Attack of the chatbots? Our writer eyes humanity鈥檚 guardian role.
Predictive chat generates responses to human input that can seem human, with implications some tech-watchers call 鈥渁s big as the internet.鈥� It will take responsible intent, not just regulation, to temper this AI innovation.
What is intelligence?聽
No matter how you鈥檝e answered that question before, you鈥檙e likely to find yourself in even more conversations that reference the newest wrinkle in its artificial form: ChatGPT technology.聽
It鈥檚 a predictor, by definition, not a 鈥渢hinker.鈥� It showcases the responsive power of computer processing, not of sentience. But it鈥檚 showing up everywhere as more businesses apply it聽鈥� doing work for student essayists, making companion apps appear more human. One major drawback:聽Its use can complicate the fight against misinformation.
They鈥檙e 鈥渆xperimenting with all sorts of stuff,鈥� Laurent Belsie tells the Monitor鈥檚 鈥淲hy We Wrote This鈥� podcast. 鈥淪tuff that is sometimes ready for prime time and sometimes isn鈥檛.鈥澛營terations will improve it, he says. (Several have arrived聽this recent interview, and this week Google rolled out , Bard, while Microsoft to two browsers.)
Regulations will eventually add more guardrails.聽In his reporting, Laurent also notes humanity鈥檚 role as a shaper. Could predictive AI help us to hack, say, climate change?
鈥淚t all comes down to what the people are doing with the [technology],鈥� Laurent says, from the programming and testing to its applications. Then there鈥檚 intent.聽鈥淭here are lots of temptations out there,鈥� says Laurent. 鈥淏ut I鈥檓 hopeful that people in general, in business, will attempt to do the right thing.鈥�
Show notes
Here鈥檚 Laurent鈥檚 recent story on AI guardrails, discussed in this episode:
Here鈥檚 Laurent鈥檚 last appearance on this podcast, talking about the four-day workweek:
You can read more of his stories on his staff bio page.
Here鈥檚 a short Monitor explainer on ChatGPT:
Here are some other outlets鈥� early stories that we found useful: on AI, a profile of , chatbots and , an account of , a look at the intersection of this technology and , on r, the perils for , the threat of a US-China , and some linguists鈥� defense of .听听
Episode transcript
Clay Collins: Artificial intelligence has captured the popular imagination for decades, and has found its way into daily life through many practical applications. Alexa and Siri are almost family. Autofill features abound. Customer service chatbots give us quick, basic answers. And it鈥檚 actually getting more lifelike.
Enter ChatGPT. Those last three letters stand for 鈥済enerative pre-trained transformer.鈥� This technology takes existing knowledge and information and reshapes and reuses it in responsive, predictive ways that feel sometimes eerily like original thinking. Some people are calling ChatGPT as big a deal as the internet.
[MUSIC]
Welcome to 鈥淲hy We Wrote This.鈥� I鈥檓 Clay Collins.
Senior economics writer Laurent Belsie, who was last on this show talking about smarter ways to work, has been watching tech and AI trends for a long time. He recently wrote about the need for guardrails around AI鈥檚 use. He joins us again to talk about that and more.
Welcome, Laurent!
Laurent Belsie: Nice to be here.
Collins: First of all, I know I鈥檓 talking to you and not to AI because I can see you in the studio. But before we get into this disruptive technology and how you approach reporting on it, I want you to tell listeners about the interesting way in which you filed your guardrails story to your good-natured editors.
Belsie: Sure. Well there鈥檚 nothing like trying the product you鈥檙e writing about. What I did was, after gathering all my information and everything, I wrote my story as I would. And then I turned on ChatGPT, and I fed it my lede, the, the beginning paragraphs, because I wanted to make it somewhat like me. And then I fed it all the facts and said: 鈥淯se these facts.鈥� And then I fed it all the quotes I used in the story. I said, 鈥淯se these quotes.鈥� You wait about five to 10 seconds, and all of a sudden it starts writing. And it鈥檚 writing faster than I can type. And out comes this thing in like half a minute. It鈥檚 taken my story and, and turned it into a really abbreviated thing. So then I took both versions of the story, and sent them to the editors and said: 鈥淥K, you figure out which is my story, which is the ChatGPT. Now in fairness, it was pretty obvious. The technology isn鈥檛 quite up to writing really sterling copy. It took out all the best quotes. The punchiest stuff, it just paraphrased. Bad move. But there was one paragraph that I have to admit improved on. It made it snappier, shorter, and it made it clear. So the technology does show promise. We shouldn鈥檛 be too dismissive of the advances that are coming.
Collins: You wrote this story in part because of how buzzy this topic is. Bitcoin was buzzy too, but it felt like it was easy to sidestep if you weren鈥檛 that interested. This new surge in AI feels different. It鈥檚 selling cars in the metaverse. It鈥檚 creating visual art and (sometimes) passable writing. In fact, it鈥檚 just a kind of brilliant impressionist, right? Can I ask you to describe basically how it works?
Belsie: Yeah. Imagine very powerful, powerful computers with the latest chips that go blazingly fast. And then imagine feeding that machine information that would take up about a quarter of the shelving in the Library of Congress. And what we call 鈥済enerative AI鈥� processes all that text, or images, or whatever it is, and then it generates an image, a text. So instead of thinking, most researchers think of it as being a very, very, very good predictor, because it鈥檚 got so much data, and has weighed that data, and taught itself, that it can then create text that can fool people into thinking: 鈥淥h! This was written by a person.鈥�
Collins: Hmm. It鈥檚 power as a helper 鈥� you鈥檝e described a little bit off mic 鈥� it鈥檚 the same thing that gives it power to do harm. So if you could talk a little bit about both the promise of predictive, generative and other emerging strains of AI, and some of the perils.
Belsie: Yeah. We can look at anything from new scientific discoveries, and the ability to predict all sorts of things. Maybe even figure out climate change. Who knows? And it can also help us at work. Imagine, for you and me, it might involve gathering far more facts than we could, then processing those facts to be able to present it to us, so that then we can write our story. That can happen in all sorts of fields. It could take away the drudgery of a professor, for example, who wants to stay up to date with the latest research, so he can rapidly keep up with what鈥檚 going on.
That鈥檚 wonderful. But of course there鈥檚 the dark side of all this. And where we would be most worried about would be fraud. You know, already people are fooled by emails, you know, asking them to give their personal information or whatever. Well, take that to a factor of 10. And imagine how sophisticated fraudsters could be if they knew far more about you and could instantly compress that information into something that looks very, very convincing.
Collins: Mm-hmm. I think you described it yesterday as feeling like a smart, articulate graduate assistant. Very persuasive, but of course also sometimes very wrong.
Belsie: Very wrong. And that鈥檚 the challenge because these machines, many of them are being trained on the internet, which is great, because there鈥檚 a lot of valuable information there. Unfortunately, there鈥檚 a lot of misinformation there too. And so the machine can make mistakes, and has, embarrassingly so, from the early versions that we鈥檝e seen being released.
Collins: Hmm. So how does the Monitor approach inform your reporting about a topic like this, that鈥檚 steeped in issues of ethics and trust? The arguments are highly charged, but your job is to be cool and constructive.
Belsie: Uh, yes. It鈥檚 exactly to be cool and constructive, and not be carried away by the hype, which there will be lots of from marketing departments, from, you know, any tech company. And just look at the promise and the need for guardrails. The challenge is that, you know, just as in any industry, the industry is moving rapidly forward. And it鈥檚 experimenting with all sorts of stuff. Stuff that is sometimes ready for prime time and sometimes isn鈥檛. But in true Silicon Valley fashion, you put it out there, it breaks, and then you fix it. You get all sorts of feedback, and then you come up with another version that鈥檚 better. And that鈥檚 how the technology improves. Eventually, regulation catches up with that, but it鈥檚 a period of months and years, sometimes decades, to catch up with all the inventiveness that is out there.
So in that interim, you want to make sure that companies are acting ethically. As one AI CEO told me, it all boils down to intention. And if your intention is to use this technology in the best possible way, to remove or alleviate bias, to educate people, or to show them increased possibilities for the financial decisions they have to make, for example. It all comes down to what the people are doing with the machine, and how they鈥檙e programming that machine, and testing that machine for even unintended errors.
Collins: Right. Market forces are going to create all kinds of demand for generative AI, and the dangers aren鈥檛 clear enough yet for lawmakers to formulate effective regulations. So what鈥檚 the next period going to look like? Is it chaos?
Belsie: Uh, it鈥檚 partially chaos. Yeah. Because all sorts of things are going to come out. We may have good intentions, but program for the wrong thing. But I鈥檓 hopeful that people in general, in business, will attempt to do the right thing. There are lots of temptations out there. There鈥檚 a lot of money at stake, because this is, you know, 鈥渢he next big thing,鈥� and possibly even more game changing than the internet. We鈥檒l have to see. And it will be tempting to take the shortcuts. But enough people have seen the power of this technology that they, I think, have the ability to make the best decisions they can at the time, and correct the errors as quickly as they can when those pop up.
Collins: Well, thank you, Laurent, for being here in real life to talk about this breakthrough technology. And I hope you鈥檙e right about your near-term prediction. Thanks so much.
Belsie: Happy to be here.
[MUSIC]
Collins: Thanks for listening. You can find more, including our show notes, with links to the stories discussed here at csmonitor.com/WhyWeWroteThis, or wherever you listen to podcasts. This episode was hosted by me, Clay Collins, and produced by Jingnan Peng. Tim Malone and Alyssa Britton were our engineers, with original music by Noel Flat. Produced by 海角大神. Copyright 2023.
