Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Deep Learning with PolyAI

Deep Learning with PolyAI artwork

028: NLP vs NLP, or AI and the art of human persuasion

Apr 4, 202423 minAudio
ShareXLinkedInFacebookWhatsAppEmail

Nikola and Damien delve into the topic of humanizing technology and AI, touching upon the preference for non-human interactions in vulnerable states, the challenges of identifying automated systems during conversations, and the importance of transparency in AI interactions. Opinions on the potential cultural impacts of humanized AI, the debate around maintaining human-like experiences, and the societal and regulatory considerations surrounding the humanization of technology also steer the convo. Follow PolyAI on LinkedIn Watch this and other episodes of the Deep Learning pod on YouTube

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Nikolathe CSAT score was higher than the human agent on a comparable task. And that was also partially to do with the fact that. In those vulnerable states, I think people. actually prefer to speak to a non human, right? Because that vulnerability and exposure, it hits you, I think a bit less when you don't have that emotional reaction with the other person. But, you know, the thing you're talking to is in fact automated.

DamienHello friends and welcome back to a another enlightening episode of deep learning with Polly. I the only podcast with the CEO wearing a suit and a host wearing an athleisure hoodie. So here we are back with another episode. Nicola, thank you so much for joining us.

NikolaThank you for having me and for the shaming.

DamienGot to see Clark, you're looking dapper. I know it's all good things. We can't talk about it right now, but it's all good things. All good things,

Nikolagreat, great things, great things

Damienfascinating stuff. It's, I wanted to, even though you do look all business today, I want to take a higher level look at, no, no, definitely not. I just love it because this is, I think the first time that you've rocked a suit coat on on, on, on just, just to be honest with you. I really appreciate it. I obviously dressed up for you too. This is fairly clean. I do believe no, I wanted to take a look, we talk a lot about technology, what we do, the general state of AI, how it's applied to business. And that's all really important, well and good, but this is truly having some severe cultural impacts.

DamienAnd I say severe, it sounds probably a little more harsh than I intend in both the positive and what the potential is for, for the future. Allowing people to engage, yeah, with businesses, but with each other with resources in a way that's never been possible before. And part of the overarching goal, I think that's been pretty. Pretty galvanized in this big AI mission is the way that we get people to adopt this technology is to just make it as human as possible. Most business applications have some sort of humanization, whether it's bringing in LLMs or having some sort of engaging aspect of the tech itself.

DamienSo I want to take today's episode and talk a little bit more about that. Explore it, talk in the philosophical, more theoretical space, as opposed to the nuts and bolts and get your opinion as someone who knows academically this world inside and out and how, what you're seeing with regards to your friends and neighbors and coworkers with regards to them utilizing this technology and how it's transforming them as, as individuals. So, first of all, I wanted to talk a little bit about. About the concept of humanizing I'm going to start with our product.

DamienSo we obviously make the world's best customer led voice assistance for enterprise contact centers, right? Yeah. What would you say is the average amount of time it takes for the average caller who rings into one of our clients and gets met with a poly AI bot? With no disclosure at the front, identifying itself as a bot, how long does it take before they become aware that they're interacting with an automated system?

NikolaSome people never do, right? I would say that there's probably about a solid quarter to a third of the people who go through an interaction just not really thinking about whether it's human or not, and they might realize at the end, or they might have something that they might go back and forth. Let me play you something. Okay.

DamienHere we go.

YanHi, thanks for calling Bella Italia Telford. How can I help you today? Book a table. When would you like to come in? Today. What time do you think you'll be here? 6. 15. And how many people? Seven. Let me just have a

Nikolalook at what space we have. It looks like we've got nothing

Yan6. 15 p. m. Is there another day or time I can check for you? No, I'm so sorry. I thought I was talking to a robot, so I was being really rude. I do apologize. Sorry. No, that's all good. We're looking for a better bot now, so that's

Nikolano problem at all.

YanThanks

Damienthat is insane. So, this is a classic counterexample of this person thought they were talking to a bot, through the course of the conversation, they actually converted and thought that they were talking to a person. That's pretty funny. How many of those do we have? It is funny. Well,

Yanthese are less frequent, but I think they're

Nikolareally interesting. Sure. They get to a point where they're like, okay, like it's robot. It's not when they think it's a robot, they speak in this unnatural way. It's like, Hey, book a table, three people tomorrow. Like they want to make it easier for the system by saying fewer words. That's actually like wrong. And, I think as people understand that the more they speak, the language model kicks in disambiguates what you said so that in the end, you actually know what you actually. Get a better subscription if you speak for longer in a more natural way That's another reason why we like to make these conversations really really human but yeah, I think that If the system works, well, like our mantra has never been to trick people into thinking that right.

NikolaIt's a human It's to help them forget that they're speaking with a machine And we always have full disclosure. If you ask, we'll tell you, right? We do not try to deceive you.

DamienRight, and there's a lot of times there's regulatory requirements that would mandate that we say something up front depending on the region or whatever. This is,

Nikolathis is my pet peeve. At the moment there aren't really strict explicit legal requirements anywhere around doing this. Oh really? To the best of my knowledge. Legislation changes quickly these days. But a lot of people interpret legislation as they have to say it. And when they do, we recommend to people that they make it part of that recorded message in a different voice, because the more you draw the attention to the fact that it's automated, the more you invoke that legacy feeling of people think of all the times that they tried and failed to use a voice assistant and they're then

Damiencreated. So they've given up

Nikolaat the front. Yeah, yeah. They just like going nuts. Great video on our channel of people who are like, losing their mind when they hear it's automated. We, we love listening,

Damienhonestly, check that out after this episode.

NikolaWe should link it. And it's probably one of the most few videos we have. But you want people to not think about it. You don't want necessarily to trick them, but you just want to put them away from that. Legacy, cache of bad experiences that mean that they won't engage and give you a shot. And you know on the percentage of people who Know or don't know we don't obsess over it My intuition says that it's about a third but we have a pretty strong bias towards not Kicking up a fuss at the start because it's kind of like saying hey, here's a tv remote You turn it off by pressing this red button and you the more instructions you get The more people just in their mind, believe that it's not

Damiengoing to work. And also probably more inclined to push the red button. Yes. Yeah, so we obviously have a stance on this that says, if you, if you do disclose that up front, it taps into the predilections or the predispositions that people have toward automated systems. And it won't even, they don't even feel the need to give them a chance. Right. Do you think ultimately as these bots become more and more human, like in various channels, do you think that that transparency is ultimately going to be necessary to maintain or to build that sense of trust?

DamienOr do you think that we are moving to a seamless human like experience that's going to be expected? And so it doesn't really matter to the caller?

NikolaI think, I think for better or worse, legislation is moving towards a need. To you've had, I think the FTC recently Bam, like telemarketing, Robocalling all together. And the EAI Act has other provisions that will apply. My dream is to have a different ringtone that will tell you that the experience is starting in an automated way. I don't know if you've noticed, like when you dial into I think Google Meets and something's being recorded. I think they might've added the sentence that reads it out at the start. But what I always really liked is just the kind of like flashing red dot that tells you this being recorded, you make it obvious and and you join and the cognitive load of realizing that it's not high, but it's instant and obvious, right?

NikolaI think that's a good way. So we had a different ringtone. And we've done this in in a few deployments, where we. But we have a different ringtone. That's my dream for how this whole thing should work.

DamienIt's easily identifiable. It's, it's, it's a ringtone that we all know. We have a selection of ringtones on our own cell phone. We pick one that's to our personality. This is something that not only indicates to the caller, Hey, you're about to have a non human experience, which is called that way, but that it's also, it could be branded and attached to that experience. I think that's, that's a really cool concept and a very plausible future state. Yeah. So, okay. Let's talk about the other side. Let's talk about the non business application of humanization within AI. I don't know if you're familiar, but obviously with the critical mass adopting LLMs coming from open AI through chat GPT, getting their experience, getting used to what this new incredible tool and resource is about, and then instantly trying to break it and do what, teenagers and 20 somethings do, which is, Turn it raunchy.

DamienAnd so there's been a few iterations of what's being referred to as the Dan jail break, which is essentially it's just a string of prompts that has been iterated a number of times. Cause every time, whoever's putting these things together on Reddit it builds a better Dan than open AI comes in and shuts it down. So then they got to build a better Dan and so on and so forth. But essentially it removes all the safeguards. In a chat, GPT experience and gets it to pull no punches. So all the guardrails are off it. It's it's violence. It's racist. It's vulgar. It's self deprecating, et cetera.

DamienAnd of course, something like this happens. And a bunch of tick tockers are essentially chronicling themselves, developing these parasocial relationships, these feelings for this bot now, because it's the ultimate in human, right? It doesn't have those safeguards in place. That of course gives rise to the anti Dan movement saying. Robots should be robots and people should be people and never the twain shall meet. So if LLMs eventually get so, so good at without these jailbreaks, mimicking the emotionality of human conversation and human interaction, is this going to lead to ultimately a lot of misunderstandings and manipulated feelings and interactions?

DamienOr do you think it's going to go the other way, which is it actually builds a more just influential and a more. A more skilled human interaction for example there's a, there was a recent study from the Swiss Federal Institute of Technology that showed that essentially through a chronicle of experiments. That AI was essentially a better debater than most people, just because one, it had access to facts and two, the way that it assembled it and portrayed it embodied those formerly non robotic or non automated characteristics, the empathy the charisma and so forth as Michelle would say the Riz.

DamienSo where do you think the direction is going as like the emotionality is captured? You're

Nikolapulling on a few threads here. I don't know. I think that like, I'll put a poll like debating. It's obvious that it should be better, right? If you and I debated and after you had to go, I had, I don't know, 12 hours to think about a good response. I'd probably be a much better version of myself than if I just had to go straight after you, right? And if you and I were evenly matched and then I had those 12 hours to prepare every single one, I'd probably win every time because I have way more time to think. Right. And war game and plan. So, an AI has that type, right?

NikolaAnd it's because she have knowledge, facts and everything, it's absorbed all of human knowledge, right? So it can come up with an argument. I might've seen an argument being made already. It's it's got all decades, but probably millennia. So debating experience through the data it's consumed. So, yeah, you're in a bit of a pickle if you really have to

Damienwin, right? Well, but, but that sort of thing, when you're talking, the way that you're referring to it, I don't see that as being as much debate as it is like a trivia contest, right? You're not going to beat an encyclopedia of trivia. Just, it has more information than you most likely. And like you said, the access and the capability of accumulating that and synthesizing the proper output. It's just, it's never been where it is today. But, that's not what wins debates. What wins debates is that ability to influence. And now we're seeing example scientific studies of A. I. Having those typically human characteristics of again, like boisterousness, lack of ego and they're able to carry that forward to tap into the human being of human judges, which in the course of this experiment, it was graded by humans who was influenced by what means.

DamienSo where do you think that plays in the ultimate scheme of things? I

Nikolamean, I think it's interesting, right? Obviously that was done over text. So, it loses Yeah, that's true. A lot of the charm of real debate where you and I are talking, there's a lot of like comedy and the charisma and me twisting your words, you twisting mine and stuff. There's a lot of rhetoric. I've done a bit of debating when I was in high school and I would constantly, I'd constantly get disqualified cause I was, it was way too much. Just actually dissecting arguments and stuff.

NikolaAudiences love me because I provided entertainment. But really think it even in a debate, it really depends on like how you've set the scene. What's allowed? What's not? Is it being judged by judges or, or instructed to ignore? The kind of like gimmicks that someone like me would pull versus like genuinely looking at the arguments, it sounds like the way that that test was done, it was much more about the arguments where I think that that cognitive knowledge and stuff that I mentioned matters quite a bit, right? Because, maybe you carry the audience with your intellect and references and ability to, provide a good example that, you just know, because more, right?

NikolaI think in terms of emotionality, we. Had an interesting example with one of our clients where about 12 percent of the calls had to do with bereavement, inheritance, people talking about losing a loved one. And We built a sub experience where the tone really changes. It's very empathetic. It takes you through the whole conversation very calmly with a very somber voice all the way through. It's not just temporarily when you say that you've lost a loved one, but all the way to the end, very patient, slower, methodical.

NikolaThe CSAT score we got was higher than the human agent on a comparable task. And that was also partially to do with the fact that. In those vulnerable states, I think people. actually prefer to speak to a non human, right? Because that vulnerability and exposure, just the reliving whole ordeal, it hits you, I think a bit less when you don't have that emotional reaction with the other person. But, the thing you're talking to is in fact automated. I think there's quite a few studies as well of this being the preferred Mo for things like STDs for people where people might have like shame associated with talking,

Damienright? I don't want to talk to a person actually. I want to talk to a robot because yeah, it's a

Nikolawhole new level of anonymity Right and avoiding embarrassment. So it's I think it's a pretty powerful thing to do. So look if you can't tell if something's human or not, if it's fully unhinged and You know trying to be as human as possible Well, what is a human? Who is a human? How do we

Damienjudge that? I know a lot of robotic people, that's for sure. So For

Nikolasure. And I think, in that case, well, it's a brave new world, right? Will people form emotional connections, friendships love interests? Probably, right? And probably see that much sooner rather than later, right? And I think while OpenAI is not in the business of doing that there are People who will be in that business. Cause I think it's potentially very lucrative. And I think it's also for providing companionship, maybe for older people, for all sorts of uses that are actually really, really good.

DamienAnd this is, but this isn't, and that's not new technology, I think is always been really good at creating comfort in physically distancing people from things or other people. Yeah, that's. Since the dawn of time, a lot of significant communication, technological advances have been at the expense of physical proximity, whether it's COVID, like normalizing, working from home and having distance instead of me seeing you. In the office, I'm seeing you through a screen and that's become my norm as your 2d presentation.

DamienAnd so when I do see you in person, it's really weird, just something like this and even instances like, Spike Jones movie, which I think is coming up on being more than a decade old at this stage with her, where a person forms a really strong bond with their voice assistant. Do you anticipate that people are going to become so comfortable as we've noticed with. With this particular use case with this customer that they become so comfortable with talking to automated systems that even beyond the hardships or the emotionally sensitive, or maybe shame inducing conversations, they're going to become addicted to talking to bots in general, and that's the only way that they're going to want to interact.

NikolaWell, no, I think that's a bit extreme, right? I think that like varieties, spice of lights, people probably do both. Right. Most men form an emotional connection to their cars, right? They care about them greatly. And that's like technology that you just becomes part of your life and you love it, right? So like if you had an assistant that actually carried state, knew a lot about you, helped you and stuff, then it's a sidekick, right? So why wouldn't you like enjoy its company? And the fact it's just part of you, right? Like it's another loved, person. They're a human experience that you live with.

NikolaSo, yeah, I don't know. I think, look it's coming, it's coming fast. We just get it to you've seen like some alternate, like discussions, maybe about an opening. I tried GPT button on a phone or something like that. That's a hard like form to get it right. And I remember when I worked at Apple and I would tell people that I work on Siri. A lot of them would just start telling me how terrible it was being the debater that all

Damienright. Speaking of awkward human error actions,

Nikolahonestly, I'm like, yo, I just met you. Like,

Yanare you deliberately picking a fight?

DamienOr like, could you pass the salt, please?

NikolaWhat's going on? What's going on. And the truth is it wasn't really meant as a provocation. It was just like, right. They felt that way. And I was like, but what is it not doing that you want And that's where people go well, it doesn't work. I was like, well, what do you

Damienwant it to do? Yeah. They don't have a specific answer. Very early.

NikolaCause if you're setting an alarm, if you're putting in an event in your calendar, if you're calling someone, if you're sending a text message, sure. Something might be misspelled, but those people that enjoy using it that way, they use it and that's fine. And they're used to the fact that something could be mis inscribed and no problem at all. So, I think it was just really like the expectation. That it should be doing a lot more that was working against the perception of Siri as like a functional tool because the functional tool it is, but, ranked against the future promise of what a voice assistant should be.

NikolaThen it fell short. People wanted something that could do anything, and yeah, we're not there yet,

Damienbut we're not there yet. But even those those behemoths behind some of the first, consumer voice assistants that the brands that, Entrance that neophytes to the whole space of having interactions with a I that's their foundation model, right? It's talking to Syria. It's talking to Alexa. It's talking to Google home or whatever. They, the 17 different names that Google applied to their various assistants. So. Do you think, and you mentioned this a little bit, and I'm wondering, like, ultimately looking further down the road, legislation, like you said, plays a huge part in this.

DamienAnd I think a bit of the party pooper, right? The idea of humanizing technology too much is scary to legislators because they know the legal sort of ramifications that something like that can have down the road. And of course the nefarious usage. Do you think as a society though, we will hit a point where we decide, you know what? Maybe we shouldn't make these things very human. Maybe that's not the end goal and just collectively try and walk back that progress. Or do you think we're on a steadfast mission to make it as human like as possible?

NikolaI think there are ways and places where it's beneficial for it to be human like in places where it shouldn't be. And I think like the best product will win each of those categories, right? Legislator wanting to legislate for the sake of legislation. Russell's looking at you.

DamienThey'll do, they'll do what they'll do. Sponsor, right. Your legislation.

NikolaLook, they'll do what they'll do, right? And they'll over legislate or they won't legislate enough. I am much more comfortable with the Anglo-American approach of like, follow the common law, see the really bad uses problematic ones and then act against them versus try to preemptively solve. Now that this technology's in flux I think that in the uk. Number 10 has organized quite a few events where they solicited these opinions. And I think that their overall attitude of we don't really know how to regulate this. So we shouldn't do it just yet is one of the healthier ones.

NikolaAnd I think America's on the same page as well. I think the EU is a bit more heavy handed, which again, won't help them become, more competitive. There's still like a distant number three behind this, like us blood block or China. I think there are places, there are uses for human like things that will be great products and we shouldn't stand in the way they should not be used to, con people out of money or something like that, but think that that'll

Damienjust, that's what humans are for. Exactly.

NikolaThat's what salespeople are for. Right.

DamienBut

Nikolayeah, exactly. No, it's, um, we'll see. We'll see. It's going to be really hard. To legislating us all this because the tooling is so widely available, right? And especially with the open source community, if you look at like Mistral and stuff, I think my co founder successfully managed to get instructions on how to, like, pack a Serbian guy into two fridges, right? And it only took a bit of nudging to get instructions on how to, like, yeah. Was that, was

Damienthat a veiled threat delivered to you? Hey, I know how to I know I use, I used, I used Bistro to figure out how to pack a Serbian guy into two fridges, just You should ask,

Nikolayou should, you should ask Sean. He did it and he did it successfully. So, I guess he's one of the jail breakers. I just think that one's a bit easier to jailbreak. But, the

Damientools, that's Sean Wen CTO and Chief Hacker here at Poly AI, by the way. That's right. That's right.

NikolaBut look, I think people will jailbreak these things. Inevitably, then, as it gets more commoditized, you will have things that you'll be able to use to, have pseudo access, right? Like, when did hackers ever not want full access to the technology they're using, right? Right. And then, what they'll build on top, there will be some amazing things and some terrible things, right? And then, once that emerges, we can regulate and see what happens. I don't think that preemptively, legislating against it will do anything, because we don't really know what's likely to emerge just yet.

NikolaAnd again, like policing for it, it's going to be really difficult.

DamienThat's it. And just as we all know, the pace of legislation is, it can't rival where this is going and the quickness with which it's getting there. So that's right. Excellent. Well, I think that that's all we have time for today. Really appreciate you. You chatting with me today, Nicola, I hope you enjoy the rest of your day and we'll catch you next week. Yeah. Thanks for having me.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. What if your contact center became a product team?

    Nov 17, 202635 minAudio

    Three years ago, Simplyhealth's PolyAI deployment was only resolving simple FAQs over the phone. Today, they run voice, in-app, email, chat, and WhatsApp conversations on the platform. Nikola Mrkšić talks with Paul...

    Credit the people on this episode

  2. What happens if your AI agents get 1% better every day?

    Sep 10, 202639 minAudio

    An agent that gets better every day compounds into a very different agent within months. That's the idea behind Wren, PolyAI's new release: an agent whose job is to make your other agents better. Some of what it finds...

    Credit the people on this episode

  3. Who's coordinating your army of AI agents?

    Sep 3, 202644 minAudio

    Companies are pouring money into AI, but not everyone is getting the returns. Ian Jacobs, VP and Lead Analyst at Opus Research, thinks he knows why. In this episode, your host Nikola Mrkšić sits down with Ian to work...

    Credit the people on this episode

  4. Why should CX leaders care about MCP?

    Aug 20, 202636 minAudio

    When PolyAI built its Builder MCP, one client helped shape it: Vixxo Facility Solutions, one of the largest facilities providers in the US. In this episode, guest host Damian Sasso talks with Vixxo's CTO Derek Neighbors...

    Credit the people on this episode

  5. Can AI really hear a call the way a person does?

    Aug 6, 202622 minAudio

    PolyAI just launched Dialog-RSN-1, its first audio-native model, and your host Nikola Mrkšić sat down with its builder, Matt Henderson, to unpack why it’s a game-changer for building voice agents. Most voice AI either...

    Credit the people on this episode