"The artificial intelligence team started some seven years ago. We talked about machine learning rather than AI and worked on classical mathematical models. Back then it was more of a fad, there were two of us and it was more about exploring possibilities than a progressive topic," says Hynek Cihlář, who is now one of the leaders of the team called VPT 4.
Today, he is working in a completely different era, with a completely different perspective and different opportunities.
Huge opportunities.
It was clear from the beginning that the use of AI could greatly help companies to be more efficient and prosperous. The first project looked at how to make it cheaper for a cosmetics company to produce liners using advanced techniques that tracked what was happening in each step of production. Today, the possibilities are quite different. At the proof-of-concept stage, the team has so-called multi-agent systems that allow people in companies or perhaps government organisations to ask analytical questions over their own data and get an answer within minutes. All this without having to program anything, create reports or complex excel spreadsheets.
"The idea is to leave the routine work to the machines and give people the space to think and find new ways to take the company or organisation in a new direction," explains Hynek Cihlář the essence of what could be called in short a local ChatGPT for companies. The reason we talk about multi-agent systems is that they have a number of small "workers", called agents, who carry out specific sub-activities, and then others who distil the results of their work. These agents are linked into a kind of chain in which they pass data between each other. In practice, this can be well illustrated by an example from a court or a city council: one agent would listen to what is going on in the room, another would convert the recording into a textual record, another would clean up the text and perhaps take out the vulgarisms or filler words and consolidate the spoken word into coherent sentences, another agent would write out tasks from the text and publish different material on the web portal or digital noticeboard of the authority.
Where to get people?
For all this to become a reality, a number of people are needed at the outset to prepare and clean the data, while keeping an open mind to see the potential in the connections between the data and the mathematical algorithms that can be used to process the data. That's why the Aricoma team has teamed up with the University of Technology, from where it recruits specific types of colleagues.
"We knew right away that our selection of people for the team had to be rigorous. We need people who want to learn, and who also understand advanced mathematics, not just traditional IT people. So we thought about how to get them. The solution turned out to be to strengthen cooperation with the BUT," explains Hynek Cihlář, adding that he has recently managed to find four colleagues among the students at the university.
Aricoma has been introducing itself to students for years, participating in trade fairs and presentations. However, the cooperation with the BUT has already taken many other forms. Thus, the VPT 4 team includes Anna Derevianko, who also teaches at the faculties of mechanical engineering, civil engineering and chemistry. Other Aricoma experts lecture externally at the faculty, they organised a summer school together with the faculty in the summer, and the topic of local language models and multi-agent systems has long been worked on in university research.
If all goes well, the current project, and the result of the work of nearly 20 people could go to companies for testing early next year. In the meantime, however, there is another growing theme of quantum computing and its potential impact on the development of artificial intelligence. It's already clear that if the two could be linked, the computational effort to build neural networks would be reduced, leading to more refined results and the development of other deeper AI capabilities. However, not only according to Hynek Cihlář, this is still a matter of at least a decade away.