Alex,
I'm glad that we finally agree. The main problem with the LLM gang is that they don't ask the fundamental questions: How is this new tool related to the 60+ years of R & D in AI, computer science, and the immense area of the multiple cognitive sciences?
For example, Stanislas Dehaene and his students and colleagues have shown that there are multiple languages of thought, not just one. And every method of thinking has a different view of the world, of life, and of the fundamental methods of thought. For example, thinking and working with and about mathematics, visual structures, music, games, gymnastics, flying an airplane, building a bridge, plowing a field, etc., etc., etc. activate totally different areas of the brain than speaking and writing English.
A brain lesion that knocks out one region may leave other regions unscathed, and it may even enhance performance in those other regions. The LLM gang knows nothing about these issues. They don't ask the right questions. In fact, they're so one-sided that they don't even know what questions they should be asking. Somebody has to educate them. The best way to start is for us to ask the embarrassing questions.
Does the visual word form area split in bilingual readers?
Minye Zhan, Christophe Pallier, Aakash Agrawal, Stanislas Dehaene, Laurent Cohen
In expert readers, a brain region known as the visual word form area (VWFA) is highly sensitive to written words, exhibiting a posterior-to-anterior gradient of increasing sensitivity to orthographic stimuli whose statistics match those of real words. Using high-resolution 7-tesla fMRI, we ask whether, in bilingual readers, distinct cortical patches specialize for different languages. In 21 EnglishFrench bilinguals, unsmoothed 1.2-millimeters fMRI revealed that the VWFA is actually composed of several small cortical patches highly selective for reading, with a posterior-to-anterior word-similarity gradient, but with near-complete overlap between the two languages. In 10 English-Chinese bilinguals, however, while most word-specific patches exhibited similar reading specificity and word-similarity gradients for reading in Chinese and English, additional patches responded specifically to Chinese writing and, unexpectedly, to faces. Our results show that the acquisition of multiple writing systems can indeed tune the visual cortex differently in bilinguals, sometimes leading to the emergence of cortical patches specialized for a single language.
This is just one of many studies that show why LLMs based on English may be inadequate for ways of thinking in other languages or in non-linguistic or pre-linguistic ways of thinking, working, living, etc. Furthermore, language is a left-brain activity, and most of our actions and ways of behaving and working are right-brain activities. The current LLMs are based on ways of thinking by an English speaker whose right brain was destroyed by a stroke.
None of the writings about LLMs ask or even mention these issues. In this mini-series on generative AI, we have to ask the embarrassing questions. Any science that avoids such questions is brain dead.
John
From: "Alex Shkotin" <alex.shkotin@gmail.com>
JFS: "Now is the time to ask deeper questions."
Exactly, and these questions should be scientific :-)
And we have a scientific phase with these creatures, GenAI in general and LLM in particular: experiments ;-)
Alex