In this episode of Science 4-Hire, I welcome my new friend Martyn Redstone, author of the H.A.I.R newsletter, founder of the recruitment chatbot tech consulting company pplbots and a pioneering force in the intersection of conversational AI and recruitment technology.
“Now, we’re seeing the move to generative AI-based chatbots… but it always comes down to having a design-first mentality.”
-Martyn Redstone
Martyn and I have a jolly fun and meaningful conversation about the evolution of chatbots and the role of generative AI in the recruitment process. With over two decades in technology and a laser focus on conversational AI solutions for recruitment, Martyn shares his journey through the advancement of chatbots from simplistic decision trees to complex systems empowered by natural language understanding, processing, and now, generative AI. We thoroughly explore the implications of chatbots and AI technology on candidate experience, the nuances of designing effective chatbot interactions, and the potential pitfalls and promises of leveraging large language models in recruitment.
This conversation delves into the significant shifts in recruitment technologies, the criticality of design-first approaches, and the careful balance between innovation and ethical considerations in implementing AI tools.
Insightful Moments:
- Evolution of Chatbots: Martyn illustrates the journey from basic chatbots to sophisticated systems enhanced by conversational AI and generative AI technologies, highlighting the transformative impact on the recruitment landscape.
- Design Challenges: The conversation illuminates the complexities behind designing chatbot experiences that are not only technologically advanced but also ethically sound and user-friendly.
- Generative AI in Recruitment: They discuss the advent of generative AI in recruitment, addressing both its potential to revolutionize candidate engagement and the inherent risks of relying too heavily on such models without adequate safeguards.
- Practical Applications: Martyn shares insights into real-world applications of conversational AI in recruitment, from enhancing candidate screening to re-engaging talent pools through intelligent, automated interactions.
————————————-
Full transcript.
S4H_Martyn_Redstone
Speaker 0: Welcome to Science for Hyre. With your host, doctor Charles Handler. Science for Hire provides thirty minutes of enlightenment on best practices and news from the front lines of the improvement testing universe.
Speaker 1: Hello, and welcome to the latest edition of Science four Hire. I am your host doctor Charles Handler, and I have a great guest today as I always do. But what I really love is having a guest that I don’t know I have a lot of people on that I know. And my travels in LinkedIn, you know, one of the things LinkedIn is great for is finding interesting people who are doing interesting stuff and saying, hey, I like what you’re doing. Let’s let’s talk and share that with the audience.
And The other thing I love, so it’s kind of a double punch here is people that aren’t really in the realm of of what I do, but are are the bigger picture, basically, of how people get hired. Right? And that’s for improvement. So my guess today, Martin Redstone of PeopleBot and I will let Martin introduce himself and tell us a little bit about your your background and what you do, and then we’ll have a great conversation.
Speaker 2: Yeah. Thanks, Charles, and thank you so much for inviting me on to to the podcast. Really excited to be here and have this conversation with you. For those that that don’t know me out, there. My name is Martin Redstone.
I I’ve been in the world of technology for twenty five years or so now. And I’ve been in the bottle recruitment for eighteen of those years. For the last twelve years, I’ve been specifically focused on technology solutions within the recruitment process. But for the last six and a half or so years, I have been focused quite laser focused on conversational AI. For recruitment. So that’s kind of chatbots, voice bots, automated messaging, WhatsApp, those kind of things. And over the last twelve months, obviously, with all the will be excitement around generative AI, but I’ve pivoted a little bit towards that as well. So, ultimately, as you said, PeopleBot is my consultancy, and I work with recruiters to upscale them on generative AI, but also to help them implement conversational AI solutions.
Speaker 1: Very cool. So one of the thoughts I had the other day as I was thinking about chatbots. And if you remember chatbots, even twenty nineteen, twenty eight teen as kind of a candidate experience thing. Boy, were those chatbots stupid? You know, I would go and test them out and, you know, you’re applying for this job and if you didn’t fit an exact script they had, they can’t do anything but refer you to somebody else, which I thought might be even a more frustrating candidate experience than, you know, not having a chatbot for for goodness sake. But it’s changed now because the chat chatbots are pretty smart. We still gotta teach them, but they’re pretty smart. So I would love to just kinda hear your take. Again, you know, anything we do sits within the context of hiring and and recruitment really has, you know, the tip of the spear all the way to the end of the, you know, a to z, as you would say there. So tell us a little bit about kind of how things have evolved before generative AI?
You know, what was the what was the prescription? What were you doing? And then how has that changed? You know?
Speaker 2: Yeah. It’s a it’s a great question. So I I agree with you. I think I think, you know, six, seven years ago, the world of chatbotsom recruitment was just getting started. And we saw some early examples of people using Facebook Messenger and, you know, twenty kind of fifteen, twenty sixteen was the the time when Mark Zuckerberg went kind of all big on on chatbots and and messenger bots.
So everyone was kind of jumping on that bandwagon. The challenge we’ve always had, whether it be the kind of simplistic chatbots in the past or even the chatbots nowadays, is more of a design problem than anything else because the technology has always been capable of delivering a good experience. The the challenge we’ve always had in, especially in recruitment is we don’t think about the design of those experiences. That’s one of the changes that I’ve been furiously trying to bring about for the last six or so years, but But absolutely, you know, now, you know, as we kind of move through the last six years, we’ve seen people going from kind of very simple binary decision tree based chatbots all the way through to chatbots that are powered with a conversational AI stack like natural language understanding and processing generation. Yeah. Absolutely now. We’re seeing the move to generative AI based chat box. Now I don’t think that a lot of the time that’s necessary for what we do in recruitment. I don’t think it’s necessary in general across most chatbots. I think that we’re seeing now some of the pitfalls of relying too heavily on large language models to power your chatbot when you’re not giving it again design guidelines, guardrail guidelines, securing against prompt injection attacks, those kind of things. So so
Speaker 1: yeah. Yeah.
Speaker 2: There’s certainly space out there for everything, but like I said, it always comes down to have your design first mentality.
Speaker 1: Yeah. So tell us about that. I wrote my notes, like, what does that actually? Mean it doesn’t mean what color the chatbot is or what little icon you use that’s a robot that’s supposed to be silly and friendly. Right?
What is design of a chatbot? What’s that all about?
Speaker 2: It’s a really good question. Ultimately, it comes back to the design of, yeah, absolutely the experience in terms of Is it the right branding? Do we have the right colors or those kind of things? But but absolutely, what you’re having is a conversation. So in much the same way that you would expect somebody who is phoning somebody on behalf of your organization or somebody in customer service in your organization, you have an expectation of the way that they handle that conversation.
You should have exactly the same thought process around how a chatbot handles that conversation. So we’re not just talking about do I use emojis? Do I say hi instead of hello? Those kind of things? We’re just talking about some deeper design principles to help people have a good experience.
So it starts with simple things like tell the user what they can do with you? What kind of questions can we ask you? Because otherwise, like you said, you know, you’ll go on. You’ll you’ll ask it something. It’ll just say, I don’t know.
Don’t know, try a different way, etcetera. And it just becomes very frustrating. So the chatbot needs to help the user to help it, deliver a good experience. So, ultimately, say, hey, you know, welcome to my recruitment chatbot. You can ask me about our current openings. You can ask me about what it’s like to work at the company. You can ask me questions about x, y, and z. You know? And so it sets up expectations. So so it’s just simple little design principles that that need to be implemented.
And there’s a whole plethora of design principles within within UX design that club together to be called conversation design. And that’s really, really important. So when you’re when you’re implementing a chatbot, you don’t just want to think about what dates am I trying to collect, what colors are the buttons. You need to also think about the fact that this thing is having a conversation and you need to design what conversational experience he’s like.
Speaker 1: Right. Right. So there’s more planning that goes into it. Who does that type of design? Is that, you know, part recruiter, part tech person.
I mean and even what what are you using? Are there are there, like, applications with a UI where you can just go in and set your chat up, I would imagine without having to do a lot of deep development. So tell us a little bit about on the back end how are you doing what you say designing this thing. Right?
Speaker 2: Yeah. So so, yes, you know, ultimately, you can get bot in the box applications where, you know, it’s a very simplistic no code. Kind of environment accessible chatbot. But from the design side of things, absolutely needs somebody who is a conversation designer involved. It it can’t just
Speaker 1: be such
Speaker 2: a recruiter
Speaker 1: Some
Speaker 2: of the vendors have those people. So some recruitment chat will vendors have conversation designers in house. Some don’t some people use something like myself to to come in and do that kind of design based. So so there’s lots of options there. A recruiter has to be involved because they know the kind of conversation that has to happen.
But they can be the person that actually goes and designs it. They they need to be the consultant on the project, basically. And so we we run various workshop design those conversations. We used to do something called wizardables design testing, which is where, you know, two people sitting back to back actually having a conversation with each other and recording that and going through kind of exactly what that sounds like. So there’s loads of cool things that you can do to make sure it’s a good conversation, but it’s important to have the right people involved. Yep.
Speaker 1: Yeah. Cool. So I don’t know if you if I’m just thinking out loud here. My brain works in strange ways. Do you have any chatbot bloopers to share?
Like, any kind of hilarious scenarios that happened because of the nature of these things and someone how they were trying to interact with it?
Speaker 2: Yeah. I mean, I mean, there’s there’s certainly some great examples of chatbot conversations that don’t go the way that you want them to. You know, we have we obviously have to think about protecting against profanity and those kind of things. And a lot of chatbots used to have, you know, a built in profanity filters, so somebody would curse and swear at the chatbot. It would just filter it out, you know, and and right.
Right. We we have place names in in the UK that that, you know, look like they have got profanity within them, but they don’t
Speaker 1: Mhmm. Yeah.
Speaker 2: So that’s that was quite difficult, you know, especially when you’re you’re bringing in something that may be American technology into use in the UK. We’re having you’re realizing that when you’re saying, oh, you know, what’s your address? Where do you live? Where are you looking for jobs? Those kind of conversations?
And and they were being blocked from from talking about where they live. So that was a that’s always an interesting situation. It doesn’t really happen anymore because we we we get around and not having profanity filters on, but but working in a conversational way rather than just a a a brute force filter But then we’ve seen recently in in the media lots of kind of bloopers around DPDA delivery company over here in Europe. They have they they launched a chatbot again, you know. They just went all in on generative AI, launched a large language model based chatbot, when somebody was able to go on, given instruction to the chatbot, which was, you know, swear at you know, you make make the chatbot swear at them, asked the chatbot to write a poem talking about how terrible the company actually was.
Speaker 1: Oh, but So
Speaker 2: that was the that was a Yeah.
Speaker 1: That
Speaker 2: was a terrible bloop that happened a few weeks ago. And then couple of months ago, there was an example of a Chevy dealership over in the US Yes. That’s our branch. Yeah. That’s everywhere.
Speaker 1: Yeah. Yeah.
Speaker 2: And and yeah. And and, you know, somebody was able to basically buy a car for for a dollar.
Speaker 1: Yeah. Yeah.
Speaker 2: It’s making sure you’re using the right technology and making sure you’re using it the right way. So so whilst they’re funny, bloopers, there’s a lot of learning that you can take out of them. And I think that’s the most important part is rather than holding these people up and saying, look how terrible chat bots are. Actually, think why did it go wrong? What can we do to to make sure that doesn’t have
Speaker 1: Right. Well, I mean, that’s how we learn. We learn from our mistakes and, you know, you can’t also, you know, you could try to think through every scenario that’ll happen and and you can’t until you actually put it in the wild. So I have a it’s not exactly a chatbot story, but it’s a chat story, and I wonder if this happens too on your end. So I work a lot of times, work in contact centers and, you know, build assessments and stuff.
So we were and we sit with people in those centers and watch what they do so we can understand the job and what human skills they need. So we were working with chat chat tech support agents. And, you know, we had a I was spending time interviewing them, and all of them said that you know, this is for phone, you know, service tech support, cellular service tech support. But they have people who just come on because they’re lonely. And they wanna talk to somebody. So they’ll just go in the chat and strike up a conversation. They have people that come on there and try to get them to go on a date, whatever. But I thought that was really interesting. Oh, wow. People are and and now that you have the the GPT store, you know, you see about others twenty five thousand GPT girlfriends people are creating.
Right? So Maybe you don’t need to call your cellular provider anymore to to have conversation. But anyway, I I don’t know how much you know, that happens or if you even know. Like, do you collect dialogues, transcripts of all the different chats and analyze that, see trends and things? Is that part of what goes
Speaker 2: on? Yeah. Absolutely. And I and actually, I’ll talk about kinda two different things around around what you were just talking about. But, yes, absolutely one of the things that we always need to do is review the performance of the chatbot.
Think and this is Mike. He’s saying to me. You have to think of a chatbot like they are a member of staff. You know, you can’t just Right. Go and get on with it.
You know, you need to put periodically review how their performance is. And so, yeah, And so when when the chatbot, that is looking at transcripts, finding trends, you know, why aren’t people completing the conversation? And where are they getting from straight at where they’re asking for a human, all those kind of things where we can we we can see the negative aspect. And and the lovely thing about a chatbot is you can retrain it in how to deal with that situation. So if it’s not dealt with the situation very well or it’s handle it’s trying to handle something it’s not even trained for, you can then train it for that situation or retrain it.
And and it will remember it forever. But unlike a human, you won’t forget it next week. So it’s so so so yeah. Absolutely. That’s a big part of of Manu and Chatbot.
I I stole a phrase off of a friend in the industry a while ago, but when I implement chatbots into businesses, we always make sure that there’s at least one bot master in that organization who is doing that role as part of their management of the chatbot is reviewing the performance of the chatbot. Gotcha. But actually, the the the going back to the the points about chatbot girlfriends
and chatbot boyfriends and and what have you. It’s there are chatbots out there that are designed specifically for that as you’ve seen within GPT store, but also kind of commercially there are lots of apps out there for that. We do hear stories about people trying to be inappropriate with chatbots to see what what how it will react.
You also hear interesting stories. So so there’s maybe not a chatbot per se, but there used to be based being on an email based scheduling tool. So it would use natural language processing to have a back and forth conversation via email. And it would take all the persona of, you know, a personal assistant for the person who’s trying to organize the the, you know, the the meeting. And so it was that impressive, the technology, the people used to actually send flowers to the offices.
Speaker 1: No way.
Speaker 2: So you basically was a person they were sending it for a chatbot. Yeah. Absolutely. So there’s some some really famous stories around that. But, actually, very interestingly, there’s a one of the biggest use cases for conversational technology now is for companionship.
Speaker 1: Yeah.
Speaker 2: And so we know that I think there was a a a survey that recently went fourteen percent of respondents. So they absolutely expect to have some kind of companion based relationship with with a conversational interface in the
Speaker 1: future. Interesting.
Speaker 2: But there are businesses out there like intuition robotics and Ferhat Robotics, Ferhat is Scandinavia and Intuition Robotics, an Israeli company who are providing companionship robots which are conversational waste to old people who are suffering from loneliness. So, you know, intuition robotics. It’s really cool. And Intrusion Robotics, great company. You know, they built fantastic bots.
Is it where it’s kind of a like, an iPad style device that can move around right And they put it into, you know, old people’s homes, especially in places like New York and what have you a lot of old people get given them by the state.
Speaker 1: Yeah.
Speaker 2: And they not only do they have conversations from a companionship perspective, but they’ll remind them to take their medication. They’ll they’ll also based on their multi modal display. They’ll be able to take them through things like stretching exercises, yoga exercises to keep the mobility out of this. Yeah, there’s absolutely this this fantastic companionship based use case, but it but it has to be done appropriately and and and and sensibly rather than just somebody trying to chat up a chatbot on a on a on a cellular phone helpline.
Speaker 1: Right. Well, we know. Yeah. We know that some point that’s gonna make the jump to some physical vessels that aren’t just like a a room by with a arm on it, you know. They’re gonna be even synthetic, you know, flesh or whatever.
I think that across the board of any kind of companionship, even inappropriate kinds, I think you’re gonna you’re gonna see a pretty big industry there. And, you know, I I mean, you see robots I read in my news feed, Elon Musk is making an army of you know, robots, there’s other companies. I I will we’ll see if there’s so much hype going on. But it’s inevitable, right, that it happens and you know, jeez, it’s it you start playing it forward in your mind and it gets really crazy to to the point like, whoa, look at Blade Runner or something like that. Like, we could end up with rep all that stuff.
So I I try to at least focus down a little bit. So So right now, you know, you’re advising a company. They they wanna use a recruitment chatbot. They’ve never tried that. Maybe they do wanna use generative AI because I think also it’s an interesting point you made earlier. I think in my travels, I see people just want to throw AI at anything and not realizing that it’s it’s not a solution by itself. It it’s find your problem first and figure out if it’s a good first encryption. There’s what I call AI FOMO. Right? People are just like, ah, we can’t not use this because everybody is and it’s taking over the world.
So and it really does require a good modicum of caution. You know, I I think it’s the the the big corporate priced people I talked to about how they’re using AI. Everybody’s still kind of in the at least generative is still in the okay. Well, let’s kinda wait and c, let’s tinker with it, which is a good thing, I think. It just evolved so fast that you could be tinkering forever. Tell us you know, you’re you’re consultant to I’m sure a lot of folks and they want to leverage this thing for recruitment. What’s your process you go through to kinda determine the use case and a good use case, and then, you know, how you help set that up? I’m very curious.
Speaker 2: Yeah. It’s a great question. There’s a lot of kind of standardization. Across the board. But actually every time I go into a new recruitment team, recruiting organization, you have to think that they have their own separate challenges.
And so that’s the bit where we start at is what are we trying to solve for because that’s the bit that drives the value. And like you said, there’s no point in just having a shiny object for the sake of having a shiny object. You need to think about where the value is going to be.
Speaker 1: Alright. So if you all out there are watching the video and you notice that Martin and I are wearing different outfits this time around. I’m not ashamed to relay that, you know, we had continuity issue. I I worked in TV production for a while, and continuity is is really a pain when
you gotta check every single little thing, but we had a the first ever in four years of doing this technical crash. So I don’t think it was Martin’s fault.
It happened right when the train whistle blew. So we’re gonna pick back up where we left off. And hopefully, you know, well, I know for sure that stimulating an awesome conversation will continue. We’re just wearing different clothes, but we’re both ready to go. My app it was in the wash.
I didn’t really check what I what I was wearing, but that’s okay. So I asked you the question, which is a really important question, and I’m super excited to hear about it because I don’t have any experience really with what you do. But when you when you go into a client, to help them out. Obviously, they’re contacting you for different reasons. But but walk me through your
process of, you know, how you get from from the initial meeting to actually producing something for them.
Speaker 2: Yeah. Absolutely. And, yes, and and apologies for the continuity think it’s warmer today. I think I was I might have been wearing a hoodie last time or something. So it’s definitely a bit warmer today, which is great because it was winter was getting to me a little bit. Yeah. So so so it’s a it’s a great question. What happens when we come into a client? Well, you know, thing is is that there’s no there’s no one standard reason that that a client calls me, and it could be they want to explore the possibilities. It could be that they’re trying to fix something. Ultimately, where we start is by evaluating the current situation. So so once we’re happy to move forward with the project is in, you know, we’re we’re mutually happy to move forward with the project. The first thing that I do with all of my clients is run a workshop and I run a something I’ve been building up over time. I call it a a Lightning Innovation workshop because some of these things can take days upon days upon days. I try and get it done in one day. Usually two, three hour sessions, one in the morning, one in the afternoon with the senior leadership or anybody that’s impacted by what we’re trying to do. And so the first session is let’s find out just what is actually wrong, you know, what you’re trying to fix. Find out, you know, where the issues are, where the challenges are, where the bottlenecks are, those kind of things in the process. And then the afternoon is solution nice for them. Come up with some ideas. Come up with some experiments because we always like a bit of an experiment, you know, and and a bit of a proof of concept situation, then we we plan towards the the solution for that in the afternoon as well. So once we’re happy that we’ve got here’s the problem we’re trying to solve and here’s the potential solution for it. It’s then about building the the plan around what that solution looks like, how it’s gonna be implemented, so what would we need, those kind So then we go into the real standard project management piece. The interesting thing about running different types of projects that are kind of in that sphere of AI is that there’s lots of different ways that people interact with technology. As you know, one of my kind of specialist areas is is conversational AI.
So running a a conversational project is a little bit different to running a a new website project or an app project or something like that. So although though it should still actually be treated as the same kind of complexity. I think that’s one of the biggest challenges that I tend to come in on is people tend to think that something like a chatbot is a simpler project than a new website, let’s say, Right. So so it’s about grounding them and making sure they understand the complexities around that project as well. That’s super important to to Manjek expectations from the
Speaker 1: beginning. Yeah. Yeah. Of course. So, you know, what are some of the more interesting and definitely, let’s stay in the in the realm of the chat bought and conversational stuff just because that’s a unique specialty that I don’t think a lot of people know about. So I’m excited to hear. But what are what are some of the more common or maybe just pick one. That’s one of your favorite projects, implementations, you know, what what did you accomplish kind of thing?
Speaker 2: Yeah. Could it’s made as being so many. There seems to be quite a standard type of use case that people want to implement chat box for within the recruitment space And ultimately, that is when you’ve got a high volume of applicants and you want to, which will that down to a long list ultimately. So so that seems to be a very, very common use case. Again, there’s lots of dependencies around, you know, their current tech stack.
Whether as an organization they’re using, you know, cloud services from Amazon, Google, might have you because a lot of the times especially over here in Europe, there’s a lot of data security process that we yeah. Yeah. For we can even think about kicking off. So, like, stick with what they’ve already got, you know, and and so kinda make it easier to get to check-in there. But yeah. I mean, there there’s there’s lots of examples of that, you know. And ultimately, again, we can do some really simple stuff around Is it a very binary role? You know, do we just need yes, nos? You know, is it is it manufacturing? Are you able to stand up in a factory for twelve hours? Can you lift twenty five kilos? You know, those three things? Or is a little bit deeper than that, you know. And so we’ve implemented solutions that will score the candidate candidate based on their linguistic response, their text response, asking them for situational questions, tell me about a time when you’ve delivered exceptional customer service. Knowing at how they respond, analyzing that, being able to pull out things like big five personality traits from those kind of answers.
I’m really just scoring it around this. There’s some really cool things we can do. Really cool things. And from there, you can, you know, you you can you can take that away you can score it against the requirements. For the world, you can also score it against the other candidates as well and give people relative long lists based not only on how they match to to the role, but also how they compare against the other applicants as well.
So that’s kind of the main use case. One of my favorite use cases is always regenerating app on an ACS. And I think that’s really important that people don’t don’t think about that. You know, you spent years as a recruiter and as a business. Spending so much money on recruitment advertising.
And so you fill up your recruitment system with so many different people that are interested in working for you as an organization, and you review that process over and over and over again. So one of the great use cases that we go into businesses with is actually, let’s let’s use messaging and conversational AI to go back out to people over what’s over SMS, what have you, and actually answer them whether they’re still interested in working there, update their profiles, get a get a really good picture of what that applicant pool looks like within your system. And then fish from that pool rather than spending more money going back out and advertising on job boards and what have you. Actually, start start finding people that you’ve already got in your database that you’re probably bringing back in time and time again.
Speaker 1: Yeah. That’s really smart. And, you know, we’ve before these kind of tools, you know, have had a lot of exposure to people talking about, how do we do that? You know, maybe that’s through email or whatever. But it’s definitely more immediate, right, to to be able to to message people out.
Well, you I should give away some kind of prize or something. I don’t know because you totally read my mind. I mean, I have my little list here that I and my next question was, hey, what about
screening. Right? Because I’ve again, my interaction with chatbots just to just to fool around. It’s been a while and you know, I tried some screening ones and they’re pretty crude. Right? And but what you were mentioning is is pretty you know, that’s the realm of of my training and expertise is, you know, how do we how do we shift and sort people based on really good, reliable, accurate, data collection relative to what the job requires. And, you know, I was a little surprised to hear you talk about some of those more in-depth methods. And I think that here in the US, we have a lot of regulations.
And I think that you know, if you’re not an applicant yet, it’s a little bit of a gray area. Right? If you’re not an applicant yet, those rules don’t apply to you. But if you are carrying that score over across when someone’s an applicant, then it can become an issue just because there’s a little bit of, you know, inconsistency, maybe some people came through a different channel. I don’t know, but it’s it becomes more in the in the crosshairs of what regulation is once you’re a candidate. So I’m assuming maybe it’s a wrong assumption that, you know, most interaction with the chatbot is gonna be by someone who’s not an applicant yet in order to then be engaged and and to move forward in the process to con to compel them if they are indeed a good fit. So tell me a little bit more about, you know, how how you go about Well, who’s taking those things? At what stage? And how do you go about actually, you know, constructing those kind of tools?
Speaker 2: Yeah. So so so you, you know, you’ve got you’ve you’ve raised some fair points. So the first kind of top of the funnel use case for a chatbot is definitely the search and apply piece. So the search could be let’s find a job, but it could also be let’s answer all the questions you have about this job as well. And and, ultimately, you know, people wanna you know, is it, you know, nowadays especially, is it remote?
Is it in the office? Is it higher breed?
Speaker 1: Right?
Speaker 2: Is it in the office? Can I bring the dog in? You know, do you pay for travel? You know, all those Yeah. And so there’s the opportunity to give people a really good opportunity to make a a sensible decision as to whether or not they want to apply for the job. Based on being able to ask these questions. So you’re actually helping people screen themselves in or out Right. Right to the top there.
Speaker 1: Perfect.
Speaker 2: And and that’s super important because nowadays, you know, in the world of one click applies and what have you. It’s it it becomes quite a burden, not only on the recruiter to deal with that boy, but also on the candidate to apply for tons of roles, but also to keep on top of all the roles they apply for. So help them screen themselves in or out based on their knowledge of the role. Once they’ve applied and they can do that either on fashion way by form or they can do that conversation, you know, in chatbot. But once they’ve applied, that’s the the biggest thing that that we have to try and fix in recruitment is that that blank space
Speaker 1: Yeah.
Speaker 2: Between applying and then hearing back, if ever, at all, from the organization. The black hole, the best way you can do The the black hole at slightly. This this ghosting area that people can’t really get frustrated about. So the best way possible is to respond to them you know, and ask them to take part in a screening conversation. And that screening conversation is a bit deeper.
It could be, are you in the right location you’re looking for the right kind of salary, those kind of really top level, you know, kind of screening questions. But like I said, you know, you can go really deeper into it and start asking people to go through situational questions. And and, yeah, and there there’s definitely ways that we can build that. So, you know, the situational questions like I said, you know, tell me about a time when you’ve accept when you’ve delivered fantastic customer experience here. Tell me about a time when you deal with, you know, a team or, you know, whatever.
Yeah. And so what we can do from there is we can infer personality traits, start scoring those stories based
Speaker 1: via NLP? Is that is using NLP? Right? Yeah.
Speaker 2: Absolutely. Yeah. Using natural language processing, to strip it apart and decipher and and the context of the answer. There’s various different technical ways of doing that. Whether you build it from scratch or whether you use a back end.
You know, there’s there’s a some great platforms that this one called Gigi, which which has done a lot of research published research on inferring personality through
Speaker 1: To do that again, I’m gonna write that down.
Speaker 2: Gigi, j u j I.
Speaker 1: Okay.
Speaker 2: That’s a fantastic chapel platform. But there’s also other chatbot platforms that are built specifically for hiring. So it’s SAPIA, SAPIA. They’re an Australian business.
Speaker 1: Yeah. I know them. I know them.
Speaker 2: Yeah. Not not only do they, you know, give the up candidate the opportunity to go through this and score them, but they also deliver a report back to the candidate. Yeah.
Speaker 1: I’ll tell
Speaker 2: them more about themselves and what have you. And it’s fantastic. You know, and it’s those kind of experiences that are delightful for a candidate because not only are they are they not being ignored? They’re being engaged by a representative of the organization whether that’s a digital representative or not. They’re being engaged by representative of the organization. They’re they’re getting in touch with them. They’re having a conversation. They’re they’re understanding that this is the first part of the assessment towards whether or not you’re a right fit
for the role. And so it’s super important. Super important to make sure that there’s some kind of engagement there because people don’t get responded to.
Speaker 1: So is that automated? So, you know, you talk about, okay, they’re gonna have a conversation. Would it happen to me that I’m Let’s see. I go on to employer site. I I interact with the chatbot.
I determine that there is or we together determine that there is a role that I would like to apply for that I’m probably suitable for. Am I instantly taken into the assessment type stuff you’re talking about? Or is there a wait while a recruiter kind of reviews it and then sends me some kind of a message to say, hey, this is the next step.
Speaker 2: Well, ultimately, you can do that for everyone that applies because you won’t have time as a recruiter to take a quick look at the applications and invite them to a conversation. And, you know, ultimately, a chatbot can have a conversation with tens of hundreds of thousands of people at the same time. And and the cost of doing that is is pretty negligible. Yeah. So so so I would just suggest do it for everybody because you never know when you might find a bit of a gold nugget in your applicants.
Speaker 1: Yeah. Yeah.
Speaker 2: And and you weed them out that way. So so I would suggest doing that for everyone. Look, you know, the the the the the Panjea of conversational interaction would be, you know, if you find the best candidate through the automated chat, be able to hand them over to a live chat with the recruiter immediately. And that takes us into the realm of kind of conversational marketing and those kind of things. But I don’t thinkcommercial producers would be interested in that because they don’t like to be interrupted in their flow and those kind of things. But that would be the fantastic thing. Oh, look, I found the best candidate ever for this role. I’m gonna hand you over to the recruiter. The recruiter can take you forward from there. That would be incredible.
But most recruiters aren’t interested in in that kind of experience. I don’t understand why.
Speaker 1: Well, you know, it’s interesting too. What you’re kind of on the tip of the spear with recruitment marketing too. Right? Because you’re you’re such an experiential component and informational component. Probably more so than a lot of people in, you know, well, I guess it depends on the application, but, you know, for me, I work a little bit further down the funnel a lot where you’re already now good experience of representing your brand critical throughout the whole process.
Right? So recruitment is sales and marketing, really, with a lot of good analytics and, you know, ability to to make some inferences, hopefully, good accurate intro. Inferences. So do you do you work in the US at all? Do you have clients in the US or just mostly in the EU?
Speaker 2: I have done in the post. Have done in the past. It doesn’t it’s not my main with my main kind of clientele or Right. Tend to be UK and you’re right. Right.
But I’ve done work across the US on that case as well.
Speaker 1: Very cool. So I’m I’m starting a project right now. I’m just doing the research, but I had an idea, which I won’t give away here. But starting a project that involves J and AI, you know, training J and AI to do something. And on the front end, I wanna have a chatbot. Right? So that someone can interact with it and say, I wanna do this, and it comes back and says, okay. What about this? This and this? So how you know, I know nothing about this yet and you’re part of my research here.
What do you do to create? So I’m assuming there’s there’s a platform or application that you access to stitch the two pieces together.
Speaker 2: It depends what you want to do ultimately. But, yeah, I mean, the the whole point of an application like that would be using the chatbot to create the data that you push into the large language model to get some kind of output back out. Right. I would have thought. So so you can do that with with a lot of different chatbot platforms.
It just depends whether you want that chatbot platform being powered by, something that involves a large language model or something that involves natural language understanding or or or just something that’s a a very straightforward decision tree. Right. Right. Most of these platforms run on APIs, and therefore, you can push data in and out as of when you want to. The one the one thing that I would I would always say to people as they’re going through a a project that involves both chatbots and large language models is we’ve seen a lot of use cases now where people are trying to run their chatbots using Gen AI, and it doesn’t work because haven’t thought about the hallucinations, the prompt injections, all those kind of things.
So I will always suggest not relying on large language models and generative AI to to power your chat I would absolutely be relying on those kind of technologies to fill in the gaps. So let’s say, yeah, for instance, when you’re building a chatbot, if you’re using kind of, you know, the standard conversational AI stack, so natural language understanding, That’s language processing, that’s language generation of machine learning. Let’s say I come onto the chatbot and say hello. In the old days, you’d need to train the chatbot to understand all the different ways that somebody can say hello and then explain to me how you wanted it to respond back to that. Where we can use natural language where we can use large language models in that process is rather than spending our time as chatbot developers training the system to understand that somebody say hello is pushing that that utterance into large language model and saying, tell me what this person’s trying to say.
And matching it to one of our intents that we’ve built in. So an intent could be this person saying hello respond this way. We’re saying, okay, large language model, tell me what intent this person’s doing because large language models are very good at capturing data and creating structure out of it. Yeah.
Speaker 1: They
Speaker 2: are. So creating structure out on structured data. So the large language model will say, I think based on x y z, this person is is trying to say hello. Enter them into the the hello intent, and then the the conversation that it takes back over and responds back with with the right
response. That’s kind of where we tend to try and use large language models because it’s the safest part and the most powerful part.
Speaker 1: Yeah. Gotcha. So my situation is a very narrow use case and I plan on again, I’m just researching it, but I plan on taking an open source one and using, like, a retrieval augmented generation type thing. So it’s it’s really just looking at specific new information I’ve given it to try and put some boundaries, you know, around it. And The use case is so narrow when someone comes on, you know, they have a purpose.
And this is just to help guide them through a purpose. So but that’s really interesting. I’m pretty excited actually about, you know, the project. But, boy, as I started to understand, how do you train a large language model It’s a lot of work if you really want a structured thing. Right? I mean, you could feed it all kind of junk to help expand its horizons. But you also have to index, I guess, and, you know, template everything. So so it knows that once it gets that in place, right, it knows what it’s supposed to do and you can pile more stuff in. So so hopefully that works more on the front end. But it’s an interesting project.
Speaker 2: Yeah. No. And I would certainly say that you yeah. You’re right. You know, they’re they’re if you wanted to use something powered by a large language you know, there are two different ways that you can control it.
You can either fine tune a large language model, which is an exceptional amount of effort. And and, you know, and and that that sometimes isn’t necessary. And you can use one of the smaller or open source large language models like, you know, large or missed or what have you and there’s loads more continuity coming out. Or you can use, like you just said, retrieval augmentation generation, rag as people kinda see it written down nowadays. And and there are fantastic platforms that allow you to to do that.
And and one of the best ones out there is VoiceFlow. Funny enough you can actually see on the cap behind me there is the logo voice flow. They’re a good friend of mine. So I’m not I’m I’m I’m not I’m I’m not employed by them or get any kind of commission form. But they’re one of my favorite platforms for building rag powered, large language model chatbots. They’re they’re proven the best out there for building what we call AI agents nowadays rather than just chat bots. So I would certainly take a look at them for that kind of for that kind of use case.
Speaker 1: So I also had I wanted to ask you this,
Speaker 2: become a free consultancy guru.
Speaker 1: Oh my goodness. Send me a bill.
Speaker 2: I’ll tell
Speaker 1: you to take it. Well, no. This isn’t a consultant. So this is more of theoretical question. I’m more re I don’t know if it’s a rhetorical question, but I thought it would be interesting. Like, do you think there’ll ever be a time when generative AI can just create a chatbot for you
where you just say, alright, I need a chat that’s gonna do these things, go out there and build it for me. Right? I mean, is that is that something that’s you know, we’re gonna get agents or are the new thing. Right?
Speaker 2: So entirely possible. Yeah. Yeah. It’s it’s entirely possible. I mean, you know, the thing that we have to bear in mind is the again, it’s it’s it’s allowing an AI agent built with a large language model to make decisions on your behalf.
Right. So I would scary it’s very, very scary thought process. Yeah. It’s not not outside the realms of reality for YouTube has given the instruction to a chatbot. To build a chatbot, it asks you all of the questions around what that chatbot needs to be.
There’s most likely technology out there that’s already doing this. But I I I just I feel slightly uncomfortable with it. But, yeah, there’s there’s absolutely, you know, and then that that’s sets off a a series of agents that that do different parts of of the build process. And then, you know, Bob’s your uncle, you got yourself a chatbot. I I think that’s totally doable.
The the challenge with generative AI like I’ve always said is the fact that it doesn’t know zero. So so it will never say, I don’t know that or I don’t understand that. It it if it doesn’t know, it it will make that up. And so that’s the problem you’ve always got is that if you’re going to build something that builds a chatbot, you need to make sure you’re asking every single question you need to know. In order to build a chatbot.
I mean, most people try and find shortcuts and what have your easy generation AI. That’s fine. But, yeah, the technologies there could actually be done. Yeah.
Speaker 1: Hey, look, we’re living in a world of deep fakes. And, you know, why not happen a chatbot deep fake you with fake information, I guess.
Speaker 2: Have a chatbot deep things to you?
Speaker 1: Exactly. Well, I mean, our agents are gonna be talking to other people’s agents soon enough. Right? I also feel like, you know, I’m interested in your take because I know that we’re getting closer and closer to being able to generate, you know, photorealistic avatars. There’s companies now that have, you know, avatar based chatbots.
They’re still a little creepy looking, but that’s gonna get better and better. So do you anticipate the the little chat, you know, the little chat bubble symbol eventually being replaced with a a human and a snappy uniform for you know, the company you’re applying to. Like, what’s the take on the personification and movement from a little bubble to an actual person for this?
Speaker 2: You know, it’s a really, really interesting subject, and there’s so much research out there around whether you even need to genderize a chat but let alone let alone create a human based habitat. Ultimately, you know, we’re we’re exceptionally intelligent being you know, people don’t realize just how intelligent we are. You know, and and we have a part of our brain that recognizes a face and and recognizes whether that face is real or not. So and that’s why whenever we see, you know, a digitally created avatar, we immediately know that it’s not real. And so we’ve got the technology’s got a long way to go before it’s it it can fake it can it can trick us into thinking it real.
Yeah. We have things like the uncanny eye where things look like a person and what have you, but when it comes to actually making people think that they are dealing with a real person, it’s a very long way off. However, you know, there there’s there’s there’s certainly research out there that says that, you know, if you present a human based interaction with a with a human looking avatar, then you get better engagement. I think that certainly so so you say you have to remember the different types of interaction we’re talking about here. We’ve got chatter interaction, which is literally just, you know, chatting as you would chat to any person on a on a piece of software. You’ve got voice which is the stuff that we’re used to when we deal with our Alexa’s and series of the world and what have you. And then we’ve got multi model. A multimodal is a mixture of all of those put together, but it’s also visual as well. And that’s where the avatar comes into it. And would you like multimodal when we when we develop things like deal with kind of our our our Alexa views or whatever they called and on on Nest hubs where we’ve got, you know, show me a recipe, and it will show you the recipe as well as talk you through it.
Right? So that’s multimodal as well. But multi modal conversation interaction with an avatar. It’s absolutely there. We’ve got lots of companies doing it, Cynthia, and lots of other businesses out there.
And and it’s great. It’s absolutely brilliant. But there’s no putting trying to do people into thinking they’re dealing with the real person. So do I see an avatar happening on the website. There’s some great companies out there that build three d holograms of actual people to me. Two number and m double e. You know, so they they do that from a kind of marketing perspective and an influencer perspective. But again, it’s it’s it’s a bit of fun. It doesn’t really make a huge difference on the engagement side of things. Not for the use cases that I work with anyway.
Speaker 1: Yeah. I feel like it could actually cause more problems if you know, oh, the the gender or the ethnicity of the chatbots different than me? You know, what does that mean? Or, oh, it is creepy looking. You know, which is a turn off for me.
Do you have clients though that kinda insist? Oh, like, we really wanna try you know, having some kind of a a persona attached to this thing. What do you what do you typically advise them? You know, sometimes clients have something stuck in their head and they they think it’s what they have to do and sometimes it’s hard to convince some other otherwise, but I’m just curious. Oh,
Speaker 2: you know, part parts of what what we do is is and and when you build a chatbot properly, you should absolutely give that chatbot a persona. But like I said, you don’t need to generalize it. You can do. Lots of companies do. They say that, you know, the persona of this chatbot is very similar to our our our consumer persona that we that that we focus on. It’s a twenty seven year old female from, you know, California and who causes a little bit, but uses emojis. You know, those kind of things that, you know, will will give give give, you know, stick back when they get to all those kind of things. But you know, it’s it’s very important to build that potato, but get it right. And and you can do that a bit easier when it comes to chat because you don’t need to think about like you said about. Genderizing about ethnicity, all those kind of
things.
So for instance, I always talk about MGM Resort, so
Speaker 1: one of my one of my clients, text clients, I mean Yeah.
Speaker 2: One of your clients. They’ve got a chatbot on their career site. And I would I would take advantage of of of pulling a bit of personality into that chat box. There’s no personality in there. And so I would say I would I would I would go back to basics and I take what’s MGM famous for the line at the beginning of the films.
You know, let’s let’s let’s put that line in as the persona because it doesn’t have to be the persona that you’re talking to because when it comes to recruitment, you could be talking to so many different people, but it could be the brand persona. Leaving the lion, whatever the lion’s name was called. I know it was a, you know, well known kind of story. Exactly. Exactly. You know, it’s it’s those kind of things that people need to think about as well. It’s that it doesn’t have to be the persona of who you’re trying to sell to or who you’re trying to attract. It can also be a brand persona. And yeah. So so really important part of of a chatbot project is building what
Speaker 1: that said. That’s very cool. Very cool. Or we’re coming up on kinda wind down time. I think if you had to leave our audience with one thing to think about, you know, let’s say, hey, we really do wanna create a warmer fuzzier piece of our process and, you know, let’s let’s think about deploying some of this stuff.
And what’s your advice? What would you wanna leave? Our listeners with around this kind of realm that we’ve been talking about. I
Speaker 2: I I think I’ll I’ll repeat what I said earlier. You know, it don’t don’t think of it as a quick and easy project. You know, I think a lot of people have fallen into this trap, especially with chat bots that are out there that are just kind of wrappers for chat EBT. You need to do it properly. If you want to ensure that you’re creating a a responsible but fantastic experience to your candidate, it needs to be done properly.
And so that’s what I would leave people with. I I think that it’s a great idea and it’s a really fun project to do to make sure it’s done.
Speaker 1: So don’t don’t buy a chatbot from a dodgy guy in a dark alley who’s who’s opening up his raincoat and he’s got a hole. Bunch of chatbots that you can choose from in there, you know. Black Market, chatbots. Chatbots. You never know what they’re gonna say. I think it was was it a giant I think it was Microsoft that had they created a chatbot that started, you know, spewing hate full vitreous people, you know, like, I mean, you have to be careful. Maybe that was a black market.
Speaker 2: YouTube. Yeah. YouTube. And and this has happened plenty of times because what tends to happen, especially in the early days of language models back back in, like, twenty twenty seventeen, twenty eighteen, twenty nineteen when we started kind of working with the earlier kind of iterations of large language models, we had chatbots like Microsoft TAE and Facebook, Facebook at the Times Blender that were trained on public conversations such as
conversation on Twitter or Reddit and what have you. And and unfortunately, very, very quickly, you can go into a bit of assess part of that those platforms and and come across a lot of problematic language that’s racist or sexist or just completely abusive.
And that’s what happened with Microsoft Tay is that it very quickly when when Uber right wing and starts becoming very, very racist and almost kind of, you know, glorifying nazism. And and the same thing has happened with with Blender bot as well. You know, the the first iteration of it, you know, because they were training it on unsupervised data. So that’s why nowadays, we see the the human reinforced training that happens with large language models where then do we train it on open data? And and just to be clear, the OGPT four from OpenAI and other kind of large language models that we’ve in the last twelve months, they’ll probably be the last iteration that’s trained on public data with run out of the enough data to train the next size models. So the next size models will be trained using synthetic data that’s been created by large language models. So so that’s the scary part. So that’s why that’s why a lot of these companies now put in a human in the loop in the training. So not only do we take the data, you know, hundred and seventy five billion parameters for GPU before from the public from from the public corpus is Corpie, whatever the the Latin is for for plural corpus. And and and start processing that data, start creating conversations, then asking humans to mark those, you know, to to score those and to fix them as well.
So so there’s there’s been a lot of bad press you know, of kind of, you know, people in Africa that are being used in in a large volume to do this, and and they’re having terrible problems with mental health and up with mental scarring just from the amount of abusive stuff that they’ve seen online. Yeah. And so it’s a real shame, but that’s the only way that you can you can do this, you know, and and and it’s up to the organizations to to make sure that if they are using humans in a loop, to make sure that they’re supported and make sure they’re they’re cared for because that’s the most important.
Speaker 1: Yeah. You’re not the first guest that I’ve had you know, my my friend, Karen, she she’s all about the model training. She’s an expert in that, and she was saying the same thing. Like, what a what a humanitarian problem it is. And it’s it’s bad because society is benefiting so much from these things.
It really stinks to think that, you know, part of it is on the back of non beneficial behaviors, you know. And it’s just it’s a it’s a real note about just how fast, all this stuff is moving, I think, and just how much demand there is and that demand I know personally I’ve never felt more overwhelmed. With information. Right? I mean, I have attention problems anyway, but there’s, you know, there’s so much interesting stuff.
It’s it’s hard to to keep up. So what a fascinating conversation, you know, Barton Ridge Stone, PeopleBox is your company. Right? So you got a minute for any kind of shameless plug where people can find you or, you know, of course, LinkedIn. I say it every time. I should say, you can find them on LinkedIn. I shouldn’t even ask you the question. But you might have some other some other things you’d wanna share. So this is your moment.
Speaker 2: LinkedIn is always the best way to find me. I’m the only Martin Red stating in the world, marking with a y as it says here. So LinkedIn is always best to basically find me through
LinkedIn. You’ll also find my newsletter. My newsletter is called hair.
Explain words. It’s AI in HR, but also, in fact, they don’t have
Speaker 1: a Yeah.
Speaker 2: API one part of it. So yeah. So so so find me a link to him. Find my newsletter there. If you do wanna go to my website, which is not as exciting as my LinkedIn profile, then you can do.
It’s people bots, but it’s it’s people it’s the it’s cool cool kid language. It’s people, p p l. So it’s p p l box dot com, and you can find me there as Robert. But But feel free to find me and connect with me on LinkedIn. That’s always the easiest place.
Speaker 1: I found you through that hair newsletter, which I like. It’s one of the many many pieces of content that I’m consuming all the time. So that’s great. Thank you so much for your time, both times.
Speaker 2: We got there in the end. I’m glad we got there in the end.
Speaker 1: Yeah. For show sharing another piece of your wardrobe.
Speaker 2: It’s honestly not a lot of changes, to be honest with you. Yeah. Seventeenth. Speaker 1: That makes
Speaker 2: the difference, and they work. Absolutely. But, no, no. It’s been a lousy pleasure. Thanks for inviting me on Charles.
Really, really enjoyed the conversation. Some fantastic questions. Thank you so much for giving the opportunity to answer them.
Speaker 1: As we wind down today’s episode dear listeners, I want to remind you to check out our website rockethire dot com and learn more about our latest line of business, which is auditing and advising on AI based hiring tools and talent assessment tools. Take a look at the site. There’s a really awesome some FAQs document around New York City local law one forty four that should answer all your questions about that complex and untested piece of legislation. And guess what? There’s gonna be more to come.
So check us out. We’re here to hell