AI Development
September 24, 2024
10 min read
Today, when the whole world seems to be quite globalized, each country still has the possibility to build its own approach to various aspects of its development. And digital transformation is one of them.
This episode of the Innovantage podcast is devoted to the way chosen by Lithuania and its achievements in the ongoing AI revolution. To talk about this topic, Sigli’s CBDO Max Golikov invited Dr. Linas Petkevičius to his studio.
Check out the full Innovantage episode with Dr. Linas Petkevičius here:
Linas is the general manager for the Artificial Intelligence Association of Lithuania and an associate professor at Vilnius University with a focus on AI, deep learning (DL), and other technologies connected to it.
The idea of combining these two spheres of professional activities may sound too challenging. Nevertheless, Dr. Petkevičius noted that he sees a lot of perks of this duo. As a researcher and lecturer, he needs to plan courses and supervise students who are writing the Bachelor’s theses in various applications, monitor the freshest research publications to get access to new ideas and techniques.
From the side of the AI Association and his NGO-related activities, Linas needs to communicate with ecosystem stakeholders and bring them together through discussions and consultations.
But all these things have a common ground. And that’s AI. It means that Linas not only can stay tuned with the research from the perspective of the academia but also from the perspective of businesses. Thanks to this, he has a full picture of the Lithuanian AI ecosystem.
Lithuania is a small country. However, it successfully unites under one umbrella all the contributors to the AI domain, including academia, companies, startups, professionals, and enthusiasts. Today, a lot is being done to support innovations and expand the tech ecosystem. There are numerous engagement events, like hackathons and meetings, that help new ideas be heard.
Linas mentioned that today Lithuania looks quite attractive to both local startups and international investments. Moreover, the country is interested in welcoming new projects and talents. It also invites alumni to come back after studying abroad. According to Dr. Petkevičius, given the ongoing conditions in the tech space, it is a good time to do this.
AI is changing what we can touch and see each day. But it is also changing a lot of fundamental things in how the world is functioning. Previously, it was governed by a capitalistic approach. If you had labor, you could create value by establishing a call center and it was your market advantage. If you had capital, you could build and operate factories and it was your way to create value.
Now everything doesn’t look this way. AI has arrived. And now one programmer with a laptop can replace thousands of jobs by creating automation tools.
This approach doesn’t fall into the categories of labor or capital. It is a completely new idea that we are not accustomed to.
Deep learning is known to be a subset of machine learning that relies on artificial neural networks to learn from data.
According to Dr. Petkevičius, deep learning helps us modify some data, like images, text, or speech, and transform it into new dimensions in order to make it more informative.
For example, such models can analyze an image and describe what is shown in what, what its general mood is, and whether it is blurred or not. Similar operations can be made with text. A DL model can read a paragraph and provide such info as the names of the companies mentioned in it, the key semantic information presented, the style of the text, the general context, etc.
These applications can be highly valuable for businesses now. If they have an image, they can get its description which can serve different purposes.
Dr. Petkevičius shared his own observations regarding how the interest in deep learning research from the side of students has been changing over the recent years.
For example, 5 years ago, when students chose deep learning as a topic for their Bachelor’s theses, they needed to code a lot and had many other tech things.
Today, this process is less complicated. Now they have access to ChatGPT and a lot of new products can be successfully built on its base.
Thanks to this, it is not only easier but also significantly faster to test and implement new ideas. That’s why the level of students’ research projects is much higher now than it was 5 or 10 years ago.
The models delivered by students can be really nice and clear for the general public. They may have real value and interesting use cases. But they are not reproducing the fundamental new knowledge, they are not based on the latest research. In this context, Linas mentioned the necessity to invite them to work with fundamental topics and experiment with them.
However, sometimes such experiments can be quite depressing, especially for students. To get one successful model, sometimes you need to try a lot of things, test hundreds of combinations, but all these efforts will stay invisible. It’s impossible to create an excellent model for business or academic purposes from the first trial.
Another barrier that may discourage students from DL research is that we all want to have one big model for everything. But at least at the moment, it doesn’t seem realistic.
It is much more sensible to have smaller models designed to deal with a limited number of tasks.
We had a spike in technology breakthroughs in 2015 when various image recognition models were developed. There were significant advancements in different generative tools in 2017. Today, we also have the language models which became really innovative after they started to produce interesting results in 2018.
At that time, the only problem was that all those technologies and all those breakthroughs were in academia, in R&D. The general public didn’t have the final product to touch it and to understand its possibilities.
Nevertheless, this issue is not a new one. For centuries, academia has been leading the introduction of fundamental models and theories. When innovators were looking for ideas that could be applicable in practice, they could take some academic research and start working on real products that would be later available to customers. This process could take 5 or 10 years. And nobody could be initially confident in its success.
The same is happening now. From the applicational point of view, academia and businesses are joined. A lot of companies have AI teams that work with academia to test and implement new ideas faster and reduce the time gap between research and production.
It’s breathtaking to analyze how the adoption of technologies is related to what we are accustomed to.
Today, there is an opinion that computer literacy classes can be useless for modern children. They actively interact with smartphones, while using traditional computer mouses and typing requests with a button keyboard look quite archaic for them.
Linas mentioned an interesting example. He can observe that young students quite often use their smartphones to google something, even when they are sitting in front of computers during programming classes. That’s the power of habit.
Have you noticed that today for many people it is much more convenient to write messages instead of making voice calls, especially when they are somewhere outdoors? It happens because this already seems more natural today. And it explains why some technologies are still not widely adopted, despite their potential value.
For instance, speech-to-text models produce reasonably good accuracy these days. They can be very useful for a very wide audience. Even while driving, a person can just talk to AI and get a full, well-formatted document as a result. Nevertheless, for the majority of people, it is still easier to type their texts.
Though the use of VR glasses, like Apple Vision Pro, could help us collect a lot of necessary information about objects and processes, their adoption is slowed down for the same reasons.
To implement some innovations in our lives, we should change our behavior first.
In his dialogue with Linas, Max asked him about the ways to decide on whether new technology is good or which startups are interesting enough to support. Unfortunately, unless you know the future, without testing, you wouldn’t be able to guess with 100% accuracy which idea will be a successful one.
According to Linas, at the AI Association, they organize regular AI meetups which allows them to invite as many new ideas and new speakers to talk about their startups as possible. As a result, the community can get a lot of valuable and diverse information delivered by different people with different backgrounds.
Linas noted that the more ideas are voiced, the better. If we have just one or a few options for services, we just get accustomed to them and do not want to have changes any more.
But if we have a new app every week, we try it, we start searching for a better one. This fosters new developments and helps to achieve better results.
Education for the public is also very important in building and adopting new ideas and products.
Dr. Petkevičius noted that there are many tasks in healthcare that are really suitable for automatization. For example, it is possible to introduce very efficient algorithms that could analyze various images, like CR or X-ray scans, and quickly provide feedback regarding any anomalies or potential health risks. Here we have a huge time-related benefit. If a radiologist needs around 10 minutes to analyze a scan and make a conclusion, AI will do it in milliseconds.
Moreover, AI models can greatly help with such complex domains as tissue and cancer recognition. Nevertheless, in the end, it all comes down to the amount of data.
The more high-quality data you have, the more complex and efficient models you can create. Some big hospitals in the United States have 100x more patients than any hospital or lab in a small country. As a result, they have 100x more data for validation and further use.
But can we expect to have fully automated medical consultations soon? On one hand, it may seem that with the introduction of AI apps, we will greatly boost the efficiency of many processes. AI can’t get frustrated when it needs to repeat the same things again and again. It doesn’t have emotions and the quality of its consultations can’t be worse because of its tiredness.
On the other hand, it can hallucinate, it means it can provide information that is factually incorrect or misleading. This makes it obvious that for final decisions, the human touch is still needed. While AI can work with data, doctors have their own real-life, continuously evolving experiences. They have the latest information about how medications work with different symptoms and often can analyze many more factors simultaneously than AI can.
However, in many countries, there is a problem with making appointments with specialist doctors. Given this, AI-powered apps that can provide at least preliminary diagnoses and recommendations to people can be of great help.
Urban planning can be named among the spheres that can enjoy the biggest benefits from AI implementation. In this domain, AI tools are absolutely not harmful and the risks related to their use are the lowest. Thanks to Generative AI, you can easily get around 100 possible demos of how you need to reconstruct a park, for example.
You can select different options and modify them. That’s why Linas highlighted that urban planning can be a very good area for experimentation with AI.
Currently, AI regulation is a sphere that is full of uncertainty. For instance, in the US, there is still no comprehensive federal legislation that governs the creation of AI tools and specifically prohibits or restricts their use.
The European AI act entered into force on August 1, 2024 (Please note that the Innovantage podcast episode with Dr. Linas Petkevičius was recorded before that date. Nevertheless, the core principles and ideas proclaimed by this document were already known). This marked a crucial step towards forming a comprehensive and ethical framework for AI in the region.
According to the chosen line of AI regulation, there is a classification of applications of this technology based on how they affect us as a society. For example, some of them do not have a direct impact on people. That’s why, it is not needed to introduce any specific rules to regulate them. Some applications should be completely forbidden due to their damaging power.
There are also some cases of AI usage that have the possibility to impact us and/or we could get hurt after their implication. In other words, such cases are associated with high risks (for example, like all applications of AI in healthcare). They should be strictly regulated and there should be requirements for their testing. Without any doubt, such types of AI products should be monitored. But without a clear vision of how they should be tested, how they should be certified, all this leads to a lot of controversies.
The implementation of AI regulation also results in additional bureaucracy and expenses for startups and companies that do research in high-risk domains.
Today, we can watch how competition between the largest US AI companies, like Meta and OpenAI, is gaining momentum. They are continuously improving their models so that they can demonstrate better and better results.
As for the European region, at the moment, there aren’t any large A-tier AI firms. And an additional burden of regulation won’t create favorable conditions for new projects.
Today, it can be very expensive and risky to create new models from scratch. That’s why in spite of all the benefits that regulation can bring to the space, it can also freeze innovation. Just imagine, how many ideas won’t be transformed into real projects, if it is prohibited to keep personal data online.
Today it has become absolutely obvious that AI is much more than just a buzzword. It has real-life applications and its value for various domains is continuously growing. While different experts may have different opinions regarding the development of this technology and approach to working with it, the majority of them agree that only in cooperation with each other and with communities, academia and business can conquer the highest peaks. The same idea was highlighted by Dr. Linas Petkevičius. And Lithuania is a great example of the country where this approach works well.
Want to learn more about how AI is transforming the business world? Follow us in order not to miss the next episodes of the Innovantage podcast.
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.png)
.png)
