@mkirschenbaum @GaryMarcus @EnglishOER @markcmarino Actually both Gary and I neither assume--much less hope--that current "AI" is an end state. Far from it. And Gary in particular is very clear about what he thinks the real dangers are to technologies like those we have now.
@mkirschenbaum @GaryMarcus @EnglishOER @markcmarino What both of us share (and what Yann LeCun also agrees w/ these days) is that the current paradigm won't stop these kinds of mistakes. If you use ChatG what you'll see is that there are many questions it won't answer (the ones that are controversial). +
@mkirschenbaum @GaryMarcus @EnglishOER @markcmarino OAI opted to make ChatG more "lawyered up" than G-3 b/c they didn't want to be embarrassed by the mistakes and toxicities. +
@mkirschenbaum @GaryMarcus @EnglishOER @markcmarino For the same reason ChatG's text generations are pretty boring: they're always reaching for the commonplace so as to be accurate & non-toxic. +
@mkirschenbaum @GaryMarcus @EnglishOER @markcmarino But what's behind those corporate decisions is a technology that--in this form--will _never_ "know" the difference between a falsehood or not, racist or not, etc. +
@mkirschenbaum @GaryMarcus @EnglishOER @markcmarino It's virtue is to give you a prediction of what a good answer might be; which is why it excels at the kind of thing you might find in a simple wikipedia entry--except when it doesn't! +
@mkirschenbaum @GaryMarcus @EnglishOER @markcmarino The storehouse of errors that Gary and others have been keeping is not about nitpicking or "chiding." And it's certainly not naive. It's about demonstrating entire category errors of judgment that data-driven deep learning (i.e., statistical models) can't grasp. +
@mkirschenbaum @GaryMarcus @EnglishOER @markcmarino For the distinction I suggest reading Brian Cantwell Smith's _The Promise of Artificial Intelligence_ which is especially good IMO at theorizing what DL does well. +
@mkirschenbaum @GaryMarcus @EnglishOER @markcmarino But Joseph Weizenbaum made a very similar distinction in 1976 in a book that is well worth reading despite its age: amazon.com/Computer-Power…
@CriticalAI @mkirschenbaum @GaryMarcus @EnglishOER Yes, I am working on a book about ELIZA with Jeff Shrager, Anthony Hay, Art Schwarz, @berrydm, @peggyweil, Peter Millican, and @SarahCiston. We have been thinking about his response quite a bit, or what could be called Weizenbaum 1.0 and 2.0 #ELIZABOOK
@markcmarino @mkirschenbaum @GaryMarcus @EnglishOER @berrydm @peggyweil @SarahCiston I would _love_ to read that book. Please DM this account. (Note that this account is run by as many as 4 different people right now. But "I" - the editor-- am almost always the person you'll end up w/ on the weekends. :)