Econ prof at Oxford.
Machine learning, politics, econometrics, inequality, random reading recs.
My slides for today's ML & Econ group meeting:
"Double descent,” and what linear regression can teach us about the coming collapse (?) of the AI bubble.
https://maxkasy.github.io/home/files/other/ML_Econ_Oxford/double_descent_slides.pdf
(Based mainly on "Learning Theory from First Principles," chapter 12, https://www.di.ens.fr/%7Efbach/ltfp_book.pdf )
📕 🔊 "The means of prediction" is now available as audiobook! 🔊 📕
Recording of my talk on
"Machine learning, causal inference, and economics"
Very nice review of "The means of prediction" by @B_Eichengreen:
https://smartthinkingbooks.com/smart-thinking-books-interview-barry-eichengreen.html
(“one brain region for one character trait”). Hard to see a priori why that should be true for trained neural nets. Moreover, not clear why mechanisms should generalize between different nets.
I remain skeptical.
Mech interp seems predicated on finding underlying computations that are neatly separable, modular or sparse, similar to the (failed) Mendelian paradigm in genetics (“one gene producing one observable trait”) or 19th century craniology
What is “mechanistic interpretability” (MI) of neural networks?
This paper gives a great introduction:
https://arxiv.org/abs/2404.14082
MI empirically explores “how neural networks work,” in a spirit similar to neuroscience - w.out the ethical constraints on manipulating living brains.
📕 🔊 "The means of prediction" is now available as audiobook! 🔊 📕
Large scale conversion of warehouses into camps to concentrate migrants is taking place:
Large scale conversion of warehouses into camps to concentrate migrants is taking place:
What is the backdrop for recent protests in Iran?
Between concentration of economic power in semi-private foundations, intra-elite fractures, rising precarity, poverty, food and fuel prices, and state repression, this article provides excellent analysis:
Looking forward to the workshop on “Causal Machine Learning for the Social Sciences” at the Newton Institute/Cambridge:
https://www.newton.ac.uk/event/cifw05/
I will discuss the big picture of possible research agendas at the intersection of ML, causality, and econ:
https://maxkasy.github.io/home/files/slides/causality_ml_econ.pdf
Science personnel and funding gutted at astonishing scale in the US:
https://www.nature.com/immersive/d41586-026-00088-9/index.html
What is the intersection of machine learning, causal inference, and economics?
What are promising avenues for research at this intersection?
I just drafted a new talk exploring these questions:
https://maxkasy.github.io/home/files/slides/causality_ml_econ.pdf
Thoughts and comments welcome!
🎄 📕 Looking for a last-minute gift for the AI-curious or AI-worried person in your life? 📕🎄
Consider "The means of prediction"!
https://press.uchicago.edu/ucp/books/book/chicago/M/bo255887145.html
One more round of booktalks on “The means of prediction:”
Dec 2 UC Berkeley (Matrix):
https://matrix.berkeley.edu/events/maximilian-kasy-the-means-of-prediction-how-ai-really-works-and-who-benefits/
Dec 3 Stanford (Silicon Valley archive):
https://events.stanford.edu/event/max-kasy-in-conversation-with-rob-reich
Dec 5 UC Davis (econ).
Come by if you are in the Bay Area!
My colleagues are organizing a workshop on
*AI for economic research*
next March.
Paper submission is open now!
https://icms.ac.uk/activities/workshop/economicsai/
🕸️ 🔢 Presenting tomorrow at UCL: 🔢 🕸️
"Causal inference on endogenous social network formation"
https://maxkasy.github.io/home/files/slides/network_formation_slides.pdf