Hi, Tim. Thanks for taking the time to write. I think it’s fair to say that language has developed for people to talk to each other. Sometimes we talk to God, or dogs, or radiators, but they don’t talk back. So to be “in dialogue” with AI — or anything that resembles it — constitutes something new; for description, for grammar, for rather deeper concerns if that system resembles a person. As you said, people acted with ELIZA (that long ago, when it wasn’t fooling anyone, really) as though it was a person. So we’re in a new domain.
Consequently, our vocabulary — in my view — trails practice. To say GPT has “analyzed” something is about the only way to say it in casual conversation. And that drift, I agree, is and will grow more significant in time, in the way that we have long-since allowed “processing” to slip into “meaning.”
As it happens, though, it does “analyze” in the sense of coughing up material that allows me to see things I couldn’t see before.
To wit: The Jihadist ideology Sayyid Qutb wrote Milestones in 1964. It was extremely influential and partly explains Hamas’s actions on Oct 7 (ideologically, not practically). It provides a moral framework, a set of goals, a set of criteria for success and I could go on.
Qutb was very concerned about the term “Jahili.” It means “pre-islamic” but also savage, barbaric, and anti-Islamic. It is because the Buddhist statues were considered “Jahili” they were destroyed.
Judaism, however, also looked at the world before the 10 Commandments. I wanted to understand why Judaism didn’t treat the non-Jewish religions, ethical systems, artistic products (etc.) the same way. Why were they not threatening and slated for destruction as Qutb thought they should be for the same of Jihad?
GPT was able to provide some “compare and contrast” observations (obviously plucked), but those allowed me to focus my questions, which continued and this became a virtuous cycle until I reached a point where I wanted to land on a distinction in theology. I then dropped GPT and went back to Google and then finally to book (!) to flush out the details of the ideas.
This is why I used the phrase “conversational partner.” A conversational tool would have been more accurate, but note how stilted the language is and how we’re always inclined back towards anthropomorphism because … only people have talked back since we invented language!
Hi Derek,
At least you call it an it, even though it, ChatGPT, probably
gives you back text that appears to present it as some kind of
being with a self, which I strongly object to, but most other
people who I've talked to about what they do with ChatGPT seem
quite happy with, and some even like it this way: it makes
them "feel like they are being treated as a real person,"
which takes all the way back to Joseph Weizenbaum's ELIZA
program (1964-1967), and Weizenbaum's worries about how some
people responded to using ELIZA; worries I think we should be
having again, only a lot stronger. There's too much about
ChatGPT, and other Generative AI systems, that is deliberate
discerption on the part of the builders and providers of these
systems.
Nevertheless, I still think you misdescribe what is really
going on when you use ChatGPT, good use, useful use, effective
use, for you, you say, and I believe you. But, you explain,
for example, that ...
"The better you know the material, the better AI can become
a conversational partner in explore [sic] various
interpretations and ideas connected to your subject. This
is not about facts and details but understandings and
analysis.
ChatGPT does no understanding of anything, it does no analysis
of anything. It's a statistical machine, albeit, an enormous
one. ChatGPT doesn't read and understand your prompt text, it
converts it into mostly meaningless tokens, and from these, to
very large numerical vectors. There is only statistical
number crunching going on inside ChatGPT, tons and tons of it.
ChatGPT doesn't write words to say things to you with, it
generates text. Writing is what you do, and when you do this
you have some notion of what you want to say, or you use your
writing to work out what you want to say, from what might be
said, and from how it might be said. Writing is about
designing combinations of words that say what you want to say
to people, as you know. ChatGPT does no designing of any
kind; it does no writing. It doesn't need to be able to write
to be able to generate text; it just needs to run the
statistical model of text it's built to have in a generative
mode, with some 'sugar coatings' added on before giving it
back to you. All and any meaning you see in the text you get
from ChatGPT, or signs of understanding and analysis you see
in this text, are put there by you, not by ChatGPT. ChatGPT
doesn't do knowing and understanding and reasoning. It
doesn't need to do any of these things to generate text that
looks to you like a person might have written. ChatGPT
generates look-a-like writing, not really writing.
A better way of describing your useful use of, and productive
experience with ChatGPT would, I think, be to describe it as a
kind of mirror; a mirror that reflects back to you the text
you present to it, but which it reflects back with distortions
and artefacts on it; distortions and artefacts you find useful
when you are doing some writing, because of the meanings you
put on the text reflected back by ChatGPT. There is no
conversation going on here, and I think it's dishonest to
suggest, albeit mildly, that there is. There is plenty of
interpretation being done by you, yes, but none by ChatGPT. We
should not, I think, attribute capacities to ChatGPT it
doesn't have, and doesn't need to have, nor attribute doings
to it it cannot do. To think you have engaged in a
conversation with some other intelligence is an hallucination
on your part, and to suggest to others this is what you have
done is a delusion, and a dishonesty. You have engaged in a
monologue with yourself via a strange and sometimes beguiling
mirror of text.
-- Tim
Hi, Tim. Thanks for taking the time to write. I think it’s fair to say that language has developed for people to talk to each other. Sometimes we talk to God, or dogs, or radiators, but they don’t talk back. So to be “in dialogue” with AI — or anything that resembles it — constitutes something new; for description, for grammar, for rather deeper concerns if that system resembles a person. As you said, people acted with ELIZA (that long ago, when it wasn’t fooling anyone, really) as though it was a person. So we’re in a new domain.
Consequently, our vocabulary — in my view — trails practice. To say GPT has “analyzed” something is about the only way to say it in casual conversation. And that drift, I agree, is and will grow more significant in time, in the way that we have long-since allowed “processing” to slip into “meaning.”
As it happens, though, it does “analyze” in the sense of coughing up material that allows me to see things I couldn’t see before.
To wit: The Jihadist ideology Sayyid Qutb wrote Milestones in 1964. It was extremely influential and partly explains Hamas’s actions on Oct 7 (ideologically, not practically). It provides a moral framework, a set of goals, a set of criteria for success and I could go on.
Qutb was very concerned about the term “Jahili.” It means “pre-islamic” but also savage, barbaric, and anti-Islamic. It is because the Buddhist statues were considered “Jahili” they were destroyed.
Judaism, however, also looked at the world before the 10 Commandments. I wanted to understand why Judaism didn’t treat the non-Jewish religions, ethical systems, artistic products (etc.) the same way. Why were they not threatening and slated for destruction as Qutb thought they should be for the same of Jihad?
GPT was able to provide some “compare and contrast” observations (obviously plucked), but those allowed me to focus my questions, which continued and this became a virtuous cycle until I reached a point where I wanted to land on a distinction in theology. I then dropped GPT and went back to Google and then finally to book (!) to flush out the details of the ideas.
This is why I used the phrase “conversational partner.” A conversational tool would have been more accurate, but note how stilted the language is and how we’re always inclined back towards anthropomorphism because … only people have talked back since we invented language!