Oh, Oh, Criticism of a Cabal Fave: Deep Learning Daubed with Doo Doo
March 14, 2022
Ask Dr. Timnit Gebru, “How’s that push back on a certain big outfit’s approach to smart software working out?” Then there are the assertions that entire conferences exclude certain information about smart software driving off the information superhighway. Finally, listen and you will hear rumblings about a cabal in smart software.
If any of these “trivial” points resonate, you may find “Deep Learning Is Hitting a Wall” interesting. I prefer the idea of a golfer stumbling into deep doo doo, but you can choose the metaphor that speaks to you.
The write up is lengthy. I will highlight briefly the points with which the essay ends:
- Computational effort to learn what can be expressed symbolically is not smart (sort of the opposite of smart software, right?)
- One-trick ponies struggle when asked to do another trick
- Alternatives like symbolic methods work well for certain situations, particularly those where deep learning goes off the rails and into the doo doo
- Black boxes mean that outsiders have no idea what’s going on, particularly when applied to ad matching.
Does this sound like flaws in a certain alleged monopoly’s method?
Stephen E Arnold, March 14, 2022
Comments
One Response to “Oh, Oh, Criticism of a Cabal Fave: Deep Learning Daubed with Doo Doo”
-
- Subscribe to Beyond Search
Feature archive
News archive
-
[…] a few things that seem obvious," Geoffrey Hinton, “Godfather” of deep learning … More: Beyond Search, High ROI Data Science, Michael's Newsletter, and Spaghetti Data Tweets: @garymarcus, […]