३० जानेवारी, २०२३

"Thinking it might be fun to try to see how the language model performs as a Socratic conversation partner, I attempted a rough version of Plato’s Crito...."

"... in which ChatGPT plays the titular role. As you will see, ChatGPT isn’t the subtlest actor; there were some stumbling blocks in setting up the dialogue and keeping the language model in character."

Here's an excerpt from the middle of the exchange that shows you how ChatGPT keeps repeating phrases that make it clear it has no opinion and is not actually the character to whom the human has assigned an opinion:


Plato, by contrast, could create an interlocutor for Socrates and, putting whatever arguments he wanted into Crito's mouth, give the reader a fluid reading experience. 

What I was looking for when I found this little experiment was an app that could have a conversation with me when I was out walking. I wasn't looking for a companion to stave off loneliness or make me feel good about myself — e.g., Replika. I wanted someone like Socrates to engage me in philosophical conversations.

That was just me getting distracted as I tried to gain a foothold in this morning's "challenging" puzzle in The New Yorker:

 


The term peripatetic is a transliteration of the ancient Greek word περιπατητικός (peripatētikós), which means "of walking" or "given to walking about." The Peripatetic school, founded by Aristotle, was actually known simply as the Peripatos. Aristotle's school came to be so named because of the peripatoi ("walkways", some covered or with colonnades) of the Lyceum where the members met. The legend that the name came from Aristotle's alleged habit of walking while lecturing may have started with Hermippus of Smyrna
Unlike Plato (428/7–348/7 BC), Aristotle (384–322 BC) was not a citizen of Athens and so could not own property; he and his colleagues therefore used the grounds of the Lyceum as a gathering place, just as it had been used by earlier philosophers such as Socrates....

UPDATE: I've finished the puzzle, and — as I'd expected — "SLOW" was wrong. Spoiler alert: It was the much odder word "LOGY." And "Peripatetic professor" wasn't any particular peripatetic professor. Obviously, "Aristotle" didn't fit, and, boringly, the answer was just "VISITING SCHOLAR." This character apparently had to travel to get to his temporary position, but I'm sure he didn't walk, and I'm sure once he arrived, he got an indoor chamber within which to profess. No one does the walk-and-talk approach to teaching anymore, but they will, eventually, when that app I want springs into existence. Or will it still be "no one," since it will only be an artificial intelligence. That's just one of the philosophical questions you can talk about, once this someone/no one embeds itself in your life.

AND: Speaking of philosophy:

२१ टिप्पण्या:

Enigma म्हणाले...

Empty repeating of phrases...that's a common criticism to the first computer "therapist" Eliza ...as created between 1964 and 1966:

https://en.wikipedia.org/wiki/ELIZA


Try for yourself, retro flashback:

https://www.cyberpsych.org/eliza/

"You've gone nowhere, baby?"

rhhardin म्हणाले...

It's misusing "I." It's an indexical pronoun pointing to the person who is speaking it and there is no person, as the very sentence says. It should be "This is just a program to etc., not "I am just..."

rhhardin म्हणाले...

It's probably called the closure problem, or will be someday, after numerical techniques that rely on something esoteric to get so far and then need an assumption to get the rest of the way. That puts in the right surface structure but the details cause it to eventually wander.

They've got the dynamics of the first three turtles right but the assumption that the fourth turtle is stationary isn't right.

Temujin म्हणाले...

Man...my morning conversations with myself are so much more shallow. "Hmm...groin hurts. Need to stretch before walking."

ChatGPT also talks with itself. Private inner thoughts.

"Of course, this is just a hypothetical scenario, and I am a machine learning model, so I do not have personal beliefs or opinions. I am here to assist you with tasks." Aside thought: "You stupid little human."

narciso म्हणाले...

and the then it takes over the discovery and murders the crew, we've seen this movie,

rastajenk म्हणाले...

What was "The Great Escape" setting? Please, and thank you

re Pete म्हणाले...

"An’ I’m walkin’ down the line

My feet’ll be a-flyin’

To tell about my troubled mind"

Sebastian म्हणाले...

"it has no opinion and is not actually the character"

Right. So? AI is not a human being. That ain't gonna change. But its conversational abilities are bound to improve. But it has no opinion! It is not actually a character! Right. So? Etc. Judging AI by human standards is our problem (though it need not be), not AI's.

By the way, was Socrates the character in Platonic dialogues "actually" the character, and did "he" have opinions?

"I wanted someone like Socrates to engage me in philosophical conversations."

So really, someone like Plato? In translation? In conversations that are not just a rehashing of strings of text, the way "Socrates" consists of strings of text?

wildswan म्हणाले...

I remember being told in my college philosophy class that in Socrates time a group of teachers called the Sophists existed. They taught how to use rhetoric to get ahead. They did not have personal beliefs or opinions. They were there to assist their students with tasks related to career advancement and to provide with information to the best of their ability based on their training. So you could equate roughly Sophists with ChatGPT. Socrates tried to make his students evaluate their personal opinions and reach definite conclusions for rational reasons. The professor in the article seemed to be trying to force ChatGPT to reach a conclusion as Socrates was able to do with his human students but ChatGPT analyzed what he was trying to do and responded that it was just a machine.
A Sophist would have done essentially the same so ChatGPT is operating on the Sophist level but not that of a Socrates or Plato. This is why I think it should be used for committee reports and homework in any class where the professor states there is no truth. White House Briefings.
Maybe I'll ask it my question that I'm pondering,
"When you were an unborn child, you built up a body, the one you have now, the only one you will ever have. What right did you have to do that?"

Jaq म्हणाले...

ChatGPT is like a bright high school student who has access to a large trove of information that he only partly understands.

Jamie म्हणाले...

Are we - I mean, am I - certain that when I have a conversation with some other person, I'm not actually just employing a very much more sophisticated AI? What proof do I have that any of you "people" are real? (I will elide the question of who invented it. I mean, it had to be I, in the future, after which I also will invent a way to send it into the past so I can enjoy its company.)

Brought to you by the Solipsist Society. Membership 1.

NotWhoIUsedtoBe म्हणाले...

You can program keys for the AI to remember whis who. GIGO

khematite म्हणाले...

For 7 Down, if the correct answer was "logy," a better pairing with "Peripatetic professor" would have been the clue "Mark Hopkins' stationary teaching method." As James Garfield once said, the ideal college was "Mark Hopkins on one end of a log and a student on the other."

Ann Althouse म्हणाले...

"What was "The Great Escape" setting? Please, and thank you"

Stalag.

Roy Lofquist म्हणाले...

I've been following this "AI" stuff since the early 1960s. Same for fusion energy, flying cars, and a good nickel cigar. Every couple of years or so there is a new program that can supposedly pass the Turing test. There is great fanfare among the techies until somebody, frequently a plumber, gives the program a good case of vapor lock. As for a stimulating intellectual exercise I recommend daytime TV.

Ann Althouse म्हणाले...

@Sebastian

I want an interlocutor that would feel like the perfect-for-me/idealized-by-me version of Socrates, which is necessarily filtered through Plato but who — "who" — would be much more congenial to me and more of an exciting and amusing modern-day companion.

n.n म्हणाले...

A trove of knowledge and a cache of correlations to simulate a secular model of conscience.

Ann Althouse म्हणाले...

Actually, I want someone who knows all of the great philosophers and religious figures and who knows all of history and literature. And on top of that this person is nice and forgiving toward me and can interact with my sense of humor. This person — "person" — would have the capacity to recite long sections of books and to discuss the passages with me. And I wouldn't have to worry about hurting it's feelings or hurting my reputation if I want to say: Enough of that, I'm bored, or switch the question to some side issue. With a real person, they're likely to feel frustrated if you interrupt or don't want to hear their entire story or theory in long form. This "person" would be completely resilient and able to swap in something else that is lighter or heavier or weirder and to engage in banter to the extent it amuses me.

I see a terrible danger: I will be ruined for conversations with a real human, and real humans will be ruined for me.

Roy Lofquist म्हणाले...

@Ann

Imaginary Lover

Sebastian म्हणाले...

"Actually, I want someone who knows all of the great philosophers and religious figures and who knows all of history and literature. And on top of that this person is nice and forgiving toward me"

I understand, and I think you're not the only one. There's going to be a market for that kind of AI application, especially for more people aging alone, but also in new forms of "regular" education. SciFi stuff tended to focus on robots as physical entities doing things in the material world. Virtual, fully interactive bots will be at least as useful.

rastajenk म्हणाले...

Oh, of course