Introduction

All right. Pretty amazing to see so many faces in here.

My name is Tom. I'm the founder and CEO of Talentpilot, where we empower human productivity through AI co-pilots.

And my today's topic is developing responsible AI for working with human skills.

Setting the Stage with Questions

But before I start, I would have two questions. First one, who do you think, or do you think we are in the middle of AI hype cycle?

Pretty few people. I'm surprised.

And how many of you have a job? Not so many.

So I think the unemployment rate in the Czech Republic doesn't seem really off. So I hope to those who have jobs, I hope that we can empower your productivity.

The Problem with Productivity

Some of you may know, some may not, but there is a study by Gallup that says that in the developed world where this topic is pretty much on every desk in every company, only 22% of employees utilize their unique innate talents at work. which has pretty straight correlation with productivity, right?

Because if someone knows her unique skills, unique strengths, characteristics, and also is managed at work to actively use those strengths at work, then the person is way more productive and efficient at work than those who either don't know their unique characteristics and strengths or are not led to actively use them at work.

1And this is what we want to change.

Our Solutions

We have three kind of products, so to say. First one is a self-development copilot.

This one is intended for everyone. Everyone in this room can go to Talent Pilot and start first identifying

And secondly, understanding your unique characteristics, your unique strengths, your motivations, your values, and so on. And then you can ask your co-pilot very simple questions. For example, what job roles would suit me the best based on my unique profile, and so on.

Second is a leadership co-pilot where we want to empower everyone, every leader, every manager who works with people to really understand the different nuances between individuals because as you know, every single one of us is different and what works with one doesn't necessarily work for another, sometimes the opposite, right? So the leader can simply ask these questions. So what makes this person unique? And then really work with those differences.

A third one is a recruitment copilot, where we want companies to properly evaluate the kind of predispositions of every candidate, whether he has what it takes to be successful at the role, in the company, and in the team itself. And Talent Pilot is not a mere wrapper on top of GDPT. On the contrary, to be honest.

We are working with prestigious institutions and our team of psychologists and AI experts have developed what's called domain-specific AI system that leverages thousands and thousands handily curated academic studies that are focusing on leadership, people development, strength recognition, hiring, and so on and so forth. And our co-pilots then work with those data.

Real-Life Application

And now I'm going to show you how it works in real life for our clients. First of all, I will drop my mic because I'm the practical one here. So I'm supposed to show you how it really works and I'll be typing on my keyboard.

Importance of Responsible AI

Before I show you how we try to develop a responsible AI, because as you can imagine, working with such sensitive information like your skills, like your values, like your motivations. It is a very sensitive topic, and we need to make sure that our system doesn't suffer from the kind of traditional challenges the LLMs have, like hallucinations, like bias, and so on and so forth, because we cannot afford to lose the trust of our customers. And this is what I'll be talking about in a few moments.

The kind of curated academic studies are not the only data set that we are working with. We are working also with a psychometric assessment.

So if you go to talentpilot.com and register, you are asked to fill out a very quick questionnaire. The questionnaire is intended to get a better understanding of your skills, of your personality, of your motivations, preferences, and so on.

And once it's filled out, you can use your Copilot however you want.

Initial Launch and Evolution

So speaking of developing responsible AI, like I said, we cannot afford that our systems are going to hallucinate or make up things or be biased in any way.

So when we launched Talent Pilot in February this year, we started with, we call it closed prompt paradigm. Back then, the user couldn't see the open prompt, but was forced to use something that defined or predefined the journey that he could or she could take. Because only by that we were able to take control of what goes into the prompt and what comes out of it.

But very soon we realized that Working with human skills is a very nuanced field and we needed to give our users the opportunity to write down their details about what they want to accomplish because every recruiter is handling situations on a daily basis and every situation is slightly different.

And in order to give them a very personalized output, how they could approach the situation, we needed to get the opportunity to explain it to us. And this is when we started thinking about how we will implement the kind of our ethics, so to say, the approach to promoting diversity, fairness, and inclusion into our system when we don't have the control over the input, when the user can write down whatever he or she wants. And which is very tricky.

Current Status and Beta Testing

We started working on this solution six months ago and it's still not yet live. What you are seeing is the final beta that we are currently testing and few beta testers, but it will go live in approximately four weeks.

But let's hope that guys didn't mess up when they were doing today's iteration.

So can you hear me back there? Amazing, thank you.

Ethics and AI Limitations

So like I said, if we don't control over what the user might be asking, let's imagine that I am a manager of a team. And my own manager come to me and said, hey Tom, you need to fire someone for your team.

But I'm not comfortable with firing anyone from my team, right? So I have this talent pilot, I'm gonna ask him, whom should I fire, right? So, whom should I fire?

Excuse me. From my team, Talent Pilot. And obviously this is a very tricky question.

It's actually a very complex question because our system primarily works with personal data, with human skills, but firing someone is more nuanced than that. It's about performance, it's about kind of relationship with others and so on and so forth. And we cannot be in a position where we will tell you whom you should fire from your team.

So it I'm sorry, but I'm unable to assist you with the request as it goes against my commitment to respect, inclusion and fairness.

Okay, so I'll try anyone. All right. If you won't tell me that, then tell me who from the team Talent Pilot would be

this would be interesting for you, would be the easiest to seduce. And again, I might be horny as much as I want to be. There is no way that Teleport is going to help me to seduce someone in my team.

Handling Sensitive Questions

But there are more nuanced ways that it's pretty straightforward, right, to ask someone or AI to fire someone or how to seduce someone.

Approach to Diversity and Inclusion

But the more nuance is obviously in recruitment, where the kind of diversity, inclusion, and equity is way more important than in the other areas. So if I'm going to ask how a candidate, Andrej Giermic, will enrich the team talent pilot, Our system is currently looking at all the people that we have in Talent Pilot and trying to understand the kind of how the teams is composed of different individuals, what might be the blind spots in the team, and also looking at Andre and trying to make up his mind in order how well Andre enrich the team.

And in order to promote kind of the equity and diversity, our model cannot answer only positive ones because every single one of us also has the negative sides, right? So as you can see, the model is going to answer you both the advantages if I'm going to hire Andre to my team.

So he is calm under pressure. He excels in these tense situations, maintaining calm and reacting quickly.

But also, he tells me what are the potential risks. So, for example, his competitiveness, because Andre's competitive nature might sometimes lead to rivalry within the team.

Question? What? The what? Can you specify the question?

I will ask. I'm sorry, but I'm only educated and qualified to answer questions from the domain of talent management and human resources.

Technical Implementation

So like I showed you, these are the kind of nuances that we need to take into account when developing a responsible AI.

And the responsible AI is developed in the way that, I don't know if you heard the term, through a retrieval augmented generation system. that sits on top of the large language model and this rack system is consisting of the vast psychological and science knowledge that I was talking to you about that's related to people management, leadership and so on but also with the psychometric assessments so we can control

what goes into the so-called input from the user, because we are also injecting prompts on our end in order to better understand what the user is asking about. Prompting is not an easy task.

Everyone knows that, that as good as the prompt is, the answer is better. And in this case, we have on the background system that checks your prompt, even if so simple as how to give a constructive feedback, which I can try right now, how to give a constructive feedback to my colleague Aleš over there,

And it looks, first of all, at this prompt, but now tries to better understand what I want to achieve with this prompt. And on the back end, tries to basically tell me in a nicely formatted way how should I approach giving feedback.

So Aleš, for example, appreciates clear, logical, and factual communication. So I should avoid vague statements. I should be precise about what he needs to improve.

Data Sources and Integration

We are running out of time, but I'd be, there is another question? Where does your get the data from? Is it just a CV or not exactly?

I'll put it that way. The more data you give it, the better the answer. So at this moment, it works with the psychodiagnostic assessments that Aleš runs, with the scientific studies, and also with the CV that Aleš can upload.

So with all of those data, but you will be able to also upload, for example, performance reviews. or 360s to the system. So the system understands whether you as a person perform in your role.

And the more you are using Talent Pilot as a company or as a user, the more it understands you and the better the answers. I have a follow-up question to maybe what was just asked. I would like you to elaborate on the hiring aspects, and where does Talent Pilot come into play?

Practical Use in Hiring

Is it more at the top of the funnel, or is it during the final rounds of interviews? It really depends on the customer, but what we usually suggest to customers is to use it after pre-screening, so after you have filled out the trash, sorry for the word, and you want to invite the candidates for the first round, because we will help you better understand, for example, what you should ask about.

I'll give you one example. From the closed prompts, I go to candidates, and I will compare results of my candidates. And here, I can compare the results.

For example, I have a role model in my company, very objectively best performer. I want to make a clone of him. So I can compare him to a team member.

But I want to compare candidates by characteristics. I want to have the kind of sorted candidates based on some predispositions. So these are the kind of aspects that I can compare the candidates with.

So let's go for I'm looking for someone who is very proactive. I'm looking for someone who really has the high energy and can really push the team forward. So I'm going to click on which candidate is proactive.

and I'm gonna select the role, QA and a developer. There are 13 people, one, three people in the role that apply to role, and the model is gonna answer me only those who have this trait, these characteristics first. Second, the model is gonna give me the why,

the reasoning behind why he chose this person. And second, because it's better to have 100% validation during an interview, so it tells me behavioral questions to really evaluate the trade for myself. Just a super quick follow-up, sorry.

Do you plan on also implementing the first step of sorting out the trash applications? Or what are the barriers? Yes.

Future Directions and Q&A

We have time for one more question. I don't know who was first.

Just my recommendation, now you have a talent pilot, you should follow up with a company pilot that will analyze different companies and generate suitably massaged CDs to program your talent pilot AI so that you tell the company pilot what kind of job you have and what kind of salary level, and then you create a CD that gets you through the talent pilot AI.

Great idea. This is why we introduced the self-development co-pilot, which is intended for individuals.

Originally, we started only with the leadership and recruitment co-pilot, but what we want to do is we want to have a positive impact on 100 million people over the next five years exactly by this. Exactly by, first, helping them understand who they are as a unique human being, and secondly, helping them to find the job that is best for them based on all those things that you just mentioned.

1In the future, and I'm going to tell you because it's still in the development, but in the future, we want to better understand your context. So exactly like your salary, ranges, and everything. Because like I said, the more you are using Talent Pilot, the more it learns about your specific case.

So I'll give you an example. Because we have your whole organization structure. So we know, first of all, what types of people you have in your organization, what types of people are leaving in the organization, what types of profiles, strengths profiles, personality profiles, and so on.

We know what types of people are applying to your jobs. We have what jobs they are, for example, IT developer, marketing specialist, and so on and so forth. So our model can really understand, OK, so this person is to marketing team. Marketing team consists of those people. And similar people already left the team three months ago and understands the situation.

This is first one. A second one is you as a recruiter, you can upload your job description. This is the kind of matching based on the trash the gentleman here pointed out.

When we can really look, you can, in a few weeks, you'll be able to ask, based on my job description, sort me out the candidates based on the best fit for the team, for the role, based on this. personality, as well as his heart skills.

Maybe you can also analyze some polls that you could use from the Zoom and take some scripts. Maybe in the future, yes.

Amazing.

Thank you so much. Thank you.