|
Tom,
I think you nailed it with this observation:
"It simply learns and statistically assesses what has been said about things in its training texts. It is the epitome of what goes wrong with our use of symbols when they zip around increasingly oblivious to the actual things they are ostensibly talking about."
I tend to think that what the AI/ML tool (ChatGPT in this case) does is use a very large number of dyadic operations to achieve the illusion of a triad. It responds to a very complex signal (that can be decomposed into simpler signals, etc) via the same stimulus/response
mechanism that causes a dog to come when his/her name is called. That's not to say that the illusion isn't powerful. It is. And it (and others) will likely only get increasingly "lifelike" in the future, ushering in an age of hyperreal nightmares that we're
only beginning to appreciate the dimension of.
cheers,
Will
From: Percy-L <percy-l-bounces+welewis=wisc.edu AT lists.ibiblio.org> on behalf of Thomas Gollier <tgollier AT gmail.com>
Sent: Saturday, February 11, 2023 11:17 AM To: Henry P. Mills <wppdirector AT gmail.com> Cc: Percy List <percy-l AT lists.ibiblio.org> Subject: Re: [percy-l] Is Walker Percy a Southern writer? FOR TOM GOLLIER (OFF LIST) Henry,
Thanks for forwarding my last message to the list. I was wondering what happened.
As to what Percy was "getting at" I look at it in the same vein as what the "movies are onto" before they screw it up. It is the everydayness of our knowledge, all of the generally accepted things we can say about things. And where hopefully there was
some direct acquaintance with the things themselves involved in our own knowledge as to what we can say about things, with ChatGPT there is virtually none. It simply learns and statistically assesses what has been said about things in its training texts. It
is the epitome of what goes wrong with our use of symbols when they zip around increasingly oblivious to the actual things they are ostensibly talking about.
All the objections I've been seeing to ChapGPT — can be wrong, no documentation, etc. — are things the program could easily be made to produce (if it were at all economically feasible). Train it on better texts, program in the sources, etc., and it would
be a match, if not better, for most, if not all, the purveyors of knowledge today. After all, we go to school primarily to learn how to talk about things. What Percy, at least for me, seems to be getting at is that all this learning what can be said about
something, these "vertical searches," are missing and obscuring something essential about why and how we use symbols in the first place.
In short, ChatGPT passes the Turing test with flying colors. So what's it missing? What exactly is peculiar to the human use of symbols? Can we even figure it out at this point?
Thanks,
Tom
On Sat, Feb 11, 2023 at 6:12 AM Henry P. Mills <wppdirector AT gmail.com> wrote:
|
Archive powered by MHonArc 2.6.24.