How widespread access to ChatGPT and other AI tools is changing the world

Hello, I am Dan Marino in San Francisco. And I'm Mihal Elhadi in Toronto. Welcome to the Conversation Weekly.

Chat GPT. Everyone's talking about it. We had to talk about it too.

Nahal, have you played with Chat GPT at all? No, I haven't.

I feel like it's a bit murky for me as a journalist and a writer and an editor to venture into that territory.

Okay, that's a very respectable position. A lot of people are taking the opposite approach, it seems, and just jumping right in.

I've heard of ChatGPT being used as a pro bono lawyer. It's been taking classes at MBA schools.

And I recently read an article that it actually just passed a coding test for a $300,000 job at Google.

That sounds like an invitation to test it out. But an article that I worked on, the author had described chat GPT as autocomplete gone wild.

And I found that a really interesting way of thinking about that particular kind of technology and what it can and cannot do.

Yeah, this generative AI, right? It's chat GPT that, like you said, is kind of like autocomplete gone wild.

But it's also Dolly, this other AI algorithm that allows you to just input some text and then it'll spit out an image computer generated.

And there was a story about a person who illustrated an entire children's book using Dolly.

And that raised a lot of heckles amongst the artist community.

Yeah, there's been a lot of conversation about it in different industries and how it could be used and applied to make work more efficient, more accurate, more reliable.

But I think it also changes how we understand and view the information presented to us in the world. Absolutely.

And I think there's a really important question here that I haven't seen examined too much out in the media right now, is underlying all these potential changes is the question of access, right?

People now have access to powerful AI. And to me, this seems like a really interesting change.

So words like artificial intelligence and generative AI are now starting to become more prominent in our everyday conversations.

But what are we talking about here, really? I think you hit the nail on the head here, Nahal.

And that is exactly where we're going to start this episode, getting some definitions down and really explaining in a simple way how generative AI and AI more broadly basically works.

And to help us do that, I reached out to Daniel Acuna. He is a computer scientist at the University of Colorado in Boulder in the United States.

And he studies not only how AIs work, but also some of the problems that arise from the way these models are trained.

Generative AI has become really popular. And

the basic idea is that, I guess, traditionally, AI has been more about you give something to the computer and the computer makes a decision based on that input.

So that could be, for example, you take a picture with your phone and you want the computer to say, you know, what is in the picture?

Like what kind of objects are in the picture? Kind of like pattern recognition. I know that was a lot of the training of more classical AI.

So that's I would say that's a narrower definition of what the pioneers in the field wanted. So they wanted like a full blown intelligent agent.

But now people like really talk about AI.

But what they really mean, in my view, is more like pattern recognition, as you say, or machine learning.

So you have that, you have like an input, an image, and then you have, okay, which category, for example, that image belongs to? Is it a tree?

Is it a car? Things like that. Now, generative AI is a little bit more sophisticated in that you give the machine multiple points, let's say in this case, multiple images,

and then the machine will learn a representation of the images so that then you can say, or you can now go, and the computer will generate images.

So instead of saying this image contains this, it will learn from the data and it will start generating data that looks ideally like the data you fed into the machine.

Most modern, powerful AI algorithms, including generative AI, are trained using a process called deep learning.

You feed the algorithms a ton of data and let the models figure out the patterns within that data on their own.

But earlier AI systems required a much more hands-on approach.

So perhaps let me explain a little bit what happened before deep learning. How did we, because we still had like AI and machine learning back then.

And so how was that different? So in the past, how these AI and models will work is that you will have a lot of human intervention.

So if you want like a model for classifying objects in an image or generating data,

you will have to produce as a human, as an expert, lots of help for the machine. You will need to create what we call feature engineer.

So you wouldn't give the machine the raw image, like the actual picture from the camera, but you will like pre-process it with lots of clever, what we call filters or features.