According to the 2019 Global AI Talent Report, there is the same number of talents in Poland and in the Czech Republic. Only that the Czech Republic is several times smaller than our country, says professor Piotr Sankowski, Institute of Informatics, University of Warsaw, in conversation with Maciej Chojnowski
Maciej Chojnowski: You have received three grants from the European Research Council (ERC). One of them was used to establish MIM Solutions, a spin-off set up at the University of Warsaw. Is there a recipe for success?
Piotr Sankowski*: I was lucky with my first project – everything came up roses for me. However, the second one and the third one required more effort; I needed to submit my applications several times. If I’m not mistaken, the application for the second grant, which would be later used to establish MIM Solutions, was submitted four times. The application for the third grant was filed three times. So you have to be persistent, but you also need to come up with interesting research ideas. It is also important to publish as many papers as you can. Being involved in various projects, not necessarily focused on publishing papers and making money but on taking actions for the scientific community, may also prove helpful.
You are the founder of IGAFIT – Interest Group on Algorithmic Foundations of Information Technology – a group of European researchers specializing in algorithms. What do you think of Polish scientists in the groups that develop algorithms or, more generally, AI?
My main field of study is algorithmics. As far as that area is concerned, internationalization works phenomenally well. Poles made a major contribution to setting up the IGAFIT group, which has been successfully integrating the European algorithmic community. We created a new conference which we hold every year. Last year we also organized postdoc workshops to integrate younger European scientists. They attracted a turnout of about 70 people.
But as for artificial intelligence, our position in the world is not as strong as it could be. According to the 2019 Global AI Talent Report, there is the same number of talents in Poland and in the Czech Republic. Only that the Czech Republic is several times smaller than our country! So there is still a lot to do.
The next European program Horizon Europe will require closer cooperation between Polish and foreign scientists. Are we ready for it?
I still don’t know to what extent Horizon Europe will be different from Horizon 2020 in terms of requirements regarding participation in grants and cooperation you just mentioned. Cooperation has always been necessary. And, to a certain degree, we always were unprepared, so I’m afraid that this time it won’t be any different. On the other hand, the number of ERC grants in Poland is growing. We have seen some improvement but the changes are too slow. We should do more.
The final version of “AI development policy in Poland” should be ready soon. What do think of it?
I haven’t seen the final version yet. What I liked in the most recent version was the idea of dedicated artificial intelligence studies. We should take efforts to start teaching artificial intelligence as early as possible and to present students with challenges so as to encourage them to explore that field on their own.
Unfortunately, as far as involvement of the scientific community in AI is concerned, a lot of things are still missing. For instance, there are no clear goals or research priorities. The Polish government could act as “an intelligent contracting party” and set long-term research goals. We should speak clearly about what scientists ought to do that could be useful in 10 or 20 years. In Poland we still embrace the idea of abstract science that doesn’t have to serve any purpose, but in the field of AI it is impossible to conduct advanced research and ignore utility.
I’m also not a great fan of the idea to create the Virtual Research Institute. It seems like this institution is not going to change much. And of course there is still the question of money. So far, no funds have been earmarked for that project. Instead, we would need an institution that would be able to compete on the international scene – a new institution that would aim for the sky.
Generally speaking, the policy lacks a long-term vision and courage to create something new.
And what do you think about the chances of Poland in Europe? Are we going to catch up or drift along?
To me, speeding up seems unlikely. The process of reforming Polish science and getting to the average European or global level is very slow. We may have made efforts to bridge the gap, but everything is moving at a snail’s gallop.
Breakthrough applications for artificial intelligence still lie ahead. Today, AI can be seen in different places, but in 5 years it will be everywhere
Can we do it faster? I believe we can. We may have our weaknesses but we also have our strengths. However, artificial intelligence cannot be further developed if the solutions are not put into practice. For example, in the USA applications are defined mainly by the private industry. This is why we should cooperate with the companies that will use the technologies on which we are working now in the future. Unfortunately, as for now, this is not the case.
Several years ago the University of Warsaw established the University Technology Transfer Center (UTTC) whose role is to support the process of setting up spin off companies. Is the environment in Poland friendly enough for such initiatives?
I wish there were more such companies associated with universities. What we have today is a drop in the bucket.
Why aren’t such companies set up?
Cooperating with universities is not always easy. Sometimes we find common ground and sometimes we don’t. MIM Solutions appreciates the cooperation and contacts with UTTC. They are trying to help us. But our domain requires expertise. Not everyone can truly understand what we do.
If someone has a new research method or a specific product, they are easier to sell. IT products, especially the ones connected with artificial intelligence, are sometimes hard to explain, so the help is limited.
At the university they don’t tell us: “Great, we’re going to help you with this and that.” In our environment, you must always think whether the value of what you get is bigger than the administrative costs generated by the university. We need a breakthrough idea, courage to cooperate and a completely new approach to set up start-up companies.
You are a member of the Council of the National Center for Research and Development. INFOSTRATEG is one of the Center’s flagship programs, which covers image processing, personalized medicine, machine learning in robotics, cybersecurity and intelligent management systems. The recruitment procedure for experts to devise assumptions for the program ended a year ago. What stage are you currently at?
We are reaching the final stages of the process. On the way, we have experienced many difficulties. For example, we had problems with recruiting experts. As we didn’t receive as many expert reports as we had hoped for, we have set up a team responsible for preparing the program, which will hopefully be soon accepted by the Ministry of Science and Higher education. It should be launched in the second half of this year.
Acting through INFOSTRATEG, we want to interest the AI community with specific issues and challenges connected with robotics, Polish speech recognition, or management of data existing in Poland. We aspire to be an intelligent contracting party.
One of the important aspects in which this program differs from the previous ones is an assumption of an internal competition between the teams implementing the projects. We are planning to divide it into several stages and to have several teams participating in every stage.
What do you expect from your approach?
We’re hoping for a change; until now, after the people had received their grants, they not always got involved in a valuable project (of course not everyone is like that). As a result, projects often end up being some laboratory prototypes and the projects, whose chances of being successful are slim, are not cancelled even though they should be.
We are very far away from setting mathematical requirements for those algorithms: from what they should do to be called fair or ethical
Apart from the above, one of the purposes of the program is to provide the community with some sets of public administration data and to create new databases. Irrespective of whether these prototypes prove successful or not, providing data sets will be a landmark.
Who INFOSTRATEG is addressed to?
To everyone, especially to companies, consortiums, and research teams.
What do you think about AI having so much hype these days?
Breakthrough applications for artificial intelligence still lie ahead. Today, AI can be seen in different places, but in 5 years it will probably be everywhere. We may witness revolutionary applications.
However, artificial intelligence itself is not subject to any revolution. Artificial neural networks were created in 1960s. Concept-wise, little has changed since then. Its rather a scale effect. We are making bigger networks and more powerful computers; we have finally managed to train deep neural networks and that surely has helped to bring better solutions to some problems. Other problems do not need neural networks because, for instance, decision trees have proved more effective, despite being an old technique.
The European AI is supposed to be ethical. And it’s not only about caring for the Europeans, but also about being more competitive on the global market. One of the components of the idea of responsible artificial intelligence is explainable AI. Will this way of thinking translate into an economic success?
It is always pleasant to think that we are doing something new, but in the United States they are also working on those AI aspects, because if we want artificial intelligence to be used for trading on the stock exchange, it has to be explainable. Actually, that is a regulatory requirement and it is the same everywhere around the world. So you can’t think of it as a competitive advantage.
But I agree that personal privacy in Europe is respected more than elsewhere. This is good but will it lead to any competitive advantage? I don’t know because in my opinion preserving privacy is getting less and less important.
Do you think it would be possible to include ethics in the curriculum of your faculty. And, if so, could it be something more than just a short course students have to attend to get a credit?
I think it could work. The problem is that this important aspect is hard to understand and to explain. There is no mathematical definition that could precisely explain what a fair algorithm means. We often get involved in purely philosophical discussions, but we are very far away from setting mathematical requirements for those algorithms: from what they should do to be called fair or ethical.
At the moment, we have only case-related knowledge: we know that something is wrong with the algorithm when we discover that it has given a job only to candidates who play polo. We should take a step back and try to find out why only polo players have been employed. In other words, we need to identify a mathematical property that hasn’t been respected.
Once we have identified such properties, we will be able to assess whether such algorithms serve their purpose. But first we need to know what we want and not only be able to tell that something has gone wrong.
*Piotr Sankowski – professor in the Institute of Informatics, University of Warsaw. His research interest is in algorithmics with the focus on algorithmic network analysis and data science algorithmic aspects. In 2009 he took his PhD in Physics from the Polish Academy of Sciences. He is the first Pole to have received three ERC grants: Starting Grant (2010), Proof of Concept Grant (2015), and Consolidator Grant (2017). In 2018 he received the individual Crystal Brussels Sprout Award and the National Science Center award. Since 2016 he has been a member of the Council of the National Center for Research and Development.
Przeczytaj polską wersję tego tekstu TUTAJ