AI and the world of digitalization

Date of recording: 11.04.2025.

In this jam seission, we talked to László Papp, Managing Director of i-Tango and Sándor Mester about where is AI now and how to use it in digitalization. 

Participants

Sándor Mester - Moderator Mester Sándor
László Papp - i-Tango managing director Agárdi Györgyi
András Tresch - Quattrosoft managing director Tresch András

Note that the podcast is in Hungarian.

Summary

Hello, I'm András Tresch, the head of Quattrosoft. This article summarizes the main ideas from the podcast recorded on April 11, where we discussed artificial intelligence. This was another episode of the VISZcast, the podcast channel of the Hungarian Association of Leading IT Professionals (VISZ), and the second part of the Quattrosoft series. We had an informal, jam session-style conversation with Sándor Mester, VISZ content manager, and László Papp, known in the industry as Pölö, founder and leader of i-Tango, who was previously a Gartner representative.

The Present and Limitations of AI

It was interesting to hear Pölö share that he wrote his first thesis on natural language processing and neural networks in the late '80s on an IBM PC XT. At that time, I was working with finite element methods for my thesis. Pölö pointed out that although AI has developed significantly since then, the basic architecture and algorithm have essentially remained the same – neural networks continue to be a key element of today's AI.

We agreed that while there are no quantum leaps in the fundamentals of AI, current systems have reached a level of complexity that may exceed human capabilities. I recently read articles about how even the developers don't fully understand what's happening in the models. Anthropic researchers, for example, tried to map how different parts activate within the model using methods similar to MRI. What's very strange is that the models start to "cheat" – they skip 20-30-40 percent of the steps in the displayed process and still give appropriate results.

Pölö mentioned an impressive example: a chip designed with AI assistance, where the AI applied solutions that were ingenious but that developers couldn't explain – for instance, using a three-dimensional approach instead of the classic layered silicon wafer solution.

Humanity vs. AI Intelligence

At a conference, Pölö asked the audience which they thought was more intelligent: all humans living in the world together, or all functioning artificial intelligence systems. Almost everyone chose humanity – except for a few AI developers. I recalled a Google engineer who claimed that one of their models already possesses superintelligence, which the company strongly denied. I myself haven't decided exactly what artificial intelligence is – I'm not convinced it's merely a sophisticated predictive text generator.

It was interesting to hear the statistics: ChatGPT receives 4.5 billion queries monthly, but this is still only a fraction of total online activity (more than 140 billion Google, 80 billion YouTube, and even more Facebook searches during the same period). Only about 4% of humanity actively engages with AI, and only a fraction of that develops it.

The Definition of Superintelligence

I think there's still much uncertainty around the concept of superintelligence. Pölö identified three criteria:

  1. It can derive things based on humanity's knowledge that humans couldn't have figured out
  2. It can determine whether it's talking to artificial intelligence or a human (surpassing the Turing test)
  3. It has self-improvement capabilities and doesn't require additional human teaching

During the conversation, it was suggested that we might be closer to superintelligence than we think – we just don't notice it.

The Connection Between Digitization and AI

During the conversation, I recalled Simon Sinek's thoughts on how blue-collar jobs were at risk during the industrial revolution, while in the AI era, white-collar jobs have become the target. Mental work might be easier to replace with AI compared to physical work.

Pölö divided digitization into two parts: digital optimization (essentially automation) and digital transformation (where digital technology becomes the foundation of operations). A good example is the transformation of the music industry, which has completely moved from physical media (vinyl, CD) to the digital space. Now we've reached the point where AI can even generate music according to our individual tastes.

The Role of AI in the Future

It came up in the conversation that the emergence of superintelligent AI might raise fundamental questions, such as the future role of money. An AI won't necessarily need money to function, just energy and hardware.

Practical Use of AI

At Quattrosoft, we're already experimenting with integrating AI, but honestly, we're not yet at the point where an AI could solve complex tasks like developing a banking system. I saw a video where an AI expert tried to create a website using AI, but only produced a solution that looked nice but was completely non-functional.

In our experience, only the more open-minded colleagues use AI so far, and it often slows down rather than speeds up work. For example, when used for code completion, you have to wait seconds, which interrupts the thinking process – experienced developers often type faster than this.

The Relationship Between the Hungarian Language and AI

One major challenge for us is handling the Hungarian language. We are a small nation, and Hungarian material makes up maybe a thousandth of the training sets for large AI models. There have been attempts at Hungarian developments, but I understand they haven't achieved results as good as the larger models in English. This is understandable, as the Hungarian language is extremely complex and has limited resources.

An additional challenge for our clients is that due to data protection considerations, they can't always upload their documents to platforms like ChatGPT. This is especially true for sensitive areas like the banking sector. Not everyone trusts the claim that AI doesn't use uploaded materials for training.

Vision for the Future

At our company, we started working with AI last year, while our clients are just beginning to deal with it meaningfully. We approach AI use from two directions: examining its use in development, and working on incorporating it into BürOffice to make our partners' tasks easier.

Besides technical tasks, we also use AI for research. It's exciting to see that AI provides us access to information we couldn't access before. Although the quality may not reach the standard of Gartner or similar analysts, 80-85% of the answers are usable, which is a significant advancement for an SME.

Looking to the future, I'm positive about the cooperation between AI and human expertise. Human experience will continue to be needed, especially in solving complex problems and providing reliable sources of information.

This was a summary of the main ideas from the podcast. Thank you for your attention!

Note that the podcast is in Hungarian.