Interview with CEO: "Pace and volume are key in AI revolution"

Our CEO Aistis Raudys recently gave an interview with the leading Lithuanian business and politics journal “IQ”, an official partner of “The Economist”, and discussed the progress of the AI revolution, AAI Labs’ expertise and regulatory challenges.

Journalist: How does your company contribute to the overall advancement of science and technology in the AI and ML field?

AAI Labs is an applied AI and ML company — we develop AI-based solutions for both private and public sector customers that search for ways to optimise their business processes. In most cases, there are no off-the-shelf solutions available and AI models have to be trained on specific data tailored to the individual needs.

For example, we recently completed a project on models for smart forests, where we developed a system that calculates lumber logs from photos. We have also used satellite images to predict the type of forest cover, the number of trees and so on. Another example of AI applications is our ongoing cooperation with JUDU, the public transport operator in Vilnius, where we are looking for ways to predict passenger flows several hours in advance and based on this data select the type and size of vehicles accordingly.

Journalist: More than half a year after the launch of ChatGPT, how can we assess its impact on the labour market?

I think the welcoming of ChatGPT in the labour market can be divided into several phases. First, there was just a lot of surprise amongst the companies and a bit of scepticism about the development of the very technology. After witnessing the exponential growth in users, the conversation about GPT became omnipresent. Yet, it was only after a few months that managers realised how much potential the application of this technology had in reality. And while everyone knows and talks about GPT today, few have a clear vision of where the technology can be applied in concrete terms.

We present a wide range of possibilities to our clients — from analysing the national legal frameworks and updating documents when laws change, to assessing employees' progress, resolving legal disputes and more. While many are already producing documents with the help of ChatGPT, this is only a small part of the possibilities that can be unlocked by tailoring the language model to the specific needs of the user.

So, I would argue that the impact of ChatGPT on the labour market is somewhat limited - few companies have so far been able to take advantage of the capabilities of the language model beyond chatting on a website developed by OpenAI, although the integration possibilities are truly endless.

Journalist: As the number of AI tools grows, which professions will be most in demand in the future, and which ones will decline or disappear?

There are two sides to this issue. It is clear that lower-skilled workers will no longer be needed. For example, in machine learning, LLMs are already able to carry out some of the tasks of our Junior staff — simple and boring tasks that used to be done by our staff can now be done much faster and usually even better. On the other hand, most in demand in the labour market will be those who can use the very GPT or other major language models most effectively. Employees who can do proper query engineering will soon be snapped up by businesses that are waking up to the benefits of AI.

Journalist: One expert I interviewed said that AI can only be considered advanced if it is universally accessible, i.e. speaks in a language that everyone can understand. In your opinion, when will we see AI fluently speaking in Lithuanian?

But AI already speaks Lithuanian, and fluently! Speech synthesis is one of AAI Labs' main activities. We have created about 10 different AI voices (female, male, children) that are fluent in Lithuanian and have correct contextual accents. Audiobooks recorded in these voices are no different from those recorded by professional actors.

We are currently synthesising audiobooks for various publishers in the country, collaborating with the Lithuanian Audiosensory Library, and carrying out projects that will allow AI to speak fluently in other low-resource languages such as Croatian, Czech, Polish, etc.

We also use Lithuanian with LLMs, which understand it as well as English and can perform the same tasks. In essence, the AI revolution no longer requires someone to translate for us, as has been the case until now. We look forward to the translation of films, books, stickers on imported goods — it's natural to imagine that AI solutions will also require translation. But this is one of the main reasons for the speed and ubiquity of the AI revolution - these solutions speak multiple languages at once, and the AI teams then just ensure that the output of the information, both verbal and written, is no different from the way we humans speak and act.

Journalist: How can AI help make data analysis more effective in business? To what extent can indicators be predicted and forecasted using AI and ML?

AI can certainly help squeeze a few extra percentages of efficiency out of a business. For big companies, that means big money. The level of foresight depends on the area of the problem at hand. If it is an area prone to randomness (e.g. financial markets), prediction is difficult. But if it is a field governed by clearly defined processes, forecasting can be quite accurate. AI can also do an excellent job of first-line basic analytics and reviewing recent trends: what is leading, lagging, and what looks like an anomaly, and presenting it all in a clear summary, suggesting an action plan or a set of conclusions.

Journalist: AI products are under active development and we cannot yet predict their ultimate benefits and risks. It is said that this uncertainty makes it very difficult to regulate new technologies through traditional laws and regulations. In your opinion, should AI be regulated, and if so, what should it look like?

The best time to regulate AI will be when we have specific and recurring problems. If we try to anticipate problems and pre-emptively ban everything, it will hinder progress.

The greatest innovations have come when people have been free to take and do whatever they want. In our field, one of the better examples of this is the Stable Diffusion models that generate images — once they were made public and open source to all programmers, they started to improve rapidly. An older example is the Internet itself — a number of organisations agreed on common protocols and from then on things started to evolve very quickly, with everyone contributing to the best of their ability and expanding the Internet network step by step. If it had been necessary to get permission from governments right away, nobody would have done anything.

In principle, the regulation only benefits large companies (e.g. Google, Meta) that have both resources and the capacity to deviate. However, it reduces competition — small businesses cannot afford to take risks, break the law and then pay fines. This asymmetry is crucial to take into account in the debate on the supervision and regulation of AI technologies in Europe.

Journalist: Is there a chance that EU regulation will inhibit European AI developers and make them less competitive outside the EU?

Well, it depends on our priorities. Do we just want AI to grow, or do we want AI to grow responsibly without endangering our society? EU regulation is not in itself an evil, because it creates a level playing field for all developers and users, which is fair, democratic and safe.

However, perhaps the most worrying thing would be excessive restrictions on the collection and processing of the data itself, which is the cornerstone of the development of AI-based solutions. The EU wants to be the first to clearly define guidelines for the development and use of such solutions. However, if this ambition leads to the creation of separate positions for employees to ensure that data inputs comply with the hundreds of points specified in the legislation, it will severely slow down and complicate model development and training processes.

Any strategic competitor in AI, be it China or the US, will undoubtedly take advantage of this to create their own rules for the AI game. Speed and quantity are key in the AI revolution. In the autumn debate on AI in the EU institutions, we will see to what extent these two priorities will be protected from over-regulation in Europe.

The full interview, dated August 16, 2023, was published in “IQ” and can be accessed here: https://iq.lt/verslas/a-raudys-di-revoliucijoje-svarbiausia-greitis-ir-kiekis/299156 (in Lithuanian language).