Post | Martin Hurych

Everybody talks about her, but nobody's seen her properly. According to the latest surveys, only an average of 6% of companies in our country have saddled it in some way. And that's even the biggest ones. In small and medium-sized companies, it's at the level of magic. That's artificial intelligence in action.

We hear around every corner that humans will not be replaced by AI, but by people who know how to work with AI. I'm not sure about that. I think it could be much worse. It has already been shown how wrong we can be in our estimates of future technical trends. Anyway, one thing is certain. If a company doesn't have AI, it's as if it (soon) won't.

That's why I took AI "in the mouth" again in Ignition. This time we asked Kuba Krchák, the executiveDirector of Telma AI and an expert who has lived with AI today and every day for 20 years, as he sees it. More specifically?

🔸 How do you realistically get started with AI in your company?

🔸 Does it make sense to force people into AI?

🔸 What can AI realistically do in small and medium-sized businesses today?

🔸 How to choose a good AI application vendor?

🔸 What to prepare if I am considering corporate AI?

HOW TO EFFECTIVELY INTRODUCE AI INTO A COMPANY (INTERVIEW TRANSCRIPT)

Martin Hurych

Hello. I'm Martin Hurych and this is another Ignition. Before we get to the topic, I'd like you to consider giving me a like, a subscription, or telling someone about Ignition right now.That way you won't miss another episode and you'll help me bring social networking algorithms to the forefront and invite even more great guests like today's. Today's topic is going to be AI, but don't run away, we're not going to talk primarily about Chat, although we may get to that as well. Today, I want to look at what's even possible in AI anymore, broaden your horizons a little bit, and

maybe even outline in a few points how that AIization of your company could theoretically take place so that you can get ahead of the competition with AI. For that, I've invited Kuba Krchak, hi.

Jakub Krchák

Hi, Martin. Thanks so much for inviting me.

Introducing the guest and what is artificial intelligence for him?

Martin Hurych

Kuba is CEO of TELMA AI and co-founder of MAMA AI. What is artificial intelligence for you today, is it still something new or is it an old coat to you, is it something that one must already have, what is the relationship with AI of someone who has been doing AI for a long time?

Jakub Krchák

It's certainly very interesting and fulfilling to me that an awful lot of people are now putting AI down. Unfortunately, my grandmother died last year, but what was interesting was that at her funeral I met my uncle who told me that for 20 years he didn't understand why I was doing AI, such a clever person and such a stupid person. But now he said he finally understood. So the penetration of AI among more people has shifted a lot in these 2 years and that's great because there's a lot more people to talk to about it. A lot more people are thinking about how to actually do it and that potential has grown. Because in order to be able to solve something with someone, to be able to sell them something, to help them with a problem, the other side has to understand that they have a problem and that there is a solution. So there are a lot more people today who understand that their problems can be solved with artificial intelligence.

How does AI help in genetic genealogy?

Martin Hurych

How do you solve genetic genealogy problems using artificial intelligence?

Jakub Krchák

In addition to the things you said, I am also a co-founder of the Society for Genetic Genealogy. That's a non-profit society, a registered society, where we're trying to evangelize this thing in the Czech Republic and show people that it's an interesting tool. In America, for example, they say that genealogy and genealogy is the second most common hobby after gardening. Of course, it's because it's a pelmel there, everybody came from somewhere else, so they uncover a much more varied past there than they do here. But I think that genealogy is also quite a common hobby here, people are interested in where they came from and who their ancestors were. The genetic genealogy is about taking a DNA test and thanks to that DNA you are better able to do the genealogy. So he can verify that the ancestors that he has in that paper form from those vital records are really his ancestors. For example, he can find living cousins, distant cousins who may have gone to Brazil or Australia or emigrated to America at the end of the Austro-Hungarian Empire. But it can also do things like identify an unknown ancestor, say my grandmother was the daughter of an unmarried mother, and that DNA can help me find out who the father was. It's a very interesting tool.

Martin Hurych

When will I get to enter my name into the internets and have my family tree run up to White Mountain?

Jakub Krchák

Someone has to do the job. There are many places today where that AI is helping, and one thing that AI can do better and better today is read even those old texts. In general OCR is getting better, we are more and more able to digitize documents and even in that genealogy area we are able to read those old things better, which helps. But we still need a person today who puts these bits and pieces together and builds these trees. But I can imagine that in 5, 10 years, there will be an agent technology, an AI-powered agent that will be building that family tree.

When did he first get into AI and what has changed since then?

Martin Hurych

You've been in AI for an awfully long time, we should say you originally worked with AI back at IBM.

Jakub Krchák

It's true, I worked at IBM for over 20 years. I started there as a student when I was studying artificial intelligence at Matfyz, so I helped out there. Then I continued there gradually as some software person and then gradually some manager in research and development at Watson.

Martin Hurych

Were you already at IBM when IBM beat the chess player Garry Kasparov?

Jakub Krchák

I wasn't, because it was 1997 and I'm just a little bit younger. I joined the company on January 1, 2001.

Martin Hurych

What has changed fundamentally in AI since then?

Jakub Krchák

Artificial intelligence in general is an evolving concept or goal. Part of one definition of artificial intelligence is that it can do things that people don't expect it to do. Some of the things that were called artificial intelligence in the 1960s, for example, may not even be called artificial intelligence today, it's some kind of a rudimentary application of machine learning. In terms of that moving target, of course, how we do that AI has changed, but that AI is still about helping people solve their problems where there's too much data, the data is flowing too fast, or conversely, it's an awful lot of tedium for those people. Compared to what we automated 20 years ago, today we dare to do higher things and help with copywriting or lawyering.

How to build a business in an environment of rapid AI change?

Martin Hurych

When I was reading the preparation today, I realized one thing. A lot of my clients kind of grumble between the lines that the last few years have been terribly turbulent, here Covid, here war, here some energy. You work in an industry where business changes under your hands in the order of months. What are the lessons for us and how can you develop a business in such a jungle at the same time in parallel to the big beasts like OpenAI, Microsoft, Google, since yesterday Apple? How do you live in that?

Jakub Krchák

We call it a polycrisis, that there are actually several of these fundamental socio, geopolitical problems now. It's certainly a very challenging environment for all of us, both for the end citizens who have to deal with it, and of course for us as entrepreneurs. The turbulence in that AI world is a very challenging thing, and I always say we have a lot of smart people, PhDs, people who have been doing AI for 20 years, who can really innovate beyond articles to things that nobody has invented yet. But still, we've got a lot of work to do to run as fast as the development is going. But it's interesting that most people in the world can't even run that fast. So just the fact that we run as fast as the development goes forward is a very strong asset, a strong value proposition for working with us. We can advise people on what to do, when it's better to buy graphics cards and do it with an open source model on their graphics cards than to pay Microsoft an awful lot of money and vice versa. To know how to do that is

hard and it's even harder for someone who just wants to do their business, so we can help with that.

What is the adoption of AI within the Czech Republic?

Martin Hurych

I don't doubt that at all. I understand that the answer is primarily directed at your company, because there must be more than one AI expert here. What do you think is the general adoption of these new technologies, artificial intelligence and machine learning among Czech companies? Are we as a country really someone who is running ahead?

Jakub Krchák

We are definitely moving forward. According to Eurostat figures, in 2021, on average, 3% of companies in the European Union had AI in place, and from the 2023 survey, it is 8% of companies in the EU. The Czech Republic is ranked 6% there, which I don't think is a bad result, because there are a lot of people behind us at the tail end.We are not at the very top, there are, for example, the Estonians who are very digital, or of course the Irish, oddly enough Spain is also very much ahead. I think that the result is not that bad, but on the other hand it says that still the 90% do not have the AI and this number is pulled forward by those big companies, financial institutions and when we look at the small, medium business, there are maybe 3%.

Martin Hurych

So should we be proud of something, or should we, in typical Czech fashion, pull out our reprimands and start massaging ourselves that we are behind the mushrooms again?

Jakub Krchák

I think we're absolutely great and we have an awful lot of smart people and great people here. The whole world appreciates it, the Czechs are seen in America, the Czechs are seen in Western Europe. I think that in general the amount of companies that are doing it here is great and the quality is also great. Our state says that we want to move from the Skoda assembly plant to doing these new things like AI, so I think that the potential is really there and that it's not just wishful thinking. I think we could do it together.

Martin Hurych

It kind of goes to my head that it's against the system, not because of the system, that we're good at AI.

Jakub Krchák

I think the state is making a sincere effort. We see that, that the various tentacles of that state are trying to communicate with both the universities and the business. There's been a lot of discussion here about the AI Act, a lot of discussion during our presidency, and there's been a sincere effort by that state to get feedback and possibly modify that regulatory framework into a form that makes sense for all involved. Ivan Bartos is in America right now and he was just writing a post that they were discussing AI research with their American counterparts yesterday and he was writing about how in America it's much more on the applied research side. Over here it's much more on the side of that primary research. So what he himself admitted as a weakness of the Czech state, I think that is definitely a place where we need to move forward together, universities, state, business. Together we need to create things that are realistically competitive, that realistically add value to end users and companies. So I think that you can see that there is an effort and that the state and the people who represent the state have the desire to do those things and to move, to find directions to move forward.

How do you realistically get started with AI in your company?

Martin Hurych

Let's turn the page and get more practical. You've already mentioned small and medium-sized companies here. I, when I'm in these companies, there's not a person who doesn't say we need AI, we need to at least get some information and training. But at the same time, I see that many times they don't know how to go about it. There's some ChatGPT, there's some Copilot, there's some Midjourney, but most people don't even get there anymore. At the same time, a lot of times it's not very clear what it's possible to do with AI today, and because of that, there's kind of a first starting hurdle for those business owners that very few people climb over. Can we give a cookbook here, a guide for someone who would like to get ahead of the competition from the next town over, where to start to start building an AI-enabled business?

Jakub Krchák

I think that what you said, putting the tools that are available, like the GPT, Copilot and so on, into people's hands is definitely a good first step. Everybody needs to get a feel for it, they need to play around with it, and then it will also increase people's appetite, but also their effectiveness in how to use AI and how to work in tandem with that technology. In terms of how to do the transformation of that company, the business and business casy always comes first. What everyone needs to realize is where they have problems and where those things are burning them. AI is terribly interesting in that it can be applied to many different places within that company. I can apply it inside the back office, I can apply it in finance, I can apply it in procurement, I can apply it in HR, I can apply it on the customer care side. Every company is different and something different is bothering them. If I do a lot of hiring, I need to have a good, simple flow of people in, so I can deploy AI in that to simplify how I hire those people to automatically cover as many steps as possible. If I'm doing a lot of tendering, the sales people are sitting on top of it, reading long selector, the AI can very quickly read the selector, check if it makes sense to go in and provide only the most important information. It always starts with the problems, the money, the business crashes, and that AI can typically solve a lot of things.

One more thing I would mention, it's important not to take too big a bite at the beginning. Some of these things sound great, but the project might be half a year, a million, it might not be successful, so it's always better to start with smaller problems especially in terms of implementation and be successful with them. That will then show that the AI works and people have the appetite to go on to the next, more complex project.

Does it make sense to force people into AI?

Martin Hurych

You actually took away the question I had prepared. Does it make sense from a business owner position to even force some vintage or some non-tech enthusiasts to try and use AI?

Jakub Krchák

I'm sure it is. I recently gave a talk on how to deploy AI more successfully than other companies and I said that it's not that AI will replace humans. It's that the people who are good with AI will replace the people who aren't good with AI because they'll be more effective, they'll be more interesting hires for the business than the people who don't work with AI. It's a process that the population is now going to go through and they're going to get used to that AI. One lady logged on and said when do I think that's going to happen. I replied that it had already happened, so the lady jokingly replied that she didn't have to go to work the next day. But the point is that we all need to motivate ourselves and our people to make the best use of the technology in this new world and to be effective at the level of each individual person and at the level of the company.

What can AI realistically do in a small and medium-sized business today?

Martin Hurych

So for now, we have to motivate or maybe even a little bit force people to try out the basics on some ChatGPT, Copilot, Gemini, whatever is accessible in that company. So the other thing is to say, where would I try this out, where it's not quite as painful and I can see results relatively quickly. What next?

Jakub Krchák

I think what's always good is to kick-start the brain on what all that AI can do today. For example, we're doing some AI academy as well, and an important part of that AI academy is that we're showing people different use cases of all the different ways that AI can be used.

Martin Hurych

This is awesome, we definitely have to stay here. I see that there is often a lack of confidence in people that you can do this in a small and medium-sized company. Just as people don't believe that innovation is for the small, medium-sized company, exactly the same thing is happening now in artificial intelligence. So let's see what really realistically I can have today in a medium-sized company and what is not for those big corporations like banks, Skoda and so on. Let's pick 3 to 5 some use cases where we'll talk about how AI can be used by a mid-sized company, an AI studio, manufacturing, services, whatever you can think of. My bubble is IT companies, small, large, medium, manufacturing, tech companies.

Jakub Krchák

I think that any programmer who doesn't use AI today should be replaced again. Regardless of whether a person is a junior, medior, senior, it moves everyone forward and typically we see that the more skilled these people are, the more they can accelerate. Again, it starts calmly just on some GitHub Copilot, it's kind of the first simple thing, definitely enabling people to use these things. But then of course there's the issue of I don't want to push my code out there, I don't want anyone to see it and not trust it. What you can do nowadays is also have local models that help with things like test generation, documentation generation. It happens that those people have old code that nobody understands today. That AI can help understand that old code, suggest changes to it, and help someone new to the company or someone who hasn't seen those things get up to speed faster.

In the production area, it can certainly be things around automatic document processing. The most traditional thing that everybody has is invoices, there are a lot of specialist companies for that nowadays, so you can digitise completely those things around those invoices. What's also easy today is to sit some robot down to do emails. It sorts those emails into some categories, it can answer certain categories itself, other categories it can pass on to other people, which in turn reduces the need to manually do the same thing over and over again. There are very useful things in that production now that check quality, so it can be visually, I'll take a picture of something, see if it's well done. It can be sound-based, I'll make a model that checks whether or not the engine sounds good. It can also be based on some other parameters, I'll measure the temperature, put it in the machine. Those kinds of checks on those manufacturing processes are definitely something that can be implemented in those mid-sized companies today. I also talked about HR, typically in those manufacturing companies there are more people, so you can automatically help select the better people who are worth interviewing at all.

Martin Hurych

I guess you have to admit that there are ready-made apps for some things. Your domain is custom development, for my specific need, and that's where I more or less put the quality control or at least the finishing touches. Every product, very likely different requirements and I have to finish it in some way. Did we miss any other big domain?

Jakub Krchák

I think we missed a lot of domains. The depth of what can be done today is really great. One thing that is very crucial certainly in those manufacturing and engineering things is, for example, predictive maintenance, or the ability to tell in advance if a machine is going to break or not break. That way, there won't be downtime. Other things that are a little bit related to that are things like anomaly detection.

When will be the first AI trader?

Martin Hurych

I'm missing a theme. There's a lot of talk about automating at least in some segments where it makes sense and where you're buying a commodity, buying. When am I going to see the first AI salesperson?

Jakub Krchák

There is a question of what makes sense and what does not make sense. In the sales area, we may be doing some reevaluation. If we know that someone is interested in a loan, we can find out what parameters that person has, whether they meet the possibility of getting the loan. Based on that, it's then passed on to live people who still typically in that sales area have that added value and it makes a lot of sense. The other thing that maybe in that sales can do is help people schedule things in their calendars, verify that those people are going to show up for that appointment and things like that. The other important thing is to make people helpers, something that knows the products well and can help those who are selling, but also those who want to buy, to know which product fits that need. Because nobody can have everything in their head. It can work again in the area of those IT companies and production to be able to automatically recommend to people, for example, a certain type of product.

Martin Hurych

The domain of TELMA AI is chatbots, voicebots, you do automated customer success management and so on. So from what you're saying, I don't have to worry yet about an automated AI salesman calling me and selling me something without me seeing it.

Jakub Krchák

We can build it, but it's up to each person to decide if they want it that way or not. That's what I was saying at the beginning, that in general I think it works better when humans work with AI. Find where the AI is strong, where the humans are strong, put them together in the right way so that the human is

more efficient, to save some resources, make more money and make people happier. All of these things the AI can do and can help with.

How to choose a good AI application vendor?

Martin Hurych

A whole bunch of AI companies have popped up. My observation is that my generation, who last programmed something in basic in school, has at most a very amateurish insight into classical programming, but they're completely lost in AI. How do you pick the wheat from the chaff when it comes to vendors? If I decide to do any of this, how do I know who to bring into the company so that it's not a mess?

Jakub Krchák

In business, it's traditional to have some references and try something that those people can or can't do. Today, it's very important not to necessarily be tempted by some tinsel demo. A lot of these companies that are trying to feed off the AI hype, they can take the AI so far that it's a nice looking demo, but then when it's supposed to actually be deployed into that production, it's often not polished. Those companies don't understand what it takes to really make that product and there's a certain disillusionment and disappointment after a while when everything looks nice. So it's important to look at whether the company can actually do the production stuff and not just the pretty demos.

How can I ensure the security of my data?

Martin Hurych

We've touched a little bit on data security here, that's another thing that's being discussed a lot even now with Apple saying, a lot of things on device, I'll take care of everything for you, I won't let anything go anywhere. How do I make sure that my competitors don't learn from my data? We're still talking about these small and medium sized companies.

Jakub Krchák

It's very generally about some decision about how our company is going to go about it. There's a lot of talk about some platforms, what kind of approach one is going to take in general. I accept that it's not just the AI, but historically in that IT community, for example, the word cloud has worked and that's also been a painful journey for a lot of people. Do I want to go into it as a company, do I want to go into it as a company, does it have any benefits in terms of maybe

flexibility, in terms of speed. But then there may be disadvantages in terms of price or just security. When I do trust some of that cloud, it often works so that on that cloud platform, within that trust that I'm giving to that cloud, I'm able to do AI that will have a similar level of that trust. Now, of course, the question is to do it right, because if it goes wrong, of course, that data can escape somewhere, but as long as I have my trust in that cloud, I can stay at that level of trust. If I don't have that confidence in the cloud, then I need to have someone who is able to do the things that I need to do outside of the cloud.

How to safely have AI on your own data?

Martin Hurych

A lot of people say they'd be happy to drop that AI and those internets on everything they have there and make some sort of internal wiki, and only then start talking about what works for what and what the AI can get out of it. So I'm not afraid to go in that direction, it just has some parameters I guess I have to follow. Right?

Jakub Krchák

For us, that's one of the most successful things we're doing today because we can do that internal brain of that company, that AI on that data inside the company, not outside. So there's no need to worry that it's going to go somewhere because we move it to where the data is, we build a neural network that represents that knowledge base of that company and then you can do those use cases on top of that.

Martin Hurych

Do you use the big models that are there or do you build your own?

Jakub Krchák

This is such a middle model, because of course taking GPT 3.5 and fine-tuning it in-house is unthinkable. This is a specially selected model and we're running at high speed, so it was a different model a year ago than it was six months ago and than what it is today. It's a continually selected model that can take your technical documents, your contracts, get them into that neural network and allow you to ask all sorts of questions, like do I have a contract. It will also allow you to ask technical questions about those products as we've talked about and it's powerful enough to do that and small enough that it doesn't cost a lot of money to do it inside that company.

What to prepare if I am considering corporate AI?

Martin Hurych

We've now convinced the more progressive business owners that it's really worth it to beat the competition and be faster. I'm going to look for a vendor, ideally TELMA AI, what do I need to prepare so that the first interview isn't a collision of two worlds? What do I put down on paper to be ready for that very first discussion?

Jakub Krchák

What we need to prepare is ideally several of our problems that we have identified, we have problem 1, problem 2, problem 3. It's good to have more than one because even though problem one burns us more than problem two and more than problem three, maybe the AI will be able to solve problem three very quickly and very well. So it's good to have several problems ready like that. If those problems are represented by, for example, communication with the machine, then have it written down approximately what that communication looks like. I want to ask our technical documents this or I want to allow our customers to ask our technical documents that. That way you can tune that expectation to that particular definition of the problem.

Then what has already been said here is that every single thing is unlocked by some data. So if I want something that's going to recommend my products, but all I have is a website done in HTML, that's always going to be harder than having some content management system in there. That's where the descriptions of those products are made and that's what the web pages are built from at the same time and that's what the AI can learn from. So, it's always the case that you need to have some digitization, you need to have some order in that data and in that information.

Martin Hurych

The last thing I was thinking, I don't want to give you a price indication here because that's nonsense, we can range from tens of thousands to tens of millions here, but can you compare it to anything? Should I imagine that this is a second investment in some ERP or what should I actually expect?

Jakub Krchák

The thing that's most complicated about it is that when the AI runs, I have to pay for it. When I buy an ERP, I put money in, they're also cloud-based today, but historically I bought the software and I had that. Here necessarily the setup fee is not that big, it's manageable, but the more that AI is used after that, the more you have to pay every month because those neural networks are still running there. I think that's the most important thing to prepare for, the AI is paid for every month and the more it's used, the more it's paid for. It's pennies or dimes per use, but if I use it 10,000 times it costs something and if I use it 100,000 times it costs more.

Martin Hurych

Thank you so much for the insight into this world. I'm going to take away a ton of new information and go build an AI positive company. Thanks so much and may you prosper.

Jakub Krchák

Thanks again for the invitation, I hope the audience will also think about how they can implement AI and how it can help them be more efficient and competitive. Hi.

Martin Hurych

If Cuba and I have got you excited, and at this point you have a scrawled paper in front of you with ideas of where you could take AI, even though you may not yet know how, we've done our job well. In that case, I'll repeat my request from the beginning, subscribe me, cast me, maybe push this piece to someone in your area who might find it useful. You'll help me navigate the social networking algorithms a little more smoothly. Be sure to check out my website, www.martinhurych.com/zazeh, where you can find all the other episodes besides this one and the bonus. I can't help but keep my fingers crossed and wish you success, thanks.

(automatically transcribed by Beey.io , translated by DeepL.com, edited and shortened)