"According to current neuroscience, human thinking is largely independent of human language – and we have little reason to believe ever more sophisticated modeling of language will create a form of intelligence that meets or surpasses our own." PAYWALL
https://www.theverge.com/ai-artificial-intelligence/827820/large-language-models-ai-intelligence-neuroscience-problems
Discussion
Loading...
@jackcole "According to current neuroscience, human thinking is largely independent of human language“
.. that statement feels like a category error to me….
@UlrikeHahn Perhaps. It is very difficult for an entity to render an unbiased judgment about itself.
@jackcole it’s also hard given the paywall 😉 but my speculative reconstruction given the quote is this: “cog, neuro, shows language and thought to be “independent” therefore language and thought are “independent”, therefore focussing so strongly on language….
if it does indeed say anything remotely like that, I think there is an equivocation on the term “independent” there that’s pretty crucial… a bit like saying
perception and reaching are ‘independent’ (because there are dissociable neural systems there…) therefore reaching doesn’t rely on perception….
@UlrikeHahn Your point is well taken. I had not considered the dependence of thinking on language, and will think about that. But obviously natural (animal) intelligence doesn't require nor is it capable of assimilating all digital data, nor do animals require fusion reactors. For LLMs and people "Learning without thought is labor lost." And primarily because of latency and to a lesser extent other factors, most digital data can never be assimilated by man or machine. This approach is doomed.
@jackcole If by 'doomed' you mean "insufficient to achieve AGI" that might well be (I certainly hope for all our sake, right now, that it is..), but I think it's also obvious that no approach could achieve AGI *without* an ability to process human language (certainly no system we could/would recognise as useful),
No approach in the history of AI has come even remotely close to LLMs ability to deal with language (whether we like that fact or not), so I'm not really sure where that leaves these kinds of arguments.
@UlrikeHahn It might be a mistake to only ascribe intelligence to machines and animals. No one knows, I believe, if for example microorganisms "think", by whatever definition might be used.As for "doomed", latency is generally considered an intractable problem, even if there were an infinite energy supply to pursue "intelligence" through LLM consumption amounting to omniscience (an unachievable superpower).