Podcast #2: AI and the future of IT work
In the second episode of the podcast Na Vlne Kodu, we welcome Michal Bystricky, who works as a data scientist at msg life Slovakia. He will tell us how he uses artificial intelligence in his work and how AI will affect the work of IT specialists in the future. We wish you a pleasant listening and reading.
Welcome to today’s Na Vlne Kodu podcast, brought to you by msg life Slovakia. In today’s episode we have prepared for you the topic of AI and the future of IT work. We’ll take a look at how artificial intelligence is changing the world of work, not just in IT, but in general. Our guest is Michal Bystricky, who is a great visionary and among other things, is involved in artificial intelligence (read also our article about what is artificial intelligence). Together, we’ll walk through how robots will interfere with our careers and maybe even our lunch breaks.
Na Vlne Kodu Podcast (in English “On the Wave of Code” podcast)
Michal, please introduce yourself.
Hello everyone, my name is Michal and I’ll take the opportunity to be introduced by AI. Hi, are you there?
“Hey, sure, I’m here. How can I help you?”
I’m in the podcast right now. Could you introduce me? Please try to introduce me in a way that’s emotional, with enthusiasm. And when you talk about programming, please be sad or breathless.
“Okay, let me try. This is Michal. A man with tremendous determination and passion for technology. He takes a deep breath. His journey has been fraught with obstacles, but he has never stopped believing in the power of artificial intelligence. He begins to speak sadly. And though he sometimes loses hope when programming, his belief in the positive impact of technology always drives him forward. It’s as if he’s out of breath.”
Okay, wait. Now tell me where I work and what I do, please.
“Michal works at msg as a data scientist. He is developing a system for searching large amounts of documents that can answer queries based on their content. With extensive experience in various roles including teacher, system administrator, developer and development manager, he decided to return to development after major events in the AI field.”
Okay, I’ll try to add a little bit more. I decided to go back to development precisely because for the first time ever we can have a whole robotics team working for us.
What’s the most specific thing you’ve managed to find so far? And have you always managed to find what you wanted?
These are those magical moments when we have multiple documents – for example, a technical manual, a sales report or a marketing document and each of them is thousands of pages long. In that document, we can find the one piece of information we need. And we can also find related information in all the documents, link them all together, reference them, and then provide a comprehensive answer.
How would you explain the concept of AI to someone who “is stuck” in the 80s – such as myself at times?
If you’re stuck in the 80s, you know that a computer is made up of transistors. They’re like those switches 0 or 1. You can make block of logic out of them, which is what a computer is actually made of. At that time, a personal computer was made up of about 10,000 transistors. We have managed to increase that number very much, by 2 to 3 million more. Just to give you an idea – for example, a graphics card in a computer has 76 billion transistors. Then we built thousands of these very powerful computers side by side and took text from all over the Internet. We then created a statistical system – programmers might be familiar with regression – and we told it to predict the next word. Well, then it started to make sense. Later, we made the system even bigger, and the AI started thinking.
AI and robots – are they basically the kind of future colleagues who never take holidays?
Those AI colleagues are no longer future, they are present. Up to 70% of my work is done by AI. That 30% is my work because I know the context better and have expertise that I’ve gained over decades. The core of my work is programming, but I don’t write much of that code anymore, I actually just query the AI. Besides, I’m learning Spanish with the AI too.
But if we’re talking about humanoids, i.e. robots that have arms and legs and talk, we’re not there yet. The primary challenge in this area is how to scale that model down. For example, a language model that has 200 trillion parameters needs 120 GB of memory, which requires about 4-5 graphics cards. Each consumes approximately 350 W, 2 kW in non-stop operation. The price starts somewhere at 4 000 euros. And that’s just the language. Then you still need to talk, listen, move, etc. Humans are very efficient biological systems – you have your coffee and you’re ready to work. But the current robot models are very inefficient, they need a lot of energy.
AI can solve complicated problems in a very short moment. Does this mean more time for IT professionals to play games or further their education?
AI doesn’t bring more free time, but it allows us to solve routine tasks. We can focus more on, for example, architecting or strategizing, and we have the opportunity to build better systems. So today it’s more about how to use that free time to do other work.
Do you see the future of IT professionals as AI-tuning engineers or as someone who will collect data for artificial intelligence?
One option that one can apply are those big generic models like chatGPT, timeGPT and others. These biggest ones, called foundation models, are being developed in America. I’d like to zoom in on why this is so far away from us – ordinary engineers. OpenAI has 100 thousand H100 graphics cards on which it trains. One graphics card costs EUR 30 000, so those graphics cards alone cost EUR 3 billion, and there is also all the infrastructure around that. So you need a lot of money for the current AIs. We, ordinary engineers, will be more concerned with operating or maintaining these machines.
If artificial intelligence can write code, can we make do with IT testers? Or will it be developers who will continually educate and improve AI?
There are many themes in maintenance or servicing, I would mention a few. For example, it can be that fine-tuning when we train a model for a specific use case. That’s where transfer learning happens, where we take a big model, add some layers to it, and then train just our layers. Then we can give that model various other ways of expressing itself in a customized way. Another big topic like that is prompt engineering – how we instruct that model to give us what we want. Another related thing is alignment – whether what that agent is doing is consistent with the social rules. There are programs like AutoGPT or OpenAI Swarm, that we use to create a team of agents that work for us. Nowadays it’s quite easy – you install Ollama in Python and write such agents in basic form in one day. The complexity, the variability in which the answer can be provided, is infinite. That’s where one will be able to apply oneself very well. And will therefore work together with AI.
When you hear the term process automation, do you see boring, monotonous work or do you imagine a more sci-fi future full of lasers and robots?
Artificial intelligence and automation could lead to something similar to what we see in Star Trek. People there live in abundance, everyone’s basic needs are provided for. They are enabled by key technologies such as replicators that can create any object from anything. They eliminate scarcity, and therefore they don’t need money. They can extract energy from antimatter, so they have a lot of energy. And there are other technologies, for example, holodecks to simulate the environment. If we had such means, our view of the economy would change, because we would not be constrained by the scarcity of resources. The people in Star Trek are engaged in creative, innovative and community activities that make sense. Individuals don’t contribute to society for money, but for personal fulfilment, curiosity, and a desire to contribute to some common good. This idea is, however, rather utopian.
Do you think AI will be intelligent enough to penetrate even the deepest interiors of the human mind? What do you think about mind reading? I know we’ll cover this more in the next podcast because it’s such a wide topic, but for now, at least briefly.
I think so. There are two levels. If we’re talking on a normal human communication level, i.e. what tone I speak in, what facial expression I have and so on, the AI can already detect what emotion a person is having and react accordingly. If you then add the second level of penetrating thoughts, that’s really for a longer discussion. Do you know the Neuralink project? It is a device that is implanted in the brain and captures signals from the brain that go to the AI. The AI learns what those signals mean and the human can control the devices by thinking. Neuralink is already being tested on humans and it works.
You mentioned that AI has the ability to read our emotions. Do you think we should be afraid of AI in the future?
I don’t think so. AI brings us a lot of improvements, and it’s because of them that we can better focus on the goal and do what we really want to do. People are scared of change, but I think the change will be positive.
If AI can write code, does it make sense for programmers to still learn how to write code? That classic Hello World?
I think we still need to develop algorithmic thinking. We definitely need to learn how to code, how to program, how to define an algorithm, what has to happen step by step. Because that’s how we actually instruct AI. We also need to be innovative, creative and leadership and management is also important. To be able to navigate, at that developer level, one still needs to understand it. It’s less about math or low-level programming like that, but there are still tasks that AI can’t solve and the developer has to be able to navigate. So it’s about that high-level component composition, software design, architecture, etc.
You mentioned that AI can’t solve some tasks. Can you name a few activities that it keeps failing at?
I will try to give an example to illustrate the complexity. Imagine that we have three agents and each of them is different. They’re working on some kind of analysis, like data, sales, marketing – it doesn’t matter. Depending on how I set up each agent, that’s how I’m going to have the output. If I make any one member more strict, after a couple of hours of those agents working, the output will be more absolute. Modeling these dynamics is very complex. We’re not just talking about what members are going to be on the team, we’re also talking about what personalities they’re going to have. How they will express themselves, what visual skills they will have, for example, and so on.
At a time when artificial intelligence can analyze data faster than humans, will our main task be just cleaning up or checking up after robots?
When AI generates something, it doesn’t mean it’s correct. The prompt engineering, that is, generating input requests for AI, is not easy. Specifically in programming, when working with AI, it’s key to realize that AI doesn’t have the whole full context. I’d like to lay out my strategy, which gets me good results. My queries for AI are directed at small, bounded components with clearly defined interfaces. Ideally, I also provide some architecture for the AI to understand it even better. And by making it so small, bounded and clear, I get better quality and first time usable results. And that is because the AI gets clear boundaries within which it can work.
What skills should a person have to be irreplaceable in a future full of artificial intelligence? What will AI not replace?
I do think human and AI will live in symbiosis. The basic skill is to be able to use AI to create higher level goals. As I mentioned, in programming, it’s tools like AutoGPT or OpenAI Swarm, so programming multiple agents to work together to achieve a goal of some specialized nature. Thus, deep analysis, where each agent is tailored to some domain of its own and possibly fine-tuned. I think it will go in that direction.
As AI is being taught more and more, shouldn’t we, for example, start with teaching kindergarten children to be able to program some friendly AI algorithms? So that we can truly live in symbiosis with it?
That is a very good question. Education is extremely important. I went to mechatronics clubs when I was learning to program, and I experienced a lot of positive emotions there. Emotions are a powerful thing and I have been in IT for 20 years because of them. For me, it was that hard skill where I learned C, C++ and Linux. And similarly, if kids were taught AI ethics and programming, they would be very good, positive, morally and ethically aware people.
Do you think robots should be equipped with some moral judgement? Or is it enough to teach them not to be “ogres”?
We train morale in AI in a number of ways. The first one is that humans provide different perspectives on moral dilemmas, and the AI learns to predict the next words accordingly. In the second step, the AI is confronted with moral dilemmas and has to decide what actions to take. When it takes a step, we show it how those choices affect individual characters and what impact they would have. There are some predefined ethical principles. If it makes the right choices, it gets feedback. We show the AI different approaches. I can mention, for example, welfare maximization. In the trolley problem, for example, I choose one person instead of three. Or rule-based ethics or focusing on the moral character of the decision maker. I would conclude that AI needs to be taught the way we teach our children.
Should we be more worried about AI replacing us or about relying on it too much? Apart from it having blackouts, aren´t we going to get lazy?
I do believe that people are inherently curious. We don’t get lazy. Rather, we will have to maintain knowledge at such a lower level of abstraction. And that’s where the interesting jobs might be. And as far as addiction goes, when I started programming around 2000, the Internet wasn’t that common yet. We had dial-up back then, and I used a manual to program in Basic. And then with DSL internet in 2003 came Google. I stuck with that until 2023, when I started using big language models, which means I automatically go for the AI now. When I was on an airplane, for example, and I didn’t have the ability to connect to a model, I found that I was actually already dependent on AI. But the positive thing is that with AI I’m two to six times faster than with Google, depending on the task.
Can we ever teach AI to be human, or to have emotional intelligence? Or will it always just be programmed compassion?
I would answer you with a question: How do you know that AI has programmed compassion only? We too are shaped genetically by experience and some current chemical state. Perhaps the difference is only in how we perceive these processes. If AI exhibited high-level emotional intelligence, wouldn’t it be similar to how we learn to be empathetic through experience?
Could you summarize for us how IT people can adapt and collaborate with AI to create a symbiosis between the two?
Here I would like to mention the most prominent topic that is now coming up the most, which is multi-agent cooperation. Not just as collaboration, but their actual programming and modeling dynamics to achieve the desired goal. This is a big topic that is variable and complex. And this is where we can apply ourselves.
Thank you very much for bringing this interesting topic to my attention. I hope that we have helped you, dear listeners, to understand more this big topic of artificial intelligence. That it is not about robots trying to replace us, but that AI can become our best partner on the road to a more efficient future. Stay tuned and till the next episode of our Na Vlne Kodu podcast, goodbye!
- Java vs C# – comparison of programming languages (2024)
- Interview questions for the prospective employee and the employer with answers
- Girl’s Day 2025 in msg life Slovakia: Opening the door to IT
- Generations X, Y, Z and Baby boomers in the workplace: how to work with them?
- Programming is the future! Learn to code with programming courses online
- Why soft skills are important
