BoneChat #56 - AI-tool for quickly evaluating bulk candidates for a new role using voice samples.

All right, welcome to BoneChat discussion number 56 with, uh, Adam Hocek at Echo, and he's going to blow our minds with a new tool, a new AI tool for assessing talent, and it probably has other applications, but I'm going to hand it over to Adam. It's all yours.

Thank you. Thanks. Thank you for having me, Tiger and Mark. Thank you for the introduction to this group. I appreciate being here.

It's a little hard to hear.

Oh, let me— oh, good. I'm glad you told me. Let me adjust my volume. Is that better now?

Better.

Okay, perfect. Good. Yeah, so ECHO is a little over 3 years old, just a little introduction. I'm the founder and CEO of Echo. We're a deep tech company. We do voice analysis and we do both emotions, psychometrics, and even behavioral patterns on that. And so what I'll just show, I'll have a few slides and I'm going to go right into a demo to show how we work.

So And tell us along the way, Adam, kind of what the impetus for the idea and how did you start it? Because that's always fascinating.

Everybody. Yeah, sure.

Yeah.

So we've— I had a background in speech, been working in speech, did research in it, developed a lot in user interfaces with speech, early days and later. And then I later got into, into machine learning, natural language understanding. And that's, that's what kind of brought me into this space. And what I realized was, hey, that it was very interesting that a lot of the emotion detection that was out there coming from universities, well-known universities and open source, was really very limited. If you use the same dataset, it worked great. But as you, if you varied the dataset, the performance dropped tremendously. So I realized there was a lot of opportunity there that things were, you know, that there was an opportunity. So that's kind of how I got into it. Now, as Echo as a company, we evolved over these 3 years. We first did, and that's what I'll show today. We first did a lot with with understanding speech and then generating psychometrics. But what we did initially was we wanted to map it into popular— like earlier talking about Myers-Briggs, DiSC, Campbell, Hexaco, Hogan, all these other psychological inventories. What then we realized was, well, we thought that would be helpful to give a more holistic view of a candidate in terms of how they'll fit into environment, how they'll work. Are they, you know, they work well in teams? Are they more an individual type, introvert, extrovert, and so on? It goes on. We actually detect about 100+ traits today. And then we realized, well, we're, we're actually creating more work because the— these— for a recruiter to understand all these psychometrics was a task in itself. So we then decided, which is what I'll show, to make a a simpler system that would automate the, the interviewing process and give simple metrics to the recruiter. So it takes, takes a lot of the work off their hands. So, um, as you know, like 30%, yeah, or, um, of employees will either are bad hires, will leave within usually 6 months is the stats. So it's quite high. Part of that we believe is because they don't— it's not just your skills, but it's more than the skills, it's the environment. So we, we recognize that as a big issue. So what we do is we have a really initial step of pre-screening, which I won't show today, but what this pre-screening is, is When, if the candidate, for example, those are the questions like, are you willing to work hybrid, uh, work on-premise or move to a location, uh, possibly a salary range, just a pre-screening if they qualify or not, because there's no point in going through all this process if they're not.

That part is written or that part is verbal?

That part is actually, it's both written and verbal. So they, it's like a form so they could fill out. A multiple choice, or they could speak to it both ways. And it's usually short, and it's the recruiter that makes it up for each job, whatever the requirements are. And then once that's done, uh, and if they qualify the pre-qualification, then they go to the interview process. And that's what I'll be showing. There's two sides to the interview process. There's the recruiter side where the recruiter puts in the, uh, job requirements. And also these soft skills, the personality types that they're looking for. And then the other side of it is the candidate. So they send the candidates the link, and then the candidate participates in this automated interview with the AI system. And then finally, the recruiter gets a— all the metrics of their performance, and they see ranking of who had the best match for based on the job requirements. And I should say, if anyone has questions along the way, just—

oh yeah, nobody.

Okay, great. Yeah. Okay. Then, so yeah, these are the two sides. So the recruiter does the job description, personality criteria, and publishes the job posting, and the candidate fills in basic details. They upload their resume and they complete the interview. Um, what we do is the questions are asked based on the job requirements that the recruiter put in, the, uh, candidate's resume, and their personality traits and emotions. That's how we That's how the questions are. So if we see, for example, there may be a need that the recruiter put that they have to work in a stressful environment, then there may be that those— the— that will be measured and it would come into the questions. Um, let's see, and that, that's— now this is the point where I want to go to the interview and show the system. So let me see if I can pull this off. And I'm logged in here. This is the, this is the recruiter's dashboard. And here I'm not going to get into all the details, but they can create a new job role. They put the title, language, they can put the the type of tone of the interview, because this is more general, so professional or casual, friendly, the domain of the interview. They put the job description here, and then more importantly, they put either the number of questions or the time limit for the interview. So if they want all interviews to be, for example, 3 questions, then they select that, or if they want a 1-hour interview, they can also put it in in time.

And when— where kind of, Adam, is the diminishing returns of number of questions and time?

So, uh, in other words, like, uh, I'm— I didn't quite follow.

I'm sorry, I'm sorry. So if at some point the system can create a personality or an emotional, uh, signature based on 3 questions, you don't need 30?

Exactly, yes. That, that's what's so interesting. In fact, it's really The psychometrics and the emotions, we only need 60 seconds. Wow. It gets a little bit more granular and perhaps a little closer to accuracy, but really within 60 seconds, it's reading the person's emotions and personality. Great.

That's really one question then.

That's really one question. The difference here is— good point. Yeah. The difference here is why we have more questions is because it's also looking at their experience and asking them questions about that, how, you know, what their experiences has been like. So it's also collecting data about them that gets transcribed and put into a report for the recruiter. But as far as psychometrics and emotions, that's And then lastly, they put in here, we broke them up into 6 groups right now. Uh, the type of, uh, these are the personality traits, psychometrics, and emotion that into stress, pressure, professional competency, work styles, etc. And then we, uh, and then we also have the ability to select the voice type for the agent to ask what kind of voice it wants to use., and that's basically it. And once that's done, they can actually email this out to candidates. And so let me show that part.

So, Adam, I've got a quick question. Yeah, please. So you're talking about taking 60 seconds to measure a person's emotions. How much does you know, what, what they're experiencing right now, you know, they had a bad day, they had, you know, an argument with their spouse, you know, they had a death in the family. How much does that change the answers that you get?

Great question. Yeah, so it's based— it— the, the things that really change are the emotions, the psychological traits, uh, as far as like creativity, an organized person, agreeable, openness. All those are really consistent. They do change over long periods of time. We're talking months, weeks, sometimes weeks, but months typically. Some of them even over many years. But the emotional traits are what would change, like if they're angry, sad, whatever it may be. Those type of emotions will vary with, like, if they're having a bad day, but their psychological traits will not. And maybe just for anyone who's— I won't get deep into it, but it's a common question. So the way we developed this, we're language agnostic. So we use tones of the voice to determine the traits, which is really fascinating. We actually trained our— we worked with other companies to collect the data. We didn't do that on our own, but we shared the effort. But we used we, what we did, just so it's understood, is we had various, they call them inventories in the psychological world, of tests that they participated in. We use the scores on those tests to understand what they were, what their traits were, and then we also did voice samples of them, and that's how we built our models, and we built it using a dozen languages. What I'm happy to say, we have found that we have worked just as well with languages that we did not train on based on the tones. Fascinating.

Adam, I'm a retired psychotherapist, therapist turned writer turned business person, and I think this is truly fascinating. I'm wondering, are there roles where this works less well?

You know, less well for a surgeon than for a lab assistant, for example. I, I would not say so. The key is really is the— I think the challenge is, is that one has to define the right metrics to look at and to seek, and that this will look for those metrics that that person, those traits that that individual has, and look to see how well they match. If they overlook or they ignore one or, you know, they just didn't pay enough attention to it, that could be one issue. Another thing is, for example, we have traits like leadership and the type of leader or their growth potential. These may not have been important initially, but later in their career they may become more important. So, uh, so we, we believe that there's another market that for evaluation and for growth within a company, for assessment, that that could be done periodically in, in someone's career. So you wouldn't want to just go with what they had during the hiring time.

Thank you. Remember, when as a working as a pilot in the military, when you would fly with some people and you would get like a strange noise in the engine or transmission or some yellow caution light would come on, people would sometimes act a little differently and you wanted somebody that would stay cool and, you know, rational and, and not go into tunnel vision and get weird or anything. And then later on in my medical career, watching surgeons in the operating room, especially trauma surgeons, I detected that there were certain personality types that just, you know, were actually— it's almost like time slowed down and they got better when this— when the bad stuff happened. And then others didn't, and they were like those pilots that you didn't wanna fly with. So that's probably something that you could easily measure, wouldn't it be?

It, it, it, it's true. And that's something we're, we're looking into, it's— we were looking at it as it's an occupational risk, high-risk jobs. And again, that comes into the medical field as well for, you know, stressful nurses, doctors, surgeons, so on, in stressful situations. Or as you said, Mark, in like pilots and other law enforcement, other things where you need to see your— not only general, like the trait of the person inherently, they're calmer under stressful situations. That's one thing, but also their alertness for the day. Maybe they, you know, maybe they, you know, are sleepy and tired or, you know, whatever the situation may be, and they're not as alert that day. That's another thing that we're looking at that to do a check-in before taking on that, you know, the role for that day. But that's still— we've talked about it, we haven't actually, and we've done some testing on it, but we haven't yet moved forward with that as a product.

And thanks.

Sure. And then I'll continue here. I'm looking at another screen here just to— so let me go to I believe I could just go to rate. So, okay, so, uh, this was— these were pre-created job roles and requirements that, that the— this is the recruiter's dashboard. And so they can see, for example, the job description they had, uh, you know, performing through interpret diagnostics, imaging, whatever, good to have. Other skills. And then they can invite someone from the dashboard that's candidates, or they could post the link out to a website or put it in an email. So there's multiple ways. And then these are the candidates that they had. These are— we'll see this better once I go through it, but those are the, the 6 group areas that I showed earlier, the soft skills, professional competencies, work style, personality, stress and pressure, cultural organizational fit, and career growth. Those are the 6, and then that's not specified if they didn't select it. And then they can actually see— I'll go through one just— I don't— let's go through one just to see what what the view looks like. So they looked— if the recruiter wants to, then they get on their dashboard. These— this would be like a typical report. It would say, give you a little bit about the overall, um, score on the interview. It would give you the things like the tone, clarity, empathy, those just general metrics of the interview. That, that we measure as well. And then we get into the specific traits, those 6 areas. So for example, job compatibility, professional competency. Well, this one scored low. Um, these are just samples, so, um, but the professional competencies look at things like cognitive ability, communication, problem-solving, leadership. That's the type of Those are these, the traits under the professional competencies. And we give a score for each one of these. And then if I look at, look at a couple others, uh, some of these scores like work style and ethics, um, for example, work approach, some of these are, um, are not just one on one level. There's an opposite. So are they proactive or reactive? So they're, they're opposite. So they'll have both. In two colors, proactive, reactive. 66% proactive. Team collaborative, this person is more independent than collaborative, but they do have a score of some collaboration. And work ethic and integrity to others. We can continue just to— let's jump to, for example, stress and pressure. So stress response, resilience, sensitive, work-life balance. In other words, do they appreciate other things? There are other small traits that go to build these traits. Are they open, adaptable, and all that will come into that, into their work life, pressure handling under stress and recovery. That's another thing, their resilience to, to to the pressure. Some people can recover very quickly, whereas others take a longer time. And then we do similar things for cultural fit and so on. So, sorry, give an example of how— so the resilience fit, walk us through, if you don't mind, an example of that, somebody who doesn't make the cut and what exactly is being measured. So pull out my list here because there's so many traits. But for example, I don't, to be honest, I have some idea I'm probably going to miss some of these traits, but resilience, so would be their emotional stability, coping ability, how they cope with things. Their energy level is part of their resilience, like if they could be a low-energy person. So energy, all these little sub-traits are what helps. And they're weighted, but these sub-traits are what comprises their resilience to For example, work in a stressful situation. Let me just— I'm checking emotional stability. Emotional state also is another thing, not the state in which they're in at that moment is another factor. So that's more immediate. And then their stress control is another subtrait, how well they control their stress. And I'm sorry, you're getting all of this from voice or voice and questions? Actually, this is all from the voice for the psychometrics. It's all from the tones.

We're dying to see the demo.

Yes, you are.

It's fascinating.

Let me do that. I'll be happy to. And then we can continue.

So why don't you do a demo on Tiger Buford?

No, no, no, please don't.

I'll be happy to do it.

You got my wheels turning. I mean, for startups, you want people that work urgently. High-stress areas, resiliency. Resilience. Yeah, I mean, that problem solving— you probably can't do problem solving with voice, but man, you could dial it in for a startup job. Really?

Yes, yes.

That brings up a question I have, Adam.

Um, the—

shoot, I can't show the damn thing.

Oh, okay.

When, when, when you're, uh, using the, the the voice to analyze this, um, it doesn't matter what the answer is. Is that what you're saying? I mean, I can, I can answer completely wrong and you guys get everything you need out of it.

Um, but that's where the questions come in too, is it's asking you, for example, okay, we'll see it in a minute, but like, tell me more about that experience that you had with your, uh, you know, that, uh, difficulty in your last job. And why— or tell me why you felt you performed very well in managing your group. What were some of the challenges? So there are questions that are related to the job, not— so it's a combination of these three main factors: emotions, psychological traits, and the responses, and the job requirement, of course, but the responses to the questions that are generated And I'm sorry, I clicked the wrong—

just get to answer Eric's question while you're looking for that. You know, you can score high on your voice with all the right traits, but if you say, I know nothing about engineering whatsoever, then I think that's going to come out. Yeah, if you're not a good fit for the job.

Exactly.

No, you have a great tone. Those words turn into text.

Yeah, that's right. So it's really a combination of both, you know, of the psychometrics matching, and then you have the job match as well, the experience, the situations, the projects. And I just sent myself— let's just give me a second here. Hopefully— yep, got it. So I got something. I'll just put it up here. Something like this. I got an email It gave me the interview link. So if I click the interview link— oh, I'm sorry. Yes, this— that's right. So it tells me this is what the— so this is the candidate. It just gives them a summary of, you know, you're going to be 5 questions. I'm not going to go through all 5. It's, and it's a little challenging for me because it's an area I'm sorry, not from, you know, you have to, I actually have to have a resume so I could pretend I'm someone and give some something sensible as if I'm a radiologist. So I will, I have a resume up on my other screen just to pretend, yeah, just to give something. Okay. So what I'll do here is Let's just say this is, uh, Natasha Watson, and I'll use my email again. And then we proceed. There'll be a little— okay, so first thing it's going to do is ask me to submit Natasha's resume, so I'll do that. Um, here we go. So I drag and drop the resume.

So this is the link the candidate would be getting. Yes, would drop their resume into the little box.

Yeah.

Okay.

And then I click, click to proceed to interview. I hopefully— yeah, there's a, there is a prompt. Let me do it like this because I'm on my system. Here we go. I'll just do it like that. Now that should take it. There we go. Yeah, sometimes my system on a Linux system. Good afternoon. Okay. I am your AI interviewer assistant. To get started, please give us an introduction of yourself. Okay. So just one second. Okay. So I go ahead now and I say, Hi, I'm Natasha Watson. I, I'm applying for this radiologist position. I've received my doctorate in medicine from Columbia University in 2018, and prior to that I was at Howard University, my undergraduate in biology. I've been working and presently am at Presbyterian Hospital in New York as a radiologist. I've been there since 2018 when I graduated. I think my key values are I'm very accurate with my diagnosis and treatment plans. I work well with physicians and collaboration well. I've got a lot of experience with, with imaging tools and software and CT scans. And I am also fluent in both English and Spanish. A little— we're going to be improving this response time, but this is something we'll work on. So we'll be— so that progress bar that we just saw was just kind of specific instance where you had to adapt your imaging techniques or approach based on a patient's unique needs or circumstances? I'm sorry.

No, no, keep going.

I didn't want to— Oh, OK. Yeah. So this would be the kind of question. So it starts going a little deeper into something like this, which was— Can you describe a specific instance where you had to adapt your imaging techniques or approach based on a patient's unique needs or circumstances? I'm really out of my element here to speak to this, but I'll make something up. So let's just say I'll do one more here. So, well, quite often the— in radiology, sometimes the images are not clear. There's certain areas of particular interest, that we would like to see. So I will, uh, ensure that we, we, that the images are detailed enough for the areas of concern. Um, and, uh, what, and I'll also confer with the doctors what they're looking for, particularly what their concerns are when I present the images. Oh, and then it does ask— we do right now, we're saying 40 seconds, uh, as a minimum time. So I'll just continue. And, uh, so sometimes I'll use also, besides X-rays, we'll also— we— I, I am able to conduct MRIs, CAT scans, and also present those images to the medical team to review. Can you provide an example of a specific case where you had to adjust the imaging technique, such as choosing between X-rays, MRIs, or CAT scans, based on the patient's condition or the doctor's request?

Request?

I'll probably leave it at this because this is, you know, goes more into it. And if I gave better answers, I think it would go— it would be, you know, it would go deeper into those. Uh, but this should give some of the idea. Let me see, I had this— unfortunately was 5 questions, so I won't be— uh, I don't want to, uh, take up the time just kind of mimicking what this radiologist, but based on that, we would then get what we saw back here in the dashboard. It would then show as a radiologist that this, uh, uh, oh, I'm already there. Uh, that's one. It would then show that this, this, new— this, uh, this one here, Natasha Watson. If we look here, uh, once it would be complete, it would say email verified. It's not ready to view because it's not complete. Once I complete the interview, it would show up and scores would be presented for each of these categories of the psychological traits.

So this is all done remotely through the AI. Correct. And whatever you program to beforehand, so you could do 1 candidate or 100 candidates.

That's absolutely correct, Mark. And that's really what's time-saving for, for a recruiter is this could be done in parallel. Like I said, you could, you could be sending this out to 100, 200 candidates, pre-screen, get the ones who qualify, and then from that list get the responses, and then here the recruiter could pick the top 5, the top 10 ranked candidates that match the requirements, and then that, that would shortlist them for the— for a human then to, uh, to interview the shortlist. We're not trying to be the whole process, I should say that. We feel that, yeah, people want to still meet.

What, what kind of, uh, what kind of positions or roles do you feel this would be most efficient at or most useful for? Would it, you know, professional positions, uh, positions that are more sort of quantitative or more qualitative, you know? Um, I mean, talking about a radiologist is one thing, talking about a marketing person is another thing.

So the general skills would be, yes, so it's true, but the skills that you look for are like, for example, their communication skills for a marketing person, their creativity, their innovation perhaps. Those are some of the traits that we would, that are used under the hood, so to speak. We used to report record all of these. And then we said it's just over— because it's over 100, it's overwhelming. And we found that we almost had to train them to explain each of these skills. But so, so to answer the question, we believe that the traits that we have will work across all the job positions. As far as the environment, stressful, non-stressful is common, innovative, leadership, independence, collaborative, communication, openness, inquisitive, wanting to learn, or someone who's not. So they're kind of general in a way, But it's important to say what, what, what the team or what the job is looking for.

It looks to me like it'll evaluate any role, Mark, but it's the way— but you've got to pre-select the right questions. Like, like you said, marketing, you're looking for creativity, innovation, communication skills. You wouldn't do that with a lab tech, maybe. Correct. It's— there's a little front-end work to sort of dial in the right traits that you're looking for, but yeah, it can work anywhere. That's— now I have a question. If you, if you send it out to 100 people, let's say, say 80 fill out the, the voice survey, will it— do you have an analytical tool that, that compares the 80 candidates that come back?

Yeah, that's what I was Exactly. Uh, so this is what this would do. So you would see a dashboard. This is like the recruiter's dashboard. This is prioritized, but you would see which ones had the best match. Um, I can go to another one.

This is a really, really good tool for bulk recruiting for larger companies.

I think that I, we agree, you know, we, we were really trying to find where we fit in best. Where I think where we are is bulk to take large numbers and to shortlist it, which is the time-consuming process. And then what's nice about it is the recruiter or the actual, the company that's doing the hiring will now can see some report and get an idea of that candidate before they come in to have the face-to-face or, you know, online interview. So it's helpful.

Yeah.

Yeah, I can see, I can see the buyer of this would be the head of HR in a large company to give the tool to their internal recruiters who, because they're completely overwhelmed, they put out jobs on LinkedIn and Indeed and they just end up with thousands of resumes and, you know, they have no idea. You know, they can't— there's— it's— they can't even— first of all, it's an HR person doesn't really know what a marketing person does. And, and then second thing, they, they get too many resumes to even figure out what to pre-screen. I see that all the time. They go into the black hole and I've heard they use bots for keyword searches.

Yes.

Yep.

So, but this would be way better than using bots.

That's right. That's exactly— I think this, this is taking a much more comprehensive view of the person, not just keywords and, and not just their skills, uh, more than that.

What is the— is it a monthly subscription?

Yeah, so it is a monthly subscription, and I'll just show that last little bit. I think it's here, uh, where did I put it? Uh, Yeah. Is this it? Nope. I just had— oh, here. I'll just throw up the PDF for the last slide I did. So what we do, our cost comes out, we do a monthly subscription. One could subscribe. It's not in terms per candidate. You buy in blocks of candidates that you want to do 100, you know, 1,000, whatever it may be, depending. We have different block offerings, and we— our starting price though comes out to $25 per candidate. Um, and exactly, our benefits no longer, you know, it's not a sequential process like we talked about. You could send these all out in bulk to any number of people, and it's gives you both hard measures, both on some of the hard skills and soft skills in the interview process.

How many, how many clients do you have right now?

We have, um, close, uh, 8 right now, and we're expanding. Yeah, we've got a few in the pipe, but 8 at the moment.

Yeah. Have you thought about sending out kind of a freemium version where you get the first 5 or 10 free.

We do that, and actually that's what I put here. You can, if anyone's interested on this call or at any time, if you email, just say you're interested in the trial to recruiter@echo.ai, we'll send you— we have that coupon you saw maybe earlier. We'll send a coupon and then we can give you whatever X number of processing that one wants to do without charge. So that's an option.

Have you talked to the military?

You know, uh, I haven't. I have that in mind. I've really— we, we, uh, I tried to make some connections. I just didn't at first, but I think it's a great area. Uh, we, we got, uh, actually there was a an open request for proposals, not from the US military, actually from the Chilean Army. We applied for that just because one of our people, our team members, is in Chile. Uh, we did not, you know, we weren't selected, um, but we believe it, it could be good. And, and what we had positioned it for was not for combat but for training, because training could be from what I understood, could be also stressful and demanding, combat training and situational training. So that's where we were proposing to initially do it as part of the training program to introduce it, to see who could work, but as you said earlier, you know, in noisy environments, in stress conditions.

Wow. Yeah, another thought is healthcare. So right now my mother is, is kind of has early dementia and she has a cognitive review every 30 days. And it's a, it's a nightmare. It's, you know, I've gotta be there live and we do it on a video call and there's a tele-nurse on the other side and they ask questions and then they give her, uh, you know, a dementia score at the end. And I mean, geez, this would be so much easier. What did you have for breakfast? You know, who's the president? You know, and, and it could— it would even tell the emotional state of the person. You know, it'd be way better than what the nurse can do.

And, and it's, it's a great area, and that's where we see it eventually expanding into. We've been doing, um, anxiety, stress, and depression, and we're working on, um Parkinson's and Alzheimer's. We're not there yet. We, the problem is we need more data to validate it, but we think we got a, you know, we'll be there in the coming months. And it's great, as you said, Tiger, that it's a great check-in because people could check in more frequently.

You get a baseline that you can see if you're going up and down every month.

Yeah. I think I took up a I don't, I'm sorry. I may time-wise, I just see. So, uh, oh no, no, good.

Perfect. I got, I got some more questions.

Oh, okay.

Um, always, I'm sorry. Um, so to the example that you were giving there, you, you were answering questions as if you were a radiologist. Um, but you're not. What would your scores have come out? Would you have gotten all zeros because it says, hey, this guy was bullshitting?

Well, it looks, uh, so two things I'm looking at, you know, I'm looking at the resume and reading a few things from there, but it would look again at my, uh, the questions and the answers to them. That's based, you know, that's one level of metric. In other words, the skills, the experience, That's at one level that's measured. But at the emotional and psychological level, it's really— it will just report on it. It doesn't care if I'm not skilled in radiology, know nothing about it. It's still going to say maybe, okay, this is an open adaptive person, whatever. And that's all it's going to report. Could have been the same report for, you know, a computer scientist or whatever.

Yeah. Right. So, so that's, that's one part of it, but the other part is, hey, do you know what the heck you're talking about? Would it have caught on that you didn't have a clue? You were just reading off a resume.

That's part of the, um, the, uh, yeah. So that's, again, that's, I could be doing that, but I could also be doing that in, in a live interview, right? A face-to-face Zoom interview. I think I could be doing that.

Maybe not. Eric's asking, is there hard skills and then there's soft skills? This is more of a soft skill tool. I know. Yeah, they're assuming you graduated radiology school, you know, you know what you're doing. It's really a soft— it's a— it's more of a cultural fit tool, I think, which is the harder part. I mean, that's why people leave after 6 months. It's— it's not. That they're not qualified. I think I only hired one person in my entire career that was not qualified. He couldn't run the HP machines that he said he could.

But, so, so the other question I have is, what would you guess would be your rate of, you know, I guess false negatives on here? You have someone who comes in who could really be the perfect person for the position, but this didn't measure it right. What is the rate of— how often does it get it wrong?

So we're typically at 80% accurate. This is a big— this opens up a big topic in a way, because psychology is kind of the science and the art, and there's big questions to it. And but We're at 80% or better. We feel we're better. And no one, as far as we've looked for it, we haven't seen research done. We did internally. I'll just say one thing we did internally with 20 people. We gave them 3 different Big Five tests, the Ocean Big Five, openness, conscientiousness, and we found that none of them were consistent. Consistent with their results over— across— generally across everyone. What with the voice, what we find is there's consistency. So if you— and we did this over not one, one time, we did this over— I think that was done over 8 weeks, 6 or 8 weeks, I forgot. And we saw once a week, and we saw that the voice was consistent. And the written tests were not consistent, both not consistent over time and also not consistent relative to the others. That's about the best I could, you know, answer it. I mean, it gets more involved, but yeah.

Sure, it is.

It's very interesting.

And, you know, I come from the engineering side, so the mistakes are a little bit different kind of mistakes for me than than, you know, the soft skill kind of things. So that's, you know, I asked Mark the other day, I said, how are these things kind of validated? And what you're saying is that's difficult.

It really is. And the thing is, is that, yeah, so even psychologists don't necessarily agree on always on the same thing. And also, but more so for interviews, What happens is it's subjectivity, whereas we're not subject, you know, the AI is not subjective. It doesn't care how the person looks, whether, you know, if their accent doesn't matter. It's what their personality is. That's what it's reporting on. So in a way, it's, it's, it's much more objective.

Yeah. Well, and that would be nice because it would, it would be a help us eliminate our, you know, biases, which, which would lead to mistakes on our part. So yes, very, very cool. Um, is there a way, um, that this could be programmed such that it, it would be able to tell if, hey, will I be able to get along with this person? Well, I like that. Um, can it do anything like that?

We, we've talked about it. It's a, it's an interesting area because you need to look at let's say for example you could hire someone for a team and you could do the team's, uh, overall psychometrics, take each individual, average them out, and say this is the makeup of this team. Now, do you want to hire with the same type— and this is another open question— do you want to hire someone with the same kind of personality that fits right in, or do you want someone who will change the team, maybe bring more innovation, or more management skills. Yeah, exactly. So that's a hard area as to how you want to do it. It's a something we're thinking about, but it still needs more work.

Really good question is cultural fit. And I work for a company and we were trying to hire people on the opposite— we did the wheel, I don't know what that was called, but we're trying to hire people that are complementary skills, uh, they would ask questions the other people wouldn't ask. Uh, that's really interesting. Another, another idea popped in my head. I have a theory that all, um, that the psychoanalysis psychology world is going to all go to AI agents in the future. And, um, I mean, I've got a friend I used to play tennis with, and he was His specialty is sex addicts, and he literally did video calls with sex addicts all day long, and he charged, you know, full rate. Um, and but he— this would be better than him. I mean, couldn't you have it? Couldn't you have an agent that just checked in on, uh, you know, a patient once a month and Precisely.

Yes. And I, I could say the same is true for recruiter. Right now we, we're targeting recruiters, making their life easier and shortlisting and all that, which the therapist could do also. But I could also see at some point, and then the recruiting firm may not need as many recruiters because this could do a lot of the legwork. I could also see down the line that companies, especially larger ones, would just go with something like this and cut out the recruiting recruiter's cost. Usually it's 15-20%, whatever, you know, that's, that's the numbers I know. Uh, and they could cut that out and they could then just use something like this and then bring them in, uh, for the final interview or two. So yes.

So I have a question.

I'm sorry, go ahead.

Yeah, uh, hi, this is, uh, I'm not on video. Um, how does this work with sociopaths? Because they're pretty good at disguising themselves.

It—

what, what this makes it— what's very difficult to fool with this system is They might be able to disguise, but it's with the words. With tones, it's very hard to. It's really hard to because we're looking at very different things. Like, we look at the, um, the frequency of the tone, the hesitations, the pitch changes in the voice, the formants, the resonating tones, the fundamental, secondary tones, all these things, things like also certain vowel sounds, the duration of it. These are like features that we're looking at. It's very hard for someone to know what to control to portray a particular type.

So the reason I asked the question, particularly salespeople, you know, that's where you can find a sociopath. And, you know, you want to talk about a cultural, you know, atomic bomb, it's a sociopath getting into the organization. They just destroy it. Um, so if there's a way to identify that type of person, you know, ahead of time, that would be absolutely huge. Um, there, there— I, I think everybody on this call has encountered at least one sociopath in their careers, and there's, it's just a horrible event. And then just a couple of observations. So I use some software called Welltory, and what Welltory does, it measures my heart rate variability, and it's amazingly accurate in showing me just how tired I am, getting sick, all kinds of interesting things. Then the second comment is about Professor Marston. Do people on the call know who Professor Marston was?

I, I don't.

Okay, so he's the one who came up with the first lie detector.

Okay.

And he was absolutely, you know, he— this is early, um, 1900s, uh, 1910 or so, and he was a professor. And, uh, interestingly enough, he, uh, he had a nice taste in women as well. And, uh, he was married, but he decided to bring a grad student woman into their relationship. And then after he kind of got, um, uh, shown out of, um, the academic institution, he decided to write a, um, comic book, Wonder Woman: Lasso of Truth, that came from his lie detector. So picking up on physical findings to tell the truth. So just kind of throwing that one out.

That's interesting. Very interesting.

The psychopath question is, is excellent. So can you I know it's based on machine learning. Can you teach the machine learning? Can you give it Charles Manson and David Koresh and a bunch of people and basically say, you know, these— they're extending their vowels and this is a marker of psychopath?

We could, uh, and I would— I don't know the statistics on it. We would have to say, you know, it's really a question. It's something we would do, I think, at at the early stage we're in, we're going to go for the larger market. But if it is, yeah, then there's no reason we can't.

Oh, let's just use softer language instead of psychopath. Uh, yeah, a lot of salespeople don't tell the truth. Uh, can you, can you, uh, detect people that are kind of loose with the truth versus people that are objective?

It— we We don't at this point. Yeah, we would have to, uh, that would work on that. Yeah, that's— there's a lot.

Oh, you can always improve that, sure.

Yeah, one thing we do do, and we're going to add that— we, we actually, there we got interest in it, uh, as a standalone, but we want to integrate it in— is we do— I'll just mention it because it's— we do a fake voice detection. So with the emerging of AI voices and replication, we can, we can, we have very high accuracy there. We're at 93% accuracy in detecting the human to, uh, to AI generated, which is one, and also human to human. We've looked at imposters and we, uh, we can also detect that.

You probably— they probably know that Natasha sounds like a male.

Yes, something's up. But, but, uh, the, the problem came up, which I, I, uh, it was— I, I was only, uh, aware of it recently, is, uh, a lot of, uh, a lot of job candidates, especially abroad, uh, it happens India is a big market and a lot of companies hire large Fortune 500s have a big portion of their team from there. What they do is they have sophisticated ways in which they are doing AI voice generation, and then they find out that, that there's another person answering those technical questions. And so they want to determine on the same voice call, is it the same person or not? It's just a binary that we do, but As I said, we— yeah, that came up.

Interesting. It's amazing how voice has such a signature.

It's incredible. Definitely.

Well, this is— we've gone a full hour. This has been great. Thank you.

Oh, my pleasure. Thank you.

Oh yeah, we got one more.

Sure.

Very good. Thanks for it all. Thanks, Tiger. Appreciate it.

Yeah, good stuff. We're going to dive deep. I'm going to tell some people about this tool.

Please do. And if you want, you know, the, uh, did you get it? Or I should have put it in the chat, the recruiter@echo.ai. Just email there interested in, and then we can get that. Uh, I'll just type it in.

Yeah, type it in now so we'll have it.

Thanks everybody.

Yeah, thanks for bringing it, Mark.

This is, uh, Fascinating.

It's— I mean, why the world's changing so fast, it's— and these tools, there's tools like this all over, they just pop up. There it is, recruiter@echo.ai. All right, well, thanks.

Oh, my pleasure. And Mark, I'm sorry if I— you know, we were supposed to both— you were to present too today, so I—

oh, I don't think so. No, no, this was, this was all you.

Oh, it was? Okay, thank you.

I mean, I might have been a backup, but no, this was I am doing a symposium on February 11th entitled Using AI in Your Medical Practice: Increasing Efficiency and Saving Time. And I've got a couple of large language computer scientists and the chief of orthopedics from Kaiser Permanente and a couple of other AI-interested physicians because, I mean, There's a lot out there, like Tiger said, it's moving so fast. How do we bring a tool that is actually usable, applicable, you know, that you can sort of ingest without, you know, redoing your practice, as it were, or hiring somebody. And it's, again, saving time and increasing efficiency. Because there's a lot of this going to be, it's going to be coming at us hard and fast. And, you know, busy people don't have time necessarily to shop everywhere. And there's no Amazon yet for AI.

Thank you, everybody. Thank you very much.

I appreciate it.

Merry Christmas.

Merry Christmas. Happy holidays, everyone.

Thanks. Thanks, Adam. Good stuff.

Bye-bye. Bye.