A brief explainer on why people might not like the British
Question: I'm planning on going over the data from this survey on stream Tuesday. Would you prefer:
1. Just me going through the results
2. A bit of live coding showing my analysis and visualization process? (All binary/categorical data can be ~a bit spicy~)
I've done up a quick five-question survey if you'd like to weigh in For Science:
I’m sure other cultures have a version of “if X gets cloudy it’s gonna rain” where X is some fat or oil. Y’all know of more?
Developer and data science educator who specializes in language technology. Making NLP boring. Linguistics PhD. Data science, NLP, Stats, ML, R, Python, FAccT. Tweets my own & CC-BY-SA. She/her.
The social network of the future: No ads, no corporate surveillance, ethical design, and decentralization! Own your data with Mastodon!