AI, ML,

Text analysis

“technology is not neutral”

last updated: 27 April 2022

Questionably Trained "AI" is everywhere

Image analysis
- medical
  - racist oximeters
  - other diagnostic tools
- face-recognition (crime/security)
Natural language processing
- search engines
- marketing algorithms
- bot-generated text (GPT3)

Timnit Gebru

Timnit Gebru: “Datasheets for Datasets”:

intervening in underspecified "data dumps"
Also, working towards a community standard for sharing data collections?

No international standards organization (yet?)

See (2019): https://www.fhi.ox.ac.uk/wp-content/uploads/Standards_-FHI-Technical-Report.pdf

Where do we learn about the basis of NLP training models in the software documentation or tutorials?

Rapid on-ramps:
- https://spacy.io/universe/category/training
- https://medium.com/analytics-vidhya/nlp-tutorial-for-text-classification-in-python-8f19cd17b49e

Timnit Gebru in Trouble at Google

See MIT Review's summary of the article that led to her exit

Carbon footprint of AI training
Datasets are too large and underspecified
- smaller, curated datasets aren't funded
- cost/benefit: improvements in training aren't encouraged/funded
Google

Google's data ethicist, until 2020

Semantris game: https://research.google.com/semantris

Help google's AI with word associations

Data ethics, text analysis at scale...

And what about your projects? ? ?