Thread by @scasagra

Deep Learning, or AI today, and it's wobbly foundations nautil.us/deep-learning-…

#quote Deep learning, which is fundamentally a technique for recognizing patterns, is at its best when all we need are rough-ready results, where stakes are low and perfect results optional. Take photo tagging.

#quote But the stakes are low—if the app makes an occasional error, I am not going to throw away my phone #endquote / Or in my case I get mad because some little mis-selections

#quote When the stakes are higher, though, as in radiology or driverless cars, we need to be much more cautious about adopting deep learning

#quote As AI researchers Emily Bender, Timnit Gebru, and colleagues have put it, deep-learning-powered large language models are like “stochastic parrots” repeating a lot, understanding little. (dl.acm.org/doi/10.1145/34…)

#quote What should we do about it? One option, currently trendy, might be just to gather more data. Nobody has argued for this more directly than OpenAI,

With the latest #chatGPT from #openAI could be tempting to write off "symbol-manipulation". But wait, I did not know that in 2021 in a competition called NetHack, a pure symbol-manipulation based system crushed the best deep learning entries, by a score of 3 to 1

here the report: nethackchallenge.com/report.html

One reason? deep learning algorithms are very good when they can extrapolate from past experience, but not when dealing with a novel situation. Also searching for past patterns is what make them prone to "bias"

any evolution in AI should mix the manipulation of large input of data, pattern recognition, and creation of new patterns, and logic symbol manipulation: a hybrid model. Why do the computer community were entangled in this battle: models versus symbol manipulation?

Good to read the full article and review the history of the development of computers and the different approaches to AI nautil.us/deep-learning-…