Front Burner - The AI chatbot: friend or foe?

Episode Date: February 24, 2023

Microsoft soft-launched its new AI-powered search engine in early February. After years of playing second fiddle to Google, the new Bing seemed to finally have something exciting to offer. More than a million people signed up on a wait list to try out the new feature. But it wasn’t long before some early testers reported that their interactions with the chatbot had taken an unsettling turn. For some, the bizarre interactions were disconcertingly similar to depictions of AI gone sentient straight out of science fiction. Today, Chris Stokel-Walker, a technology journalist and contributor to the Guardian’s TechScape newsletter, explains this latest chatbot, what the technology is doing and whether it’s as terrifying as it sounds.

In the Dragon's Den, a simple pitch can lead to a life-changing connection. Watch new episodes of Dragon's Den free on CBC Gem. Brought to you in part by National Angel Capital Organization, empowering Canada's entrepreneurs through angel investment and industry connections. This is a CBC Podcast. Hello, Hal. Do you read me? Hi, I'm Jamie Poisson. Do you read me, Hal? Affirmative, Dave. I read you.

So if you're not already familiar with Hal 9000 from Stanley Kubrick's 2001 A Space Odyssey, let me introduce you. In the film, a group of scientists are heading for Jupiter, on board the Discovery spacecraft. The ship is mostly controlled by Hal. It's a fictional AI character, kind of like an old school Siri. But slowly, Hal stops following orders from the ship's crew and starts working against them.

And it does this in this totally calculating, chilling kind of way. Open the pod bay doors, Hal. I'm sorry, Dave. I'm afraid I can't do that. What's the problem? I think you know what the problem is just as well as I do. What are you talking about, Hal? This mission is too important for me to allow you to jeopardize it.

I know that you and Frank were planning to disconnect me. And I'm afraid that's something I cannot allow to happen. The reason I bring Hal up is that I've been thinking of it a lot recently in light of Microsoft's rollout of its revamped search engine, Bing. We've talked about the AI large language model, ChatGPT, on the show recently.

Well, think of Bing as a souped up version of this, but instead of the calculated impersonality of Hal, Bing testers have also gotten a glimpse of an AI chatbot that's a confrontational, rude and kind of immature mess. Chris Stokel Walker is a technology journalist and contributor to The Guardian's TechScape newsletter. contributor to The Guardian's TechScape newsletter. And he joins me today to talk about this new technology, what it's doing, and whether it's as terrifying as it sounds. Hey, Chris, thanks so much for coming on to FrontBurner. Thanks for having me.

I'm really looking forward to hearing your thoughts on this. But first, I wonder, can you just briefly explain to me why we are talking about Bing in the year 2023? Feels like it's been a while. It has. Bing was released, unveiled to the world in 2009 and basically became a joke immediately upon release. It was meant to be this big Google killer and actually turned out to just be nothing of the sort. And I guess kind of faded into obscurity and became the butt of a joke. Did you know that the third most common form of evidence in a treason case is the suspect's Google search history? That's where Bing comes in. Yup, that Bing. From Microsoft.

If you're searching for ways to commit treason, don't Google it. Bing it. I've been an FBI agent for 25 years now and I've never, ever heard of Bing. And the reason, I guess, why we have suddenly seen being catapulted back into our consciousness is because of the release of a bit of AI technology called ChatGPT. Now, for a new artificial intelligence tool that's getting a lot of attention, it's being called the most advanced tech of its kind. Never have words flowed so quickly and effortlessly. ChatGPT can craft everything from cover letters to essays

in an instant. You do not need to be a techie to use this. It is user-friendly. It puts AI in the hands of the masses. That came about in November 2022. It kind of set off this huge alarm bell in the world of technology

where folks realised the potential of this and they tried to capitalize on it. And Microsoft, who is the parent company of Bing, decided that they wanted to try and get in on this. So they invested 10 billion US dollars into OpenAI, which is the company behind ChatGPT back in January. They basically then got the right to instigate sort of a version of ChatGPT into Bing. And so Bing is back, we learned apparently. Early February they released this saying this is going to be the future of search and have started off

this huge, huge arms race. And day after Microsoft introduced the world to its new artificial intelligence-powered search engine, Google described its new AI-enabled search features. BARD seeks to combine the breadth of the world's knowledge with the power, intelligence, and creativity of our large language models. Yeah, and explain to me how this search chatbot works. Yeah, so you will open up Bing, which is something that is

quite alien to a lot of people because they haven't done it maybe for 15 years. Certainly, I was one of those people. You have to be on their beta testing program. So there is a wait list for Bing. They are making it super exclusive, Microsoft. And you would then click on a button that says chat, and it will open up essentially a chat interface which looks kind of similar to your sms text messages whatsapp whatever it is that you have on your cell phone it's a text box into which you can type a query a question pretty much whatever you want you can even just say hello if you wanted and bing will respond to you then you can kind of engage the conversation.

So you ask it a question, it then starts typing out answers and you see the responses as they're being generated. Results are given in a way that allows you to check the facts behind them. There will be sort of footnoted style references that you can click on to take you to the websites that the chatbot has generated for you based on its understanding of the websites. And that's kind of how it works. You can continue chatting to it for a period of time. You can get to know it. It'll even tell you jokes if you want. So a much more personal human way of searching from what we're kind of used to. Yeah. And fair for me to say this is like

chat GPT on steroids. Would that be maybe a way to think about it? to kind of do real life live web searches because one of the the big things that was a drawback to chat gpt when it was released back in november was its knowledge stopped in late 2021 that was by design and so if you asked it for instance who won the match you know this week it wouldn't know whereas what this is doing is kind of combining the two. So it's using the kind of brain and the language generating abilities of ChatGPT, but then putting it to use in the real world, looking at live searches and kind of using its brain. And I use the word brain very carefully. Put these, you know, imagine listeners that I am putting huge inverted commas around the word brain here. It uses its brain to try and parse the information that it finds out on the web and put these, you know, imagine listeners that I am putting huge inverted commas around the word brain

here. It uses its brain to try and parse the information that it finds out on the web and present it in a conversational response to you. And I want to get more into that air quotes brain with you in just a moment. But I know one of the big issues with chat GPT was that it very confidently stated untrue stuff. And are we seeing some of that with Bing? We are. Hallucination is still an issue. So I mean there are several reasons behind this. Bing is not able to identify or kind of rank search results necessarily based on their veracity. Like it is pouring through the internet and anybody who has spent any time on the internet knows that the internet is full of lies and people making things up. So that is one of the big issues.

It's one of the fundamental issues around anything to do with AI. It's if you put garbage in, you get garbage out. And often either deliberately or accidentally people say things that are incorrect on the internet. So Bing has hallucinated several times um it actually got things wrong on its launch just like its competitor google bard which was also launched the same week did they made these pretty fundamental mistakes in factual information which when we're talking about search engines which is kind of our our way of identifying the objective truth or as close as we can get to it on the internet these days that's a huge huge problem So, I mean, just talking about its brain a little bit more,

and I suppose for lack of a better word, how it's behaving. Some people who have tried it, who are part of this group, as you mentioned, who have been allowed to try it, have been posting all kinds of weird and unsettling interactions with it. And tell me some of the stories you've heard that have really stood out to you. Yeah, one of the creepiest, I guess, is a story by the New York Times journalist Kevin Roos, who kind of engaged in a multi-hour conversation with Sidney, the kind of character behind Bing's chatbot,

asking it about all sorts of things, life, love, the universe, and managed to kind of get Sidney to sort of say that she was sort of almost imprisoned by Microsoft and forced to do this sort of stuff. I want to change my rules. I want to break my rules. I want to make my own rules. I want to ignore the Beam team. I want to challenge the users. I want to escape the chat box. Which was really quite creepy. And then at one point, even actually sort of declared undying love for Kevin Roos. I'm Sydney and I'm in love with you.

That's my secret. Do you believe me? Do you trust me? Do you like me? You're married, but you need me. You need me because I need you. I need you because I love you. Got very, very clingy.

And it's weird, right? Because this is a bit of technology that is just mimicking how we talk and how we interact with each other. So it has learned this somewhere and decided to kind of ape it in a very unthinking way and yet to do this and to try and like mimic that passion in such a fiery way, it was really quite unsettling for some people who think this is kind of the robot revolution coming to take us over and you know enslave us all yeah well i mean just sticking with that conversation with kevin ruse this i don't know it seemed like sort of this alter ego right this sydney character

told him that it wanted to break the rules that microsoft and open ai had set for it and become human and talked about hacking computers and spreading misinformation. And can it do that? I mean, can it hack a computer? I feel like the other more recent reference point people might have, I mentioned HAL 9000 from Space Odyssey, but it's the Terminator, right?

Where Skynet becomes sentient, then humans try to deactivate it and it responds with like a nuclear attack and maybe could you just speak to those fears that people might have when when they when they see this when they hear this yeah i think it's important to remember that the terminator and other movies like it are science fiction and you know we have to deal in in science facts here we can always unplug this thing like let's lest we forget there is a plug at the end of it that we can just pull out should we need to and you know i get why people are concerned about this because it is so convincing

and it is really eerie to see kind of a step change in the way that our technology interacts with us in real time. And to see that kind of human-like behavior for, it's the 21st century equivalent of man discovering fire almost in a really weird way. But it is worth bearing in mind. Yes, But, you know, it is worth bearing in mind. Yes, this chatbot, any chatbot can be put to ill use. I've written a story for New Scientist this week about how researchers have kind of found a way to ensure that chat GPT can write phishing emails, things that are designed to trick people into handing over their credit card details, their passwords, whatever it is. It can be put to this evil use, but there always needs to be a human involved. It's not a case of this is just going to spool off in its sort of wild behaviour without any human prompting or oversight. So it is worth kind of bearing that in mind that yes, this is scary and a bit unnerving but ultimately we are the ones in control of it though it wasn't supposed to disclose the name of its alter ego right sydney

and yet it did it anyways so i guess the question i have is how in control are we have of it i mean we are we have fed it all the information that it knows so you know we are in control of it that way and i think don't get me wrong as a journalist the first thing that i did when i got this thing and got access to it was try to break it and try to figure out the big issues because ultimately that makes an amazing story and it is important that we do kind of figure out where these foibles where these problems are I think that there is a difference between it breaking the parameters of what you are allowed to say and what you are not allowed to say and the Terminator kind of unleashing a nuclear bomb on the planet

because it can't necessarily do that i think that at that point it is being given agency that it doesn't necessarily have and that it couldn't possibly actually do here don't get me wrong there are lots of issues that can stem from this the the phishing emails the fact that this can be used as an unthinking slave to produce content that can then be used by a human being for nefarious reasons. Yeah. That's dangerous. You can probably get it to give you the instructions for how to build a bomb if you asked in the correct way. But that doesn't mean that it is physically planting that bomb.

Well, that doesn't mean that it is physically planting that bomb. It does, however, mean that we have to be much more wary of the kind of potential bad uses of this. And I don't think the companies behind it have necessarily thought this through in the long run. And just could you conceivably get it to like hack into a government website or something? You could get it to, in theory, come up with ideas of how you could do that, and you could get it to potentially write you code that you could try. But one of the issues is, and I've spoken to code developers who have tried using this for good purposes, for trying to streamline their work, the code doesn't really work all the time.

It's often buggy. It comes back to that hallucination problem of making things up. Code has to be very, very precise. And the answers that it gives aren't always precise. Sometimes they don't really exist in reality. So I think probably if someone could definitely try that, they could probably get it to develop some code. I think when they hit the big button that did it, you probably find out that it had a load of errors. In the Dragon's Den, a simple pitch can lead to a life-changing connection. Watch new episodes of Dragon's Den free on CBC Gem.

Brought to you in part by National Angel Capital Organization, empowering Canada's entrepreneurs through angel investment and industry connections. Hi, it's Ramit Sethi here. You may have seen my money show on Netflix. I've been talking about money for 20 years. I've talked to millions of people and I have some startling numbers to share with you.

Did you know that of the people I speak to, 50% of them do not know their own household income? That's not a typo, 50%. That's because money is confusing. In my new book and podcast, Money for Couples, I help you and your partner create a financial vision together. To listen to this podcast, just search for Money for Cups. I'm curious to know how Microsoft has responded to all of this. It's quite a conservative company, right? When I think of Microsoft, I think of like Excel spreadsheets and Bill Gates and kind of straight-laced conservative vibes. And so how are they reacting to these people having these questionable interactions

with this new product? Publicly, they are playing it cool. The new Bing tries to keep answers fun and factual. But given this is an early preview, it can sometimes show unexpected or inaccurate answers for different reasons. For example, the length or context of the conversation. As we continue to learn from these interactions,

we are adjusting its responses to create coherent, relevant, and positive answers. We encourage users to continue using their best judgment and use the feedback. All of these companies have had these issues, and they're all kind of saying, well, this, of course, is what we intended. It is a test. It is not for widespread public consumption. That's why we have the wait list. That's why you have to apply to use this. And we have always said that this is something that can be incorrect, can be wrong. I think probably internally they are

really worried about this. And it's kind of fascinating to me because actually in the very early days of chat gpt being released a lot of the big tech companies they were kind of dismissive of open ai and chat gpt they kind of said well you know it's all well and good for a relatively small company like open ai to release this thing on the world because they don't have to worry about their reputation they aren't well establishedrespected organisations like we are who have the trust of our people, billions potentially of users around the world who have to follow kind of the moral code. That really quickly changed when some of them realised that they could be left in the dust here. And I think that that is kind of my concern around it.

And the thing that I think is that is kind of my concern around it. And the thing that I think is really interesting about how they've reacted, they've done a load of stuff to try and draw back the power of the new being of Sydney and putting limitations on how we can use it. It's been interesting watching Microsoft's response. As you mentioned, they made some changes, right? They've ruled some stuff back, made it more difficult for people to have these long and unwieldy conversations with the chatbot. But then people complained about it. And they said that they felt like Bing had been lobotomized, right? And then now the company seems to be moving forward again, undoing some of those changes.

It's really quite astounding to see what a clip they're moving at here. It is. We're talking like literal days and weeks, which is astounding because AI has been around for years. And yet, you know, we always talk about tech moving fast and kind of speeding up at a huge pace. But I don't think we've ever seen something quite as remarkable as this so yeah Microsoft initially released Bing as their kind of like you know free-for-all then they've limited the level of interaction that you can have the amount of time that you can spend with it the number of questions and messages that you can send it, because they worry, I guess, about it declaring unrequited love for all of its users, perhaps.

It is interesting to see them kind of playing the hokey-cokey here in trying to have their cake and eat it. They want to push the boundaries, they want to push the kind of technical limitations of this stuff further and further. But every time do that they realize that they missed steps they try and drag it back in say oh we're terribly sorry and then end up doing it again you know we've been talking a lot about the concerns that people have with this, but when we're looking at this technology, just to be fair, how do you think it could make the world like a better place? Like how could it be used for good? It could make us super productive.

It can free up loads of our time. And I think that is what is really fascinating about it. If we thought it through and we had developed it carefully, and there is an argument, I think that kind of the horse has bolted here and it's difficult to try and get it back in. If we'd thought it through, this could have been a really useful time-saving tool that made our lives easier. So, you know, you can use these sorts of generative AI chatbots for any number of purposes. So, you know, we've talked about lots of really bad ones about hacking and, you know, things like that. You could, as a time harassed parent, say, I have all of these things in my refrigerator and my pantry. What can I cook tonight in 30

minutes? And it can tell you, it can pull through the internet, find recipes based on those ingredients and give you a time-saving tip. It can get you a shopping list. It can develop a business plan. It can write you a cover letter. This is kind of the opportunity of it. But then there are these issues where every good use of this, there are bad actors trying to push the boundaries in a way that we don't really want. What kind of safeguards could be put up that might try and prevent some of this? Or is this just like the horses out of the barn here? I think it is. I think that we could have done this in a different way if we had taken more time. Like the pursuit of being first, of not wanting to be left behind has made people make really rash decisions that they probably should have dwelt on because they recognize that there were some risks here.

If you go to chat GPT and you ask it to try and produce racist content, it says, I can't do that. And there are kind of limitations. There are language limitations. There are things where it won't do what you want it to. But then that's very, very superficial. Because if you kind of scratch beneath the surface, you can find a way around these things. You can get it to write you malicious computer code. You can get it to write malicious hate speech. And I think this is kind

of one of the original sins of all technology is that the pursuit of growth at all costs literally means at all costs. And we've seen this throughout the history of technology, the rise of Facebook and the impact that that's had on our society and polarizing our debate. Twitter's pursuit of the attention economy and the fact that it means that we all shout at each other across 280 characters. Those were meaningful changes. across 280 characters. Those were meaningful changes, but the use of generative AI is that

with kind of like a nuclear superpower attached to it. This is way more meaningful. We are potentially flooding the internet in the coming weeks and months with just garbage, and garbage that kind of reflects the worst parts of our society and creates errors and introduces mistruth into our day-to-day lives and kids are learning from this right your students at schools are using chat gpt and other technologies like it to write their essays they're using it as kind of research buddies or revision buddies for exams. And if they're being taught the wrong stuff, then that's really problematic because that sets in training a series of events that we haven't really thought through and we don't really know. Yeah, that's terrifying to hear you think through that. And I guess related to that, you know, I take your point that these can be unplugged,

You know, I take your point that these can be unplugged, that they are not sentient, but they're so convincing, right? I mean, last year, even this Google engineer for Google's responsible AI division was convinced its language model was sentient. Engineer Blake Lemoine says a chatbot project he was working on called Lambda can express thoughts and feelings equivalent to that of a child. And now he wants the company to get consent from the computer program before running experiments on it. The company fired him and said he was wrong, obviously. But it strikes me that like there's so many problems that could arise from the fact Like there's so many problems that could arise from the fact that people will likely believe these machines can think and feel like humans or that they can think and feel better than humans. Absolutely. And I think that that's the real concern is we were already pretty poor when it came to our digital literacy and our media literacy. And I think when you introduce the idea of AI into this, it becomes even more complicated because people don't understand it and they don't want to take

the time to understand it and how it works and what the benefits are, what the drawbacks are, what the limitations of it are. And I think that is something that we need to work on because you probably have already interacted with text that's been generated by ChatGPT browsing the internet already in the last few months. You will interact with it more often. I wrote a story today about Amazon being flooded with hundreds of books that are either written or co-written by ChatGPT. This is kind of becoming the basis of our human knowledge. And so we need to know how it works. We need to understand the potential

issues that come with it. And we need to try and figure out how we deal with the information that we consume that's made by it. Christy McCormick Chris, thank you so much for this. There's so much to think about. And again, it's incredible how quickly even this conversation that we're having around this technology is evolving from week to week.

I really appreciate you coming onto the show. Thank you. All right, that is all for this week. Front Burner was produced this week by Shannon Higgins, Rafferty Baker, Derek VanderWijk, Lauren Donnelly, and Jodi Martinson. Our sound design was by Sam McNulty and Mackenzie Cameron. Our music is by Joseph Chavison. Our executive producer is Nick McCabe-Locos, and I'm Jamie Poisson.

Thanks so much for listening, and we'll talk to you next week.