Estonia, the Baltic country of just 1.3 million people, is today recognised as one of the most digitally advanced nations in the world. It is known as one of the most tech-savvy countries on earth not only because of its four unicorns (Skype, Playtech, TransferWise, and Bolt), a start-up company valued at over US $1 billion, but also because of its government that has been actively implementing AI.
In this regard, Ott Velsberg, Government Chief Data Officer of Estonia, who oversees data governance and data science within the government system, was invited to speak to Almas Serikuly’s program entitled ‘LIFT’ (Law, Innovation, Finance, and Technology). The guest spoke about the use of AI in Estonia by public and private sectors, a hotly debated AI judge, and the ways of how less technologically developed countries may catch up with the developed ones.
Estonia is one of the few countries in the world, if not the only one, that has the Chief Data Officer position in the public sector. Would you please tell us more about your job responsibilities and challenges that you face?
- I started last year in August, and as you mentioned, I became the country’s first Chief Data Officer as well. My role includes taking responsibility for everything that has to do with data: from data governance to data science. Those two areas are quite broad in that sense. Data governance ranges from data quality to meta data, semantics and so on. Data science goes from a more practical, application side. If we talk about artificial intelligence (AI), then there is also simple analytics, education, research, and so on. The topics are extremely broad. Regarding your question about difficulties, the area itself is broad and maybe still a bit underdeveloped. So, you still have to work often on the basics, even in Estonia.
In one of your recent interviews, you said that the Estonian Government has an ambitious goal of having 50 AI use cases by 2020. Can you please mention some of them and give their brief description?
- Right now, we have 23 AI examples or practical solutions live in the government. We plan to have 50 AI solutions in the government by the end of the next year. For example, we are using machine learning to detect whether grassland has been mown or not. The idea is that we pay subsidies of roughly 200 million euros to farmers and we expect them to mow the farmland at least once a year. Previously we did on-spot analysis. So, we went to the farm and observed whether they had mowed the farm or not. Now we can do that from satellite images. Then we also detect what kind of trees are in the forest, again from satellite images and also from airborne LIDAR measurements. We also detect anomalies and incidents from x-road. X-road is our data exchange layer. We use video analysis to count the number of cars on different roads, and we also use different road dependent information to decide investments. So, where and how different investments affect the number of cars, waiting time and so on. We also use AI in the healthcare sector. It is to understand when different people should be brought into the hospital or not to understand their medical needs. This project was initiated and funded by the World bank. There are other simple things like chatbots. We also use transcription from voice to text; clean different documents from private information because it is extremely important in today’s world – we do not want private information to be public. We use analytics to understand where different points of interest are, for example, road signs. We also measure the quality of different streaming processes also to understand how users use different portals, for example, our national television. We also use machine learning from the security side to understand when different anomalies happen with people’s computers. There are many different examples of AI that we use in the government.
Do you have the approximate numbers of how many jobs can be automated?
- That is hard to say because what is important to understand is that you do not automate jobs, you automate parts of the work that is done. For example, if a person is right now doing A, B, and C, then we are actually automating only A. So, they can now concentrate on things that are not so easily to be automated. It is hard to predict how many jobs can be automated because even in the government itself we see that people now can do more fulfilling jobs.
How much does the Government invest in these projects?
- In AI as a whole (machine learning, data science), we roughly invest 10 million euros in the next few years. That also includes some things that are not so strictly common related to AI. For example, our Data Science Master’s degree programme that is now going to start next year but it is still related to the field from the perspective of creating that base competency. We also support scientific research and so on that is carried out by the government as well. So, the project money (research and so on) together is roughly 10 million euros for almost three years. Not much.
One of the projects that has recently become widely known is the development of the AI judge. Could you please tell us more about this project and the goals that you set before it?
- To be honest, there is not much to talk about right now. It is an ongoing project, and it was delayed due to funding reasons, but it is really about just fully automating the order for payment procedure. This is already half-automated. Payment procedures themselves are not typical civil court cases. It is not like people go to the judge and make findings and so on. It is more information stating how much was owed. And regarding that reference, final part the judge signs it off. If you are not happy with the decision, you can dispute it and so on.
How did the Estonian people react to this project (the AI judge) and what do the real judges think about this because the robot-judges may take over their jobs?
- As I said, it does not take anyone’s job because this is more than half-automated, and the rest of the job is done by a referent. There are people who help the judge. As I mentioned, you need to keep in mind that proceedings are evidence-free: there is no evidence to consider, and only the facts given are controlled. For example, whether the claim fits the set out the maximum financial limit. Because of that maybe in Estonia this project generated zero interest from the lawyers and legal community, interest was mostly from the rest of the world.
From your point of view, what are the main benefits and risks associated with the use of AI in the judiciary?
- In my opinion, many of the risks that people think about are overstated, to be honest. But at the same time, you need to keep that in mind as well when we take, for example, bias. Bias is something that we work on as part of data quality that is always an important thing regardless whether it is AI or analytics, but we often forget that there is bias in every decision we make: people themselves are not bias-free. Even data that we use already has bias. We have bias beforehand when we collect data. Bias is definitely an issue that we need to work with, but we need to understand that bias exists, nevertheless whether we use AI, machine learning or some other analytics. As a whole, data quality is a problem for this practical approach. This takes quite a lot of time.
In addition, we often do not really understand where different data is situated. From my perspective, it is about preparation, understanding what data we can use, how, and for what reasons. We should not use data for cases that are not allowed by the law, and we actually cannot do that. Part of data quality is being transparent: how we deploy these different tools. In Estonia, people can go to the government portal and see when the different ministries and different information systems have accessed their personal data and for what reasons. I think the heart of everything is being transparent, so people trust the government. People can always ask how and when the government has accessed their data and even take away the right from different participants to access their data, for instance from healthcare providers.
In Estonia, is the public sector ahead of business in terms of AI implementation, or do you know any successful examples from the private sector?
- There are many great companies from the private sector part. Before I go to the examples, I will just say that Estonia's public sector is ahead in that sense from the private sector. Globally Estonia stands out as a country where public sector strongly applies AI. From the private sector part, we have Cleveron, which is operating courier robots. Bolt, a competitor of Uber. Even the same satellite image detection solution that I mentioned earlier – SATIKAS that is actually offered by the private sector as well. Everything the government does is always done in collaboration with the private sector. We really do have many different exciting examples. Another example is Starship Technologies.
Estonia’s population is just over 1.3 million people, but it has 4 (Skype, Playtech, TransferWise, and Bolt) unicorns, a start-up company valued at over $1 billion. What, in your opinion, are the key ingredients for this success?
- I really believe that it is about the mindset. Estonia, as a whole, has entrepreneurship-like, the “let's try things out” mindset. The public sector always relies on the private sector; we have the mindset that the private sector should do things. And the public sector should only do when the private sector for some reason cannot do that, or it is just more beneficial for the public to do that. It is really about a mindset, and I think here Estonians are really kind of unique.
What can you recommend to the countries, which currently are not technologically advanced, in terms of the development and implementation of digital technology?
- That is really a hard question to answer without knowing the exact background of the city or the country itself, but I truly believe that the same model that worked for Estonia will not necessarily work for other countries because we started from the whole different starting point. For example, if you go to some countries in Africa, they may not have the legacy that we had back then or the legacy we have right now. They can start from scratch and might make a bigger leap. It is really about understanding about why you do what you do. When you go to AI, every project we do, we ask “what is the reason?”, “how do you measure the success of the project?”. I think that in every project we do whether it is AI or another IT investment, we need to ask why we do it and how we measure the success. So, there must be a reason.