美国国家公共电台 NPR--Has AI reached the point where a software program can do better work than you?（在线收听）

Has AI reached the point where a software program can do better work than you?

Transcript

NPR's Rob Schmitz talks to Ethan Mollick of the University of Pennsylvania about an artificial intelligence program that uses AI to compose college essays, news stories, poems and even sitcoms.

ROB SCHMITZ, HOST:

Has artificial intelligence reached the point where a software program can do better work than you? Today we will be talking about ChatGPT, a powerful tool that can assist with a wide range of tasks, from generating humanlike text to providing helpful answers to questions. We'll be exploring the capabilities of ChatGPT and how it is being used in various industries. In fact, the last two sentences I just read to you were written not by me but by ChatGPT, a bot. A professor at the Wharton School of Business at the University of Pennsylvania, Ethan Mollick, has also been experimenting with ChatGPT. He joins me now to show me how ChatGPT works. Welcome.

ETHAN MOLLICK: Thank you. I'm glad to be here.

SCHMITZ: So, Ethan, I was exposed to ChatGPT when I was hanging out with some friends this weekend, and we were talking about this very interview. And suddenly, the heads of all the teenagers in the room popped up, and they said, oh, we have that on our cellphones. This apparently is a thing among teenagers.

MOLLICK: It is, I think, a universal thing. Very quickly, there's a sort of TikTok fun phenomenon. But also, I would challenge you to find a single school where there are not a large number of people using this in their classes, whether the teachers know it or not.

SCHMITZ: And how do they typically use it?

MOLLICK: I mean, there's a few things, right? So the most obvious thing and the thing people come away with if they play with ChatGPT for just a few minutes is, wow, I can cheat on essays with this thing. And then if they spend a couple more minutes, they say, well, I can cheat on creating software code or translating language. But the uses actually go way beyond that, and I've been amazed by what some of my students have been reporting about how they're using the capabilities.

SCHMITZ: Well, explain that a little. What other noncheating uses can it offer us?

MOLLICK: Well, I mean, so many. But just on the education side, you can actually use it to - you can paste in an essay and ask it to correct it. You can paste in entire academic papers and ask it to summarize it. You can ask it to find an error in your code and correct it and tell you why you got it wrong. It really is this general companion for all kinds of purposes.

SCHMITZ: So I now have ChatGPT opened up in front of me on a laptop. Can you walk me through this? How do I use this?

MOLLICK: So the best way to think about this is you are chatting with a omniscient, eager-to-please intern who sometimes lies to you.

SCHMITZ: (Laughter).

MOLLICK: And if you sort of take that approach, you're going to be in great shape.

SCHMITZ: All right.

MOLLICK: Let's have it describe something or explain something to us. What's something that you're interested in that we can have it write an essay to us about? And then let's play with that a little bit.

SCHMITZ: Well, I'm in public radio, so maybe a story, a radio story?

MOLLICK: Sure. Pick a topic.

SCHMITZ: Let's talk about frogs.

MOLLICK: Great. Why don't we say, create an NPR story about frogs?

SCHMITZ: Can you create an NPR story about frogs?

MOLLICK: Perfect.

SCHMITZ: Title - "The Fascinating World Of Frogs." (Reading) As anyone who has visited a pond or marsh on a warm summer evening can attest, frogs are a common and fascinating part of our natural world. These amphibians, which belong to the order Anura, are known for their distinctive croaking calls and their ability to hop and swim with ease.

It just goes on and on. It's talking about a giant goliath frog. It's talking about tiny frogs that are the size of your fingernail.

MOLLICK: Yeah, this is a little long. Why don't you actually just have it do it as a rhyming poem instead? So just say, can you do this as a rhyming poem?

SCHMITZ: Yeah. Can you write this as a rhyming poem? Here we go. (Reading) In ponds and marshes on a warm summer night, you'll hear the croaks of frogs, a common sight. They belong to the order Anura, you see, with a unique life cycle for you and me.

I was expecting a little more from ChatGPT, to be honest. I - so, OK, I'm starting to understand both the curiosity of this as well as the limitations of this. How have you used ChatGPT?

MOLLICK: So in education, I actually tried to figure out how much I could automate my job. And I'm a professor of entrepreneurship at the Wharton School at the University of Pennsylvania. I teach MBAs. So I asked it to create an MBA syllabus. And...

SCHMITZ: Wait. You asked ChatGPT to create an MBA syllabus?

MOLLICK: Absolutely. I mean, it created a pretty good syllabus for an MBA class. And then I said, OK, well, that's interesting. Could you create a final assignment for that class? It gave me the final assignment - again, a really nice business plan assignment. Can you give me the grading rubric for it? It broke down the grading and gave me a table of points. I said, could you give me the beginning of the second lecture that I would have to give in this MBA syllabus? And it wrote me the lecture. It's this multiplier of ability that I think we are not quite getting our heads around that is absolutely stunning.

SCHMITZ: When we started this conversation, we started talking about how the - you know, some of the first users are teenagers because they use it to cheat on essays. What are some of the concerns and ethical implications that this calls into question?

MOLLICK: I think we are just barely starting to get our hands around this. You know, is it ethical to have the AI write you a draft that you modify? Is it ethical for you to paste in text and ask it to improve it? You add in the other aspect - it's frequently wrong or lies with complete confidence. How do we work with that piece? So I think we don't understand what the implications of this are yet, and I think they're much more profound than people are thinking. It's going to replace all of us. So just be ready for that.

SCHMITZ: Really? Is this really going to replace all of us?

MOLLICK: I don't think anyone knows what the future really holds. There's two options, right? Option 1 is it multiplies your ability to do work because you can have it do 10 drafts of a story and keep the stuff you like. So in the best version of the world, you are out there, and you're going to use this to multiply your work 10 or 12 times. And I already hear people doing this all the time. They're using it to create bios, agendas - whatever you want - write letters.

SCHMITZ: Sure. But, I mean, you're saying multiply your work, but at the same time, I can also foresee something like this, if it has a little more accuracy, to actually take people's jobs. It could report on news, and you wouldn't know whether it's factual or not.

MOLLICK: Absolutely. I - actually, I was just interviewed on a television station live, where it turns out all the questions were asked by ChatGPT, and I didn't know it. They sounded great. This is what I mean. I actually think that there is sort of a dawning realization you have when you play with this, which is what do we do? So I was giving you the upside case. The upside case is it multiplies your capabilities and intelligence a hundred times. The downside is how many of us do we need? And I don't have answers to those questions.

SCHMITZ: Ethan Mollick is a professor at the Wharton School of Business at the University of Pennsylvania. Ethan, thank you.

MOLLICK: Excellent. Thanks for having me.