Increasingly rapid development of IT and AI means a growing demand for employees who are constantly poached from universities. Although the problem of financial conditions in the case of doctoral studies may seem down-to-earth, it is far from being unimportant, says professor Krzysztof Krawiec, Ponznan University of Technology, in an interview conducted by Maciej Chojnowski
Maciej Chojnowski: You said that the policy of Polish authorities regarding artificial intelligence may have a significant impact on the future of our country. Why?
Professor Krzysztof Krawiec*: There is a saying that artificial intelligence devours software. It is actually a paraphrase of an older saw that information technologies (IT) take over all areas of the economy. Today every business has to be, to some extent, IT business; it is hard to imagine an area of human activity where a certain amount of computer assistance would not be useful or necessary. Computer techniques and computerisation are very widespread, not only in the case of pure information processing.
At the current stage, this software is being supplanted by a new generation of software, which is no longer developed by humans, but automatically derived from data. It is a revolutionary change. Many functionalities in IT packages, which only a few years ago were laboriously implemented by programmers, are replaced today by functionalities which are automatically obtained from data in the process of machine learning. Obviously, this allows for much greater efficiency in software development.
How fast is that development?
Today, we are seeing an exponential increase in the pace of development of applications and services that involve AI. And since nowadays IT is commonplace, the progress in AI, which is based on IT, affects all areas of life. This process cannot be stopped. We may try to regulate it to a certain extent, but trying to suppress it would not do anyone any good.
Why?
One of the reasons is that in the world there are economies and cultures that do not really care about negative consequences of such technologies, And if we do not try to keep up with the changes, then in a few years’ time we will be forced to import those technologies.
In our everyday life, we use the term “artificial intelligence” rather loosely, often referring to “machine learning”, “speech recognition” or “machine vision”. Does it make sense to use such a generic expression?
I believe it would now be unnatural to try to introduce a new term to describe it. Obviously, there are many clichés and negative perceptions of artificial intelligence, as some people still associate it with Terminator or hostile robots, but they are disappearing. What is more, I have noticed that in the popular science media the term “machine learning” has also made headlines. We can say that the awareness is growing. And the language is evolving independently. Not much can be imposed.
Besides, I believe that using the term “artificial intelligence”, despite all its constraints, is purposive because it opens people’s eyes to a broader understanding of intelligence, broader than a traditional, intuitive understanding to which we got used.
Today, we are seeing an exponential increase in the pace of development of applications and services that involve AI. And since nowadays IT is commonplace, the progress in AI, which is based on IT, affects all areas of life.
Let us take a look at a thermostat that controls the temperature of a heater. Most of us do not perceive this device as particularly smart although it is capable of interacting with the environment around it. And if I were to define intelligent systems, I would say that they are systems which are fitted with sensors allowing them to perceive the world around them and which are able to react to changing circumstances on the basis of data collected by such sensors. Surely, the structure of the thermostat can be very primitive allowing only to open the valve when it gets colder and to close it when it gets warmer. For most of us, this is no intelligence. But if this device remembered the values of temperature from the last few hours, and was able to adjust the size of flow passage more precisely, then perhaps some of us would call it intelligent.
And what would we say about a thermostat which would collect data on temperature in a room for several months and which would learn about the habits of the user? What if it knew that the user is at home on weekends and wants the temperature to go up in the early morning?
So “intelligence” is not so easy to be clearly defined.
Yes. The question where artificial intelligence begins and where it ends can be addressed to psychologists or cognitivists, but they would also have problems with giving a straightforward answer. It is no coincidence that psychologists measure intelligence in so many different ways. We talk about verbal intelligence, non verbal intelligence, intelligence connected with spatial awareness, and so on. So, as computer scientists, we should not feel particularly guilty for not being able to determine exactly where artificial intelligence ends, since the concept of intelligence itself has not been clearly defined yet.
During the recent conference of the Polish Agreement for the Development of Artificial Intelligence (PP-RAI 2019) you said that you were not a supporter of long-term planning, especially in the AI domain. However, you added that it was worth identifying the most important areas in which we should specialize. Who would identify them? Government? Researchers? Both government and researchers?
Definitely government and researchers. Neither the government nor the artificial intelligence researchers want to turn their backs on each other, as both parties are committed to the development of the country.
However, I believe that in today’s world, where there is no time for everything, we cannot wait for years to develop a strategy or vision. We also do not have time to study piles of documents whose authors split hairs trying to describe everything in great detail. Strategies are necessary, but they should be simple and concise and identify the most important areas.
It seems that the community of researchers and government authorities are building consensus. I am referring here to the most promising domains that have the biggest social impact, such as medicine, robotics, natural language or transport.
How should that cooperation look like in practice? Should there be a discussion followed by a decision about particular investments in specific areas? Or should government authorities outline their problems in a given area without specifying what technology ought to be used?
In my opinion, the latter idea is better. The needs are most efficiently identified by the industry which approaches Polish artificial intelligence centers and invites them to collaborate. And often the results of such cooperation are very positive and meaningful.
Of course, the nature of collaboration with government authorities is different from what we experience in the case of research and implementation projects. It is a different scale of challenges, it is a different time perspective. Yesterday, I had a conversation with professor Jarosław Arabas and we both noticed, half jokingly, that we envy Czechs one thing.
What is that?
It is a small country where the artificial intelligence community is concentrated. As a result, the concentration of know-how and research projects is natural. For example, Czechs are very good at computer vision. Recently, they have been very active within the Confederation of Laboratories for Artificial Intelligence Research in Europe [Editor’s note: CLAIRE, European initiative for research, innovation and cooperation in the field of AI established in June 2018]. They quickly offered their services as a regional organizational center and they already have their office.
It seems that the community of researchers and government authorities are building consensus. I am referring to the most promising domains that have the biggest social impact, such as medicine, robotics, natural language or transport.
We are a big European country, so it is harder for us to coordinate. Establishment of the Virtual Research Institute would be therefore strongly recommended. It would be required to make sure that the institute is financially stable and receives orders from the government.
What is the cause of that Polish fragmentation?
Some blame the model of how science is financed. For example, research projects financed by the National Science Centre are rather small. The fragmentation of the grants prevents small research teams from growing. Consequently, it is difficult to reach a critical mass.
We, computer scientists, having a computer at our disposal and knowing the current state of research, are able to kick off a project without anyone else’s help. This trend is particularly relevant to theoretical computer science. That is not the case of robotics, which relies heavily on equipment and costly instrumentation; by definition, your guiding principle should be “strength lies in numbers”.
The Horizon Europe program, which is scheduled to be implemented in 2021, will favor consortia. What perspective for Poland does this program bring? Are we ready to collaborate?
I think so. After 15 years in the European Union we have become much more internationalized. As far as computer science is concerned, we are trying to do everything on an international scale. National scientific conferences are becoming things of the past. Each of them wants to be international in scope.
We, Polish researchers, travel around Europe, we have our networks of contacts. Many people from Poland participate in CLAIRE, which we have already talked about; they are either members or work in advisory groups. There is also ELLIS [Editor’s note: European Laboratory for Learning and Intelligent Systems], an organization which brings together institutions specializing in machine learning. Many of CLAIRE or ELLIS institutions, acting as consortiums or under an agreement, are jointly participating in a big undertaking launched this year by the European Commission to identify four network projects. They are to form the basis for European excellence centers focused on robotics, high performance computing, explainable AI and trustworthy AI.
If we had our Virtual Research Institute, it would be our candidate for a member of such a consortium. But as we still do not have it, we decided to participate as Poznan University of Technology. It goes without saying that if we are through, we will try to represent not only our university but also other centers.
Explainable AI and trustworthy AI are elements of ethical AI, which Europe is now trying to develop. Can ethical AI work to our advantage in the technology race with the United States and China?
It can surely be a distinguishing feature. It was spelled out at CLAIRE meetings, for example by professors Holger Hoos and Philipp Slusallek, founders of the organization.
It is about promoting fair AI, distinguishable from highly commercialized American AI or from Chinese AI, which has some issues with respecting human rights. It is a great idea, because this approach is entrenched in European values. Europe is ready. Not only do we have great AI experts, but also renowned ethicists, philosophers of science, sociologists and psychologists.
In August, the Policy on Artificial Intelligence Development in Poland was presented for consultation. What is your opinion about it?
It certainly had some sensible ideas. One of them was the Virtual Research Institute. Another pertained to doctoral schools. Those schools need to be sufficiently financed, which is key for the project to bring practical benefits. Unfortunately, for now, we are rather disappointed with how the Constitution for Science “reformed” the ways of financing doctoral students. Let me give you an example. This year my faculty could accept only five doctoral students as compared to several dozens in previous years.
You cannot treat all disciplines in the same way. Increasingly rapid development of IT and AI means a growing demand for employees who are constantly poached from universities. Salaries offered to young people, even to students, are in most cases far better than the money they would get for their doctoral studies. This may seem a down-to-earth problem, but it is far from being unimportant.
The question where artificial intelligence begins and where it ends can be addressed to psychologists or cognitivists, who would also have problems with giving a straightforward answer. It is no coincidence that psychologists measure intelligence in so many different ways.
I would also add the issue of cooperation with the industry. Since artificial intelligence is an applied discipline, it should respond to social and economic needs. And it is already happening, especially at technology universities, which are, in a way, statutorily established to respond to the staffing needs (although not only) of the industry. Most of us want to collaborate with the industry sector. Direct funds from the industry sector are often something that helps us to retain our staff; within certain projects, we can provide financial support to our employees and prevent them from resigning from the university.
In this case the medical phrase primum non no nocere should apply. I believe that with the right regulatory framework and with unfettered cooperation between universities and the industry sector, their collaboration has a chance to flourish. My experience at Poznan University of Technology shows it is very likely.
Your research team is part of the Laboratory of Intelligent Decision Support Systems and specializes in computational intelligence. What projects have you carried out in collaboration with the industry sector?
For example, we started collaborating with StethoMe, a Poznań-based company. A few years ago this start-up came up with an intelligent stethoscope which can not only record the sound of breathing, but is also able to analyze data collected and make preliminary diagnoses, mainly for children. This year, we held consultations at our faculty for the project team of this company.
We have also been cooperating with Roche IT Polska, with which we have launched a pilot project on the possibility of detecting anomalies in lung computed tomography scanning. Me and one of my master’s students made some interesting observations and presented our results in the Roche headquarters in Basel.
We have also established cooperation between our faculty and the PGE Group. Together, we have been working on intelligent processing of huge amounts of data collected by the company. Our activity in that scope is financed by the National Center for Research and Development.
We also collaborated with TomTom, a renowned producer of digital maps, on providing updates to their maps.
What dreams would you like to come true for Polish artificial intelligence?
I wish we spent as much time as possible doing things that are substantively related to our projects. This is what we enjoy the most and this is how we spend our time efficiently.
*Professor Krzysztof Krawiec is an acting deputy director of science in the Institute of Computer Science, Poznan University of Technology, where he does research in the field of machine learning, computer vision, and heuristic optimization algorithms. Professor Krawiec is a recipient of the Fulbright Commission scholarship, awards granted by the Ministry of National Education, the Foundation for Development of System Sciences, the Association for Image Processing, as well as several Best Paper Awards at international conferences. He is the author and co author of four monographs and over 100 articles published in journals and conference materials. He served foreign internships as a visiting professor at the University of California, Riverside, and at the Massachusetts Institute of Technology. He was the chair of the Polish Computational Intelligence Section at IEEE. Currently, he is a member of the presidium of the Committee on Informatics of the Polish Academy of Sciences.
Przeczytaj polską wersję tego tekstu TUTAJ