Science on Saturday lecture: How to Recognize AI Snake Oil.

[NNadir]

PPPL's science on Saturday lecture tomorrow morning by Princeton University Professor Arvind Narayanan is entitled "How to Recognize AI Snake Oil."

Sign up here:

Science on Saturday, on Zoom

Dr. Narayanan spoke a few years back when Science on Saturday was being held live at PPPL. I recall it as being quite interesting. (It's very, very, very rare when a PPPL Science on Saturday is not interesting.)

As the talks have moved to Zoom, they are now accessible around the world. They are held at 9:30 am EST, with a Q&A session after the talks ending usually by 11:30. The talks themselves are about an hour generally.

[Bernardo de La Paz]

1. Train a net to recognize snake oil

(probably one of the ways to recognize it is if someone claims you can train a net to recognize snake oil)

[Bernardo de La Paz]

2. PDF

https://www.cs.princeton.edu/~arvindn/talks/MIT-STS-AI-snakeoil.pdf

[Bernardo de La Paz]

3. Interesting. All boils down to slide 10, and I generally agree with the content of that slide

He says predicting social outcomes is fundamentally dubious and he is right, in my opinion, because social outcomes depend on neural networks (human) that are vaster than any ANN now or envisaged for some number of years. And even then, human neural nets have rights.

Genuine, rapid progress
•Shazam, reverse imgsearch
•Face recognition
•Med. diagnosis from scans
•Speech to text
•Deepfakes

Imperfect but improving
•Spam detection
•Copyright violation
•Automated essay grading
•Hate speech detection
•Content recommendation

Fundamentally dubious
•Predicting recidivism
•Predicting job success
•Predictive policing
•Predicting terrorist risk
•Predicting at-risk kids

Perception
Automating judgment
Predicting social outcomes