Jan Kralj & Anton Donchev: Research = necessity + curiosity

The fourth conversation in the Scientific Cognition series features Jan Kralj, a junior researcher at the Jožef Stefan Institute (JSI) in Ljubljana, and Anton Donchev, a PhD student in philosophy of science at the New Bulgarian University in Sofia, Bulgaria.

Jan Kralj obtained his Masters degree in mathematics. He is now pursuing his doctoral thesis at the Department for Knowledge Technologies at JSI. He is working on machine learning, particularly data mining in a network setting. He also participates in a flagship European Human Brain project, and collaborates with The National Institute of Biology. He also has experience working on computational creativity.

Anton Donchev is pursuing his doctoral thesis on topics, not dissimilar to those of his Masters thesis. He deals with a question that is inevitable for any experimental scientist, i.e. to what degree does the evidence at hand confirm a scientific hypothesis. He is building a formal model, which would explicate the relation between the best explanation of evidence and hypotheses.


Jan and Anton both work with data. Jan deals with data on a computational level, in the domain of semantic data mining. This means that he also takes background knowledge into account when developing computational systems that look for patterns and new knowledge in large datasets. Anton, on the other hand, deals with information regarding scientific evidence and explanations, and how they relate to scientific hypotheses. He is using computational models to clarify the relations between the explanation of evidence and the probability of hypotheses.

Our guests talk about the curiousity that led them to their present research topics. However, this begs many questions. Is curiosity alone sufficient for research of particular problems? Is something else also needed? What is creativity? Can computers be creative? How is creativity defined in different scientific ares? Why is it important for broader audience to understand what mathematicians actually do and what artificial intelligence is? What does The Terminator have to do with all of this?

Do scientists actually use the models provided by philosophers of science? Of course, our guests did not avoid their future plans — especially the pros and cons of research in academia and industry. More questions arose: Can philosophical theories be empirically tested? Why does mathematics provide a bridge between computer science and philosophy? Why should this bridge be passed more frequently?


DRIVING THE RESEARCH: ON CURIOSITY AND NECESSITY

Anton Donchev: Could you tell us what are your current research interests, what are you currently working on?

Jan Kralj: Sure. My name is Jan Kralj, I work at the Jožef Stefan Institute, Department of knowledge technologies, Ljubljana, Slovenia. Particularly, my research focuses on machine learning and data mining in network settings which means that I am interested in studying data sets which have a network structure. I do not just work with data instances, but rather instances which are interlinked in some sort of network. Ultimately, I try to extract knowledge by both exploiting the data itself and the way that the data instances are connected with each other.

Kralj: Creativity means adding new lines or new nodes in the network of knowledge that humans have. There is also art, but in science, creativity means either developing new ways of looking at the existing things or discovering new things to look at.

I don’t want to go into too much into detail. I started off by studying mathematics, I have a Masters degree in mathematics. During my math studies, I drifted from the more theoretical mathematics towards the more applied numerical mathematics and from there I was just one step away from machine learning and what I am doing currently.

I currently work on a project of my own—I am a junior researcher—which means my mentor is receiving funding for my PhD studies. Next to that we’re also involved in the Human Brain Project, which is a European flagship project, and then several other projects, especially with the Institute of Biology which we consider to be a nice source of data for our experiments.

And you, Anton, what is your research about? What are your current projects?

Anton Donchev: I’m a PhD student in philosophy of science at the New Bulgarian University, Sofia, Bulgaria. I am in the second year of my Phd and at the outset I was generally interested in what’s going on around us which immediately implies going in science. Before that my central question was: Given different hypotheses or theories in science, how do we know which one is most strongly supported by the evidence that we’ve got?

It turns out this is a philosophical question which brought me to the philosophy of science and in fact to confirmation theory, which is a field in philosophy of science dealing with exactly the question of connection between scientific hypotheses and theories and the evidence, the data that we have. I wrote my Masters thesis on this topic; specifically, I defended the advantages of the Bayesian approach in confirmation theory and I compared it to classical approaches.

Donchev: Given different hypotheses or theories in science, how do we know which one is most strongly supported by the evidence that we’ve got?

In my PhD project, I continue the trend, as you might say, but I now explore the connection between scientific confirmation—so the connection between evidence and hypotheses and theories in science as a confirmator—but also the connection between confirmation and explanation. Therefore, I am also interested in how scientific hypotheses and theories explain the data that we have. In particular, I am trying to build a Bayesian explication of inference to the best explanation.

Inference to the best explanation tells us that out of many hypotheses and theories, we should choose the one that best explains the evidence and it is the most probable or the closest to the truth. But inference to the best explanation does not really have good formal explications up to this point. There is an ongoing debate whether it’s compatible with Bayesianism and with Bayesian confirmation theory or not. So, I argue that it is and I’m trying to show how it is compatible.

Anton Donchev: What motivates your research? What drives your research? Maybe tell us a bit more about that.

Jan Kralj: Well, I started in mathematics. I believe mathematics is probably the flagship scientific discipline where research is done for research’s sake, at least in the more theoretical part of mathematics. In part, of course, it must be curiosity which drives any research effort. Curiosity, especially, not only drives but motivates the researcher. If I wouldn’t be curious, that is having a new idea without being curious whether it works or not, then I would never get the motivation to actually do it.

Kralj: I am primarily developing and improving algorithms for machine learning or data comprehension. Which direction I will go depends also on the data I have at hand and the data that I have to analyze. In brief, my research is directed by necessity and driven by curiosity.

However, since I am working on various projects, my specific field of research is also directed by what is needed. Currently, I am primarily developing and improving algorithms for machine learning or data comprehension, for example. Which direction I will go depends also on the data I have at hand and the data that I have to analyze.

Overall I would say that the motivation for new research comes in looking at what I already have and then being curious about and reading a lot about how this data was already tackled, what was already done with it and, well, going through my own personal library of what I know that could be done, trying to find new ways to look at the data.

I think most of the scientific research can be in some way or another explained by: “This brilliant scientist looked at this data in this new way and got a new explanation.” I’m not saying I’m a brilliant scientist, but all the time I try to find a new way of looking at things which will maybe provide a simpler explanation, that is, a simpler and more effective explanation of the data. In brief, my research is directed by necessity and driven by curiosity.

Anton Donchev: Thank you. I completely agree with the point about being interested in what’s going on around you. That’s what brought me to philosophy in the first place and it’s really the greatest motivation that actually drives my arts. I work on my own project, which means that I don’t have to listen to anyone else’s ideas if I don’t want to. I’m completely driven by my own interests, which really works for me and I’m glad of that.

I would say that my motivation is kind of split. Part of it is internal – I’m into philosophy of science because I’m interested in the questions that I research, like scientific confirmation, scientific explanation and so on. But I would be dishonest if I said that here’s just the internal motivation and no external motivation. Funding is very important. I wouldn’t be able to do what I’m interested in if I didn’t have the financial means to do it. I would be lying if I said that it’s just my internal motivation and that I’m completely unfazed by any external matters.

CREATIVITY: A PURSUIT FOR NEW CONNECTIONS

Anton Donchev: How would you define creativity? That’s a very interesting question for me.

Jan Kralj: My department has just finished a series of projects on computational creativity. This is a topic of ongoing debates – how can we make computers creative if we can’t even agree on what it means to be creative? I remember a nice talk by professor Simon Colton, who made a computer program which was able to paint paintings. Colton said he’s not focusing on what creativity is but what it isn’t. He would always explain exactly how the computer comes up with ideas for the paintings it makes and then he asked people: “Do you think this is creative?” And, of course, because it was made by a computer, people said: “No”. Then he asked: “But why is it not creative?” The response was: “Well, it’s only doing something from inside.” This led him to add an algorithm which was influenced by outside events.

Then Colton asked again: “Is it creative now?”

“Well it’s not.”

“But why isn’t it? It is now affected by outside events.”

And so on. Overall, I think it’s hard to say what creativity is apart from the nebulous “Coming up with ideas which did not exist yet in the world before.” In my personal philosophy, I’d say creativity means adding new lines or new nodes in the network of knowledge that humans have. There is also art, but in science, creativity means either developing new ways of looking at the existing things or discovering new things to look at. An example of this is the theory of evolution. The creative step was to think of a new way of how this abundance of species that exist would come about. Therefore, we tried to find a new explanation for an existing phenomenon.

We have an example that we always share in some of our projects. It is about an explanation of how bisociations were formed – we call them bisociations – from two distinct fields of research, chemistry and health. There was a field of research which covered the fact that low levels of magnesium in the bloodstream can cause migraines in people. There was also another, completely separate field of research where they showed that drinking some sort of fish oil or some nutrient increases the amount of magnesium in the bloodstream. The creative lines of research were trying to put these two together and try to mitigate the consequences of migraine using some new medicine.

Donchev: We cannot change the power of our brain given by the nature. But we can change the amount of work we’re putting into our projects.

Anton Donchev: I was just about to say that it’s not only about discovering new things but discovering the connections between things that are already there, but you already said it. It may therefore be a lot of different things. Creativity could be discovering things that we haven’t known before, it could be learning new connections between already existing things, etc.

It seems that we have very different definitions of creativity in different fields. In cognitive science, there is another definition of creativity, which is quite different from everyday understanding of creativity, namely coming up with a new idea. It seems that in computer science there is a difference as well as you just said.

In philosophy, one definition of creativity could be that we follow one compatible method. That means that we follow a logical or a probabilistic method that comes up with results which are not in the premises themselves. The methods comes up with new information, but not new information about the logical connections, but actually something new.

Kralj: I’m not exactly sure how creativity appears, but the necessary condition for it are practice, knowledge, experience.

Another interesting question concerns the role of creativity in our own research. For example, do you follow any kind of heuristic or pattern when you do your own research in order to be creative, in order to come up with something new?

Jan Kralj: Personally, I’m not actively trying to say “Okay, let’s be creative” because that’s the best way to stifle any new idea, as it puts you under the pressure of “I’m not thinking of anything new right now.” The best way to be creative in research is to let yourself be exposed to a lot of ideas because if you want to draw a new line in the knowledge network of humanity, you have to know as many nodes as possible to get the idea, that is to draw a line between two concepts.

For example, in one project, I’m trying to connect the field of network analysis techniques in data mining and something called semantic data mining. My idea is to look at the background biological knowledge of genes in a different way. If this idea works out, it is a nice example of  looking at any existing phenomenon in a new way with new methods. But for this idea to come up I had to know both sides of the argument.

My point is that the creative part comes in when you see two similar things and start asking yourself the question “Wait, are these things somehow related? Can I draw a connection between them?” But you can only get this idea by being exposed to as many views already as possible.

Creativity is something that pops up once you have enough experience. I think it’s similar in art. I can’t just get up and draw a Picasso. If I want to be a good painter I have to teach myself to paint well, to paint in different styles, to see the world in different ways and after a while I would maybe get an idea. I’m not exactly sure how creativity appears, but the necessary condition for it are practice, knowledge, experience.

Anton Donchev: Absolutely. In fact, I read one of your articles and you’re not only connecting things that already exist but you also developed the Hedwig system?

Jan Kralj: No, but it was designed at our department, specifically by my colleague Anže Vavpetič.

Kralj: My point is that the creative part comes in when you see two similar things and start asking yourself the question “Wait, are these things somehow related?

Anton Donchev: Back to the topic of creativity. I didn’t hear you say anything about the talent and I’m not a believer of talent either. In my opinion talent plays a very small part. However, I’ve seen people do very productive work in far shorter time than I would imagine. However, any kind of genius or talent doesn’t become successful without putting his or her back into it. Hard work has to be involved either way.

We cannot change the power of our brain given by the nature. But you can change the amount of work you’re putting into your project. Therefore, I would agree with you again on the point that creativity requires a certain level of experience. You simply have to do a lot of work before you can get any ideas. It’s the same with me.

My so-called creative process is not very different than the creative process of any other PhD student. The first base is choosing the important things that you should read. Then you read a lot. Then you try to reconstruct what you read in order to see whether you actually got it or not, because when you try to reconstruct it, some of the things that you thought you got in fact come out as “You didn’t get it. You don’t understand this, period.” This thought process takes as long as it takes and you hope that at some point you connect the dots that were previously not connected or see the position of another something that no one else thought about. That’s creativity for me, personally.

ON PROMOTION OF SCIENCE, THE TERMINATOR AND PHILOSOPHY’S PUBLIC IMAGE

Anton Donchev: Would you say that promoting science to the general population is a good thing or not? How do you think it should be done?

Jan Kralj: Of course, I think it’s a good thing. I’m a mathematician by training and what really annoys me about people, whenever I mention the fact that I studied mathematics, is their first reaction, and any mathematician will confirm: “Oh, I always hated math, oh, I was never good at it …” My reply to that is: “Well, that’s why you have us so you don’t have to bother with it.”

Overall, I think science has a bad reputation. A lot of people consider scientists to be introverted individuals only interested in doing what they do because of their own curiosity without any societal benefit. That they are just using public funds for their own amusement. Promoting science, especially educating the general public, is something I consider more or less vital for the Western civilization.

Kralj: It is vital for our society that the public maintains a positive outlook on science. Because we need people in science, and we need good people in science to do good work. This can only be achieved by making the public interested in what we do, especially children.

Our world is the way it is, for good or for worse, because of the scientific breakthroughs of the last 300 years. Science is what raised the average life expectancy by a factor of 2 in the last 450 years. It’s what killed tuberculosis, it’s what won over polio, it’s what brought us to the moon, it’s also what got us Hiroshima. However, if you look at the state of the world now and compare it to any other point in human history, it’s beyond comparison. The problem is that people forget what brought us here and start to mistrust science. What is more, they doubt the scientific method and the rational way of looking at the world. This can all go down very quickly if people stop thinking rationally.

I’m lucky enough that what I’m doing is fairly easy to explain in general terms, which is good for promotion of science. It’s fairly easy to explain to people what a network is, it’s easy to explain that I’m doing something with these networks by trying to learn something from the network. However, I work in machine learning and machine learning is a subfield of artificial intelligence. Of course, people have seen The Terminator, and artificial intelligence is open to misconceptions from the public. We have to explain what we’re doing and how we’re doing it. And that it will not end the world. If we don’t do that, the public will soon turn against us and mistrust our efforts. It is therefore almost a necessary evil to promote our field of science. If my field doesn’t explain to people what we’re doing, people will start to mistrust us.

But I don’t think it is just a necessary evil. People need to be familiar with our work such that they can be curious about it. Once they get interested, they might join us, help us. Researchers are no elite clique doing things no other people can do. We are people who are interested in things and other people can also get interested in these things.

It’s important that the new generation, also new generations of parents, know what artificial intelligence and machine learning are. If their children ask them what this is, they won’t just say: “Oh, it’s some hocus pocus that some people in white lab coats are doing.” They have to be able to explain that these are methods and they are improving our world. Maybe that will get children interested in these methods.

Donchev: It’s not just the media’s responsibility to cover science well. It’s also the people’s own responsibility to filter out good information from the bad.

To conclude, it’s vital for our society that the public maintains a positive outlook on science. Because we need people in science, and we need good people in science to do good work. This can only be achieved by the public being interested in what we do, especially children.

Anton Donchev: Concerning the topic at hand, does your field also have civil science projects? Do you involve citizens in any way? Because in some scientific fields there are projects which involve citizens. In astronomy, astronomers have civilians looking at the photos from deep space and classifying what they see. Scientists found out that people do this a lot faster and a lot more accurately. They now need them to do it. I was therefore wondering whether in computer science there is a way for civilians to do the actual work even if they’re not experts.

Kralj: Of course, people have seen The Terminator, and artificial intelligence is open to misconceptions from the public. We have to explain what we’re doing and how we’re doing it. And that it will not end the world.

Jan Kralj: I can’t think of an example, but a related example would be the recent way that Google, the company, is making all of its machine learning and AI software open source. What Google is doing and what they’re developing is called Tensorflow and that’s basically the structure that allowed them to win at the game of Go a year and a half ago. It’s possible for anyone to just go on the particular website that they set up, download all of the software that they use and run it on their own computers.

I don’t know of a case where people would actually join a project, but enerally it is not difficult to involve a new person into some project. The knowledge that we require can be accessed by having a computer and a keyboard. Astronomy is different, you need specific equipment, you need a telescope to do astronomy. But everyone has a computer.

Anton Donchev: Regarding promotion of science, I am for promotion of science, of course, when done properly. I don’t suppose that there are many people in either science or philosophy of science who’d be against the promotion of science, that would seem strange. But the idea is that too much attention has been paid to the topic of whether the media covers scientific projects and the results in a good way.

In my opinion, this is connected to the very important topic of educating the general public to filter out results which are badly covered or the yellow press from actual scientific results. Because it’s not just the media’s responsibility to cover it well. It’s also the people’s own responsibility to filter out good information from the bad.

Jan Kralj: It’s getting increasingly difficult due to the fake news media, social media and the whole propaganda that we see nowadays.

Anton Donchev: Of course, that’s why the problem falls more heavily on the people managing to filter out bad information by themselves rather than censoring the media.

Philosophy of science, as most parts in philosophy, also has a bad image, a bad public relations image. There are questions whether it is really needed, whether it does something that’s productive, whether it’s any good for scientists or not. In my view, this could be amended with even more effort put in telling the general public, in a meaningful and clear way what we do, how it is connected with science and why it’s important. The most helpful thing would be to hear scientists on the side of science saying that they found something useful or something meaningful in philosophy of science.

ON FUTURE PLANS, ACADEMIA AND THE INDUSTRY

Anton Donchev: Would you tell us a few words about your future research plans? What are you inclined to do?

Jan Kralj: I sort of have two threads of research, both connected to network analysis, going on. My immediate plans are to wrap both of them up and to wrap them up in a nice package which would carry my name in front. The more long-term plans concern an important dilemma of researchers – whether to go into industry or academia.

People in machine learning need to be exposed somewhat to industrial environment, to some environment where we need results. In my undergraduate years I was already doing some programming and I was already got to know the industrial aspects of my eventual career.

The last year, being a teaching assistant, I had my first opportunity to teach a course on machine learning, and I always found it fun to teach, I always found it exciting to explain things to people who didn’t know it and to see that look in their eyes when they got it.

Donchev: I would be interested in looking into the philosophical questions which could be re-formulated as empirical and empirically tested.

I don’t want to completely abandon the academia. Once I have enough knowledge I also want to transfer it. Transferring the knowledge has to be in some sort of academic way where the bottom line is not the only thing that counts. But … I don’t know yet where I’ll be in two years. I would like to work on a project within a company which is interested in the final result of the project, not only the method. My plan would be to reach out, but not sway completely out of the academic waters.

What about your plans, Anton?

Anton Donchev: Mine are more or less the same. My immediate plans for the future are finishing my PhD, of course. I hope that I can get some publishable results which is a pretty good thing to have in the beginning of an academic career.

I always thought that in philosophy of science there are many questions which could be tested empirically, but they’re not really formulated as empirical questions. I know that we cannot get the answer to every question in an empirical way, but I think that there is much potential in testing empirically specific questions in philosophy of science. If I stay in academia, which I would very much like to, I would be interested in looking into these kind of questions which could be re-formulated as empirical. They may not be solved by empirical research, but they could be informed by empirical research and this in my view is quite okay for now.

Kralj: I always found it fun to teach, I always found it exciting to explain things to people who didn’t know it and to see that look in their eyes when they got it.

I, too, would like to stay in academia, if possible, because with philosophy of science, it’s slightly difficult to find work in industry. If you’re not in academia, skills that you have in philosophy of science, e.g., analytical skills, are not all that applicable in industry. You can, however, pick up some programming or something like that, like a side project, which can also help inform your research in philosophy of science.

Jan, thank you for this lovely talk and discussion, it was very pleasant. I wish you good luck with your research and I look towards more collaboration between philosophy of science and computer science. I think that that is a venue worth pursuing.

Jan Kralj: I think mathematics is already an example of such venue. Mathematics is a form of philosophy and computer science is a form of mathematics. This means that we are not that far apart.

Anton Donchev: Absolutely, I agree. Thank you again!

Komentirajte prispevek