Read on Twitter
Listening to @timnitGebru talk over lunch today with some @MathWorks colleagues, including @anoushnajarian !
The talk is "Hierarchy of Knowledge in Machine Learning & Related Fields & Its consequences". 
Talk is starting off talking about how social science and history are missing from computational rigor. Citing @ruha9 's talk. at CLR 2020 (?)
Interesting point: talking about engineering and how it was related to war, and inventions, but no reference to William Shockley and eugenics, even though he was the same person inventing some of the tech. Important context missing in the engineering education.
"Some of these people who get the highest honors in these fields, advance some of the worst ideas of society." - @timnitGebru
"Feeling that science is done from no one's point of view, but that's false." She feels algorithmic harm and bias can originate from this point of view.
Being a "good scientist" was about being neutral, and just focusing on the science and not the ethics. This sounds very similar to a lot of the #libraryneutrality conversations that are still going on. Can't ever be neutral.
[Paraphrase] "If we can add physics and game theory to machine learning, why not history or other areas? Will just make it better." Definitely agree with this.
Such a good path to point out that's she's going down: If we add history or other social sciences into ML/AI, then need to decolonize that information and data. Such important and foundational work, especially when it all starts getting fed into these large systems.
Data collection from archives is brought up: Eun Seo Jo, an archivist at Stanford. "Strategies for Collecting Sociocultural Data in ML". Love this so much, as I am a library scientist and archivist.
She just gave a big shout out to LIS and archiving. Love it.
I feel like I've been talking into the void for years about LIS needing to be more integrated into the tech world, and it's great to see it being put into action.
One of the next points is about making sure there is domain expertise and context in consideration when talking about ML/AI. 100% on board with this as well. Can't have information without the context.
Funnily, just wrote a blog post about context and organizing, specific to race/racism: https://www.gameronomist.com/blog/xqr99c5tw0duldrm55eq45uhvx42r2
Hierarchy of knowledge: gathering and annotating knowledge isn't valued. Referenced earlier LIS that this is also gender bias because libraries are very female dominated.
Core LIS issues being widely talked about in the library community right now are not just library issues.
Core LIS issues being widely talked about in the library community right now are not just library issues.
Talk now is focusing on issues with ethics boards. It's nice they exist, but it's the same elite institutions with no diversity, with the same people that lead everything else. Perpetuating the same problems.
Hierarchy of knowledge is a result of power structures, and starts at a foundational level. Such a core idea.
"Categorizing things affects a society." - @timnitGebru summarizing "Sorting things out" by Bowker and Star. Such a succinct way to show the power libraries could be having because of the time and work put into categorizing and classification.
Overall such a great talk. So much good information from such a smart woman. And she's so charming and accessible on such a hard topic!
Recording of the talk from @timnitGebru that this thread is about. I recommend taking the time to watch, it was great.
