Good morning. Today we are featuring Dr Emanuela Girardi in our interview. Emanuela Girardi is the founder of Pop-AI (Popular Artificial Intelligence). She is a member of the Artificial Intelligence Expert Group of the Ministry for Economic Development, which wrote the Italian Strategy for the Artificial Intelligence. She is member of the board of the Italian Association for Artificial Intelligence and of the Industry Task Force of CLEAR (Confederation of Laboratories in AI Research in Europe), of which she is the coordinator of the task force on Artificial Intelligence and Covid 19. She is also a member of the board of AI Data and Robotics (ADRA), which is the new European association that will work with the European Commission to implement the Horizon 2020 programme. Thus, she is an extremely relevant and extremely competent person who will guide us in this half-hour journey on some interesting things in the field of AI. I thank her again and I would proceed to ask her some questions.  

QUESTION (F. ANANASSO)

Thank you for your considerations, which in large part coincide with some of the doubts I had. I understand, however, that the essential thing that we have more than others is precisely the concept of the “human centric”, right? The famous “Human in the loop” that many people say “on the loop”, rightly so, that is, at the end / on top  there must be a “human”, and I am very concerned personally on what the European Commission has said explicitly it cannot deal with, that it is military topics, we already dealt with in other interviews, very worrying because there is no Human In the loop. However, this is a topic that has nothing to do with what we are addressing now, it is only a consideration between us who study and address these topics .

So we have seen the World and Europe. How about Italy? What is the situation of Italy? Among other things, In Italy, as well as the rest of Europe, but we in particular, lately there is also the term “sustainability” that we often refer to. What is the Italian approach, the Italian strategy and our approach, let’s say, to Sustainability and to the importance of using artificial intelligence and investing in artificial intelligence systems?

ANSWER (E. GIRARDI)

Italy has Developed different strategies. At the moment I was involved in the work that led to the development of the artificial intelligence strategy promoted by the Ministry of Economic Development (MISE). There were actually two other initiatives, so we have actually developed a lot of strategies, but to date we have not published a single one. Now the news of the last few days is that a group made up of three ministries has created a new task force that has to re-analyse the document we wrote at MISE, and I hope really they will propose the phase of execution, that is, to actually implement the strategy, to carry it out, the execution that is the fundamental thing, so we really hope so. That will be the turning point for Italy as well, it is a pity because in Italy we have so much excellence, in my opinion, especially as regards, e.g., the research and development of artificial intelligence, but unfortunately there is a lack of strategic vision at the moment, of what could be the applications and uses and developments of artificial intelligence applied in the various fields of the society. 

On the other hand, the proposed Italian strategy that we have made is based on three pillars. The first one was AI for the human being, so absolutely in line with the European Human-centric vision, the second one AI for a productive development sustainable and of excellence, and this is very much in line with the EU vision, with the European White Paper on artificial intelligence that was presented in February 2021 and which promotes precisely the creation of a system of excellence and trust at European level to promote the development of artificial intelligence technologies, and then to invest in education to bring them to companies, schools, society. The last point, however, is the point that I think is the most important and is also the most innovative as far as the Italian artificial intelligence strategy is concerned. And it is AI for sustainability, that is what we have proposed and that we think really needing a paradigm shift, in the sense that it is no longer enough to just put the human being at the centre, because he / she actually lives on the planet and within the ecosystem. And so to say that we only use technologies to improve human life is no longer enough, and we therefore proposed the use of artificial intelligence technologies to achieve the sustainable development goals (SDG) of the United Nations 2030 Agenda. And so just changing a little bit the paradigm and proposing this aspect of sustainability or an approach, let’s call it “planet-centric” if you like, this vision is  very innovative at the moment. In fact it has also been adopted then by the OECD, by the UN and I think it has also been included actually in the last coordinated plan of the European Commission, and I think it is probably the only possible vision. Even  because, if there are 156 Countries that decided that the UN 2030 Agenda is THE Agenda, as well as the SDG goals that, even if we think that 2030 is practically tomorrow, are the most shared goals that there are today at European level, then, in my opinion, once the goals to be reached are defined, as we were saying before, the definition of artificial intelligence are systems that in some autonomous way allow us to achieve certain goals. If we are clear about the goals we want to achieve, at this point, even if they are complex goals, we can use these technologies to achieve them. And so we then analysed for some SDG how to use technologies to achieve them. In particular, one aspect, let’s say very important, we have tried to devote to trying to increase the inclusion and accessibility of people with disabilities. And this is thanks to the use of artificial intelligence technologies, and so these aspects are very important, but they are also very important in helping us to counteract climate change or how to improve the impact on the environment. For example, in this regard, there is a very interesting project that has just been promoted by the European Commission, called “Destination Earth”, which has created a digital twin of the earth that makes it possible to monitor and assess climate change, the effects of climate change, and to test, let’s say, on this digital twin of the earth, the new environmental policies that we want to develop using artificial intelligence technologies. This is also to promote an optimisation and rationalisation of the scarce resources that there are on earth.

Fabio Del Frate received from the University of Rome “Tor Vergata” a Master Degree in Electronics Engineering (1992) and the Ph.D. degree in Computer Science (1997). In 1995-1996 he was Visiting Scientist at Massachusetts Institute of Technology, and from 1998 to 1999 he was with European Space Agency. He then joined the University of Rome “Tor Vergata”, where he is currently Associate Professor of Remote Sensing and Applied Electromagnetism in various Master and PhD Programs. On those topics, he has been lecturer at several European Universities, as well as principal investigator / project manager in several European Space Agency (ESA) and Italian Space Agency (ASI) funded projects, leading research activities in the field of Artificial Intelligence applied to satellite data for Earth Observation (referred to as “EO” in the following).

Q2. What is Machine Learning?

Well, I would say that machine learning is a possible, I guess at the moment the most used one, technical process to generate artificial intelligence inside a machine. We’ve seen already that there are  two main actors: the data and the model, more exactly a mathematical model. The challenge is to find the parameters of the mathematical model so that it can represent the relationships among the data. Everything relies on setting-up a learning process based on the data themselves. We may have different types of models, for example neural networks or support vector machines and, consequently different algorithms to learn the model. The final goal however is always the same: to extract knowledge from the training data used in the training process and to apply this extracted knowledge to new data. By the way, the evaluation of how good the performance of the model is on the new data, that is data not considered during the training phase, is very important. It says to us how much the model is applicable in the real world, how much it is able to generalize. 

In any case, the quality and the quantity of the data used during the learning phase are crucial factors for a successful performance. In particular, the quantity of data needs to be statistically significant to provide the relevant features of the phenomenon we want to investigate. Another important key issue is to decide how complex our mathematical model should be. It would be a serious problem if we choose a too simple model to look for very subtle relationships among the data but, at the same way, if the rules underlying the data are less complicated, also the structure of the model can be lighter, sometime…. less is more, as we say, where less is referred to the  number of parameters characterizing the model. 

By the way, the model complexity  has also to do with the time we need to train the model and with the amount of data which is necessary for the implementation of the complete learning process.

Q4. What is Supervised Learning? Can you give some examples of applications that use it?

Well I would say that Supervised learning is the most common sub-branch of machine learning today. Supervised learning is a form of learning where, for a given input, a “true” outcome is available and this truth can be taught to the model. In other words, as a teacher or a supervisor, for each input given to the model we know what the output of the model should be and we want the model to incorporate the general rule. Typically, this happens by minimizing what is called an error or a cost function, which represents the distance between the desired output values and the actual output values generated by the model during the training phase. Actually, supervised learning can be assumed to be successful when the final overall error over a set of examples, which have not been used during the training phase, is under a predefined threshold. The training phase can be very long, but once we have concluded it properly, we have a very powerful tool, basically operating in real time, for the processing of the new data. Supervised learning can be useful in many fields. A rather massive application regards image classification. In this case the true data are data labelled by human image-interpretation. The classification can be performed at various levels: pixel level, patch level, object level. The level of classification guides the choice of the input that we want to give to the model. More in particular, supervised learning is now very much applied in Earth Observation where we want to make estimation of bio-geophysical parameters from data collected by satellites. An interesting example regards precision farming in agriculture. If we want to monitor the biomass of a crop in order to optimize the amount of fertilizer to be used, we can exploit satellite measurements taken on that crop. However, especially at the microwave region of the electromagnetic spectrum, the extraction of vegetation parameters is not trivial because the measurement is also, let’s say, contaminated by the underlying soil, so AI supervised models can definitely be one of the tools which allow us to retrieve that part of the information in the measurement more connected to the vegetation, discarding the one more conditioned by the soil.

Q6. There are various groupings of Machine Learning algorithms, e.g. regression, classification, and so forth. Can you explain their relevance and difference between them?

There are various possibilities to group machine learning algorithms. I think that the main one relies on the type of the mathematical model. More in particular, as mentioned before, we have models for supervised training and models for unsupervised training. Regarding the former, among the most important ones we have neural networks, deep or shallow, support vector machines, decision trees, ensemble techniques. For the unsupervised models we can mention k-means, principal component analysis, auto-associative neural networks, self-organized maps, and so forth. 

But as your question was pointing out, another possibility to group machine learning algorithms addresses the distinction between regression and classification. Once a certain input is provided to our mathematical model, the difference between regression and classification mainly relies on the type of output we’re interested in. In the case of classification, the output will be a label, a class. For example, let’s consider again a satellite application where remote sensing is used for meteorology. Let’s therefore assume we have to make a meteorological analysis, and we want to decide which pixel in an image of the sky has to be associated to a cloudy sky and which pixel of the image to clear sky, so that the final purpose is to produce a map where we represent with the blue color the clear sky pixels and with the grey color the cloudy pixels. We say in this case that we want to obtain a classification map. For the regression, basically, we want that our mathematical model, as a response of its data processing, computes, or estimates, a number, more in general a real value number. So, if we consider the same scenario set on meteorology, we want to know from the remotely sensed data how much is the humidity or the precipitation rate, real numbers, in a given area. However, it is interesting to note how a regression problem can be translated into a classification problem. This is what happens if I say: within this range of precipitation values, so between these two numbers, I have the class “low precipitation”, in this other interval or range I have “moderate precipitation”, and so on. In this case I map ranges of values into classes.

Thank you, Prof. Fabio Del Frate, we really appreciated. It has been a clear & interesting snapshot of Artificial Intelligence features and relevant applications.

Hello everyone. My name is Simon Delakorda. I’m director at the Institute for Electronic Participation from Ljubljana, Slovenia. The Institute for Electronic Participation is a nongovernment organization which is active in the field of information society, development and digitalization. We are also relating dealing with the issues of artificial intelligence. For example, last year we implemented a citizen dialogue which was also covering actual relevant issues with the development of the artificial intelligence. How this technology is relating to the everyday life of the citizens.

So now the next view or issue relating to the artificial intelligence I would like to focus on is the social challenges of the machine learning. First of all, what is machine learning? Machine learning refers to algorithms and techniques that learn from by themselves, who are confronted with data, observations and interactions with the surrounding world. Basically, machine learning is about algorithms. It’s based on the algorithms and algorithms are simply set instructions how to solve tasks. And now when relating algorithms to the machine learning means that a particular type of algorithm can be learned, but not by the human programmers, but by themselves. Meaning that machine learning is learning by themselves by using statistical approach. And usually there are two phases of machine learning. The first phase is that the machine learning algorithms are being trained by using data, by using the training process. And the result of this process is machine learning model. And in the second phase, this machine learning model is applied to the desired area of application. The idea is that this machine learning model is then solving a problem that we would like to address. For example, machine learning models for prediction if a patient has cancer, tumour or machine learning model, or for prediction how the stock markets is going to behave in the future. Of course, there are some issues issues with machine learning because machine learning can have a serious impact on people’s lives and especially when there can be errors in machine learning which can happen during data collection, data preparation or during the training process. And also we can have issues when we misinterpret the results of the machine learning process. Looking at the social level, a great danger is that the training data used for the training process for machine learning is considered perfect. But we know that perfect data sets would require all possible factors that constitute or influence the situation that we are addressing. To mean that it’s very hard to train algorithm, that they are using all the data from the complex society. And as a result of this imperfect data which is then applied to the machine learning algorithm, it can happen that the results can be too simplified and a situation they are addressing only has a limited number of factors taken into account. The results of this is that the machine learning can lead to bias prediction, especially for the disadvantaged groups of society which are not so much present in data or in the society as a whole and are less represented in the data set. For the case of the bias in the machine learning, this means that the bias data will also result in biased prediction. This is for case when it comes to the rare phenomena or social contexts that are hard to quantify. This means that we really have to be very aware of those specifics with the society which are not seen or included into the machine learning results. Because machine learning is actually operating based on the patterns. And if those patterns are lacking occurrences or being of poor quality, we can get very biased and poor results.

So and my last view on the artificial intelligence is relating to that I would like to share today’s relating to the role of the civil society organizations, especially the role of non-governmental organizations. I think that the role of non-government mentalization in the application of artificial intelligence is very important from the viewpoint of explaining citizens, explaining people, users, everyday people how this technology is affecting their lives, especially in terms of trust into the artificial intelligence. Also in terms of helping them to understand the applications and the impact of the everyday life of the citizens, because there are different interpretations and fears and maybe unfounded negative, their relationship towards the artificial intelligences, which are needed to be discussed, learned educated in trustworthy environments and non-governmental organizations with their members, with their voluntary activities are a link often between the high tech research and everyday people because they are having trust because people having trust in non-governmental organizations. And because of their information, their education activities, and can help citizens to understand what the negative and positive effects of artificial intelligence are. This is one role of the NGOs. And the second role of the NGOs is that they act as watch dogs. That they act as a not exactly supervising body, but someone who is paying attention to the possible discrimination, breach of human rights or negative usage of the artificial intelligence, which can lead to discrimination, which can lead to worse conditions, for immigrant groups, for specific social groups like immigrants or women or Socially vulnerable groups which are not as represented in the society as other social groups. And the NGOs must take the role of safeguarding their rights and also battle combat for human rights when they see that the artificial intelligence is about to bridge those standards. I think these are two main rules of the NGOs relating to the artificial intelligence. There are two minor roles. One is also NGOs as a usual user artificial intelligence, especially those NGOs who are generating large data on this society, and then can then use this machine learning, tools to provide some conclusions, to solve problems, to create visualization on different aspects of society. And the second this role is also important is that the stakeholders are acting as a policy actor, meaning that they are taking part in the forums, coalitions, discussions with other stakeholders on the economy, from the politics, from the research. And so that and those forums, those discussions, they are pointing out the issues of accountability, responsibility, privacy, trustworthiness of the artificial system, basically, meaning that they work as advocates of the society and the humans and strive for the human oriented use of the artificial intelligence. So this would be my, short but not really short viewpoints on the artificial intelligence and the society.

What is the main difference between machine learning and ordinary programming? There are not  many differences, but at the same time they are important. With ordinary programming, we already  have a fairly accurate idea of what is programmed and how things will look. With machine learning we  have a thesis that offers itself as a possible answer through machine learning. Some of these possible  answers are more accurate, they are more elaborated, and some are more like guessing. In fact, it is  all about more or less accurate guessing.

It is about the cooperation with artificial intelligence and human interaction with a machine. Can you  tell us more about the human-machine interaction? Yes, I’m increasingly convinced that artificial  intelligence is not about some kind of intelligence that is in any way similar to human intelligence, but  that these algorithms are really so capable that they become our dialogue partners. As if when I am  sitting around the table with my colleagues we debate sometimes about the program, sometimes  finance, sometimes about politics and so on, we have another chair where this machine is sitting. It  offers us an echo of our thoughts. We have not yet in such a way implemented the artificial  intelligence, but we intend to, because we are developing an interface which will help the innovators  innovate. We would like to see that one of the innovators is a kind of AI. Perhaps it can contribute an  inspiration in a brainstorming session or a talk with humour or uncanny way of folding things together.  Especially if there is a group of experts around the table who are more egocentric and emotionally  exhausting. On the other hand we have a machine, which creates approximation and suggestions that  release the tension in these conversations. At the beginning I have mention that this machine iterates  what is said or put in by the humans. In artistic circles we are dealing with the idea in which these  machines can emancipate through conversation among themselves, machine to the machine. They  develop some lexicon, some level of communication which is incomprehensible to us humans. But can  be meaningful and through this communication we can better understand the way the algorithms  work. We have art projects where data is taken from sensors which are attached to plants or some  environments in which the animals move and act which are equipped with sensors. Then suddenly you  don’t get any more human-human dialogue (excluding that a person is already included, since he has  written the algorithm), but communication between extracted human in algorithm, the animals, or  plants. Art projects that we do encourage interactivity that is not just responsiveness, for example you  press a button, and something jumps, but true interaction, meaning you press, something happens,  and it influences you back that you press again. It creates a communication loop. We have a situation  of intercognition between more forms of liveliness, machine, and plants. Through this communication  which can be very different from that of a human – plant, we are exploring our own prejudices as to  what the plant is and what the machine is. It acts as a bioindicator and so on. The artists are using  these tools in very innovative ways, so we can better understand these tools on one hand and to better  understand the nature or the environment in which we live on the other hand. The amount of data is  so big that sometimes it is bigger than the average human brain can process and because of that it  makes sense to develop algorithms.

…

Well, my name is Bambos Papacharalambous. For most of my career, let’s say almost 30 years, I was mostly involved with the telecom part of the IT business. I’m involved with a wide range of projects that are related to IT in general. I’m the CEO and founder of Novum. Novum is a company that does software development and ICT consulting.

And could you now expand on ethical use of artificial intelligence? Are there some artificial intelligence applications with ethical implications? Okay. As more machines are performing more and more complex tasks nowadays, this immediately directs people’s minds to the use of machines in replacement of human labour. That’s the first thing that comes to people’s minds, and it’s a valid concern. It’s the same concerns that were raised during the Industrial Revolution, and today we see them again as relevant. Is it difficult to allow machines to harvest wheat instead of relying on people’s hands? Is it difficult to allow robots to display human workers in the car assembly lines? So, these basic same questions we see them again coming up, but now in a more, let’s say, complex environment, as machines now have developed, are performing more complex tasks. So, if people had to adjust to the new environment created by machines in the workforce, they would need to go through another readjustment process really soon. For instance, where would the truck drivers find work if the trucks now are driving themselves? Right. Where would the bus drivers find employment if the buses are driven by AI software? Where would you employ an airplane pilot if the planes fly on their own? So, the more complex now tasks that are being developed and are being run by machines, the more these types of jobs are being threatened. And it’s not just hard labour anymore. I mean, why would anyone become a doctor if a robot can perform the same operation better than any human, right? So, these concerns are valid. And of course, there are other examples that raise questions of ethics in the use of AI in our lives as well. It’s not only the labour force. Shouldn’t we be concerned with the use of machines that are used for our security? First, can we be sure that the AI software powering up a security camera in an airport has made the correct decision in identifying the terrorists? Can we be sure that the security system in a mall is following the real thief to his car and not some other innocent bystander? So, these issues are valid. There are a lot of ethical concerns within the AI space, not only through areas like employment or security, but also in the use of military weapons as well. Shouldn’t we be concerned, for instance, by the fact that the armed drones are today doing our fighting for us? Are we sure that the autonomous robot soldier tomorrow would be able to distinguish between the enemy and the friend? So, these are very valid customers. And whenever something big happens, a change happens within the industries in general. Whenever something magical happens, then there is also the chance of negative implications of its use. So the industry needs to eventually find a way to use AI for the good of the society we live in.

And so you mentioned already the impact on artificial intelligence, on employment, on security, on welfare, in the military service, and they also hold on new laws of robotics in other fields than this street field you mentioned. Yeah, this issue of the rules and the laws for robotics. It’s not an easy issue to solve. We are in unatched territory when it comes to a widely accepted set of rules that should define the ethical framework of how robots should work, even in areas outside the military, but also in the space of security, in the space of the welfare of the elderly, let’s say the space of the personal security. At home, we will see robots that will start taking decisions for us. As I said earlier, this is a very new area, and even though there are groups both in academia, in the industry, and in governments also, that try to touch upon these rules, this is at the very, very early stage right now. Right now, the discussion is being done on what the robot should or shouldn’t shouldn’t do. For instance, a robot shouldn’t harm another human. That could be like a primary, let’s say, directive for a machine. But at the same time, a robot should obey its owner and its creator. So that’s another directive, primary directive for a robot. But what if the robot’s owner is giving instructions to the robot that is outside its main directive? So what if the robot is asked to harm another human? Right, but should we limit the scope of harming something to a human? What if the robot is asked to harm an animal? Or what if the robot is asked to harm another robot? What if the robot needs to harm a human in order to protect another human? So these are ethical questions that are not easy to answer. And they’re not easy even if they are answered. They are not that easy to be put into an algorithm. So don’t forget that eventually, all these rules of the framework, they need to be translated into a mathematical, let’s say, equation. Right? So this is a pretty tough thing to do. Even though it would be better for mankind to define these rules before and not after the fact, I don’t believe that this is an easy task. Companies are fighting for product dominance on a global level. Countries are fighting for military dominance on a global level again. So it would be hard to make rules that could be agreed by anyone and then be enforced by everyone because of just how society and mankind works. A more likely scenario would probably be that this set of acceptable rules are to be defined and enforced at least after the fact. I mean, after we saw that chemical warfare during the First World War was unethical code and code, we designed the framework around it. Right after we dropped the nuclear bomb, we saw that we needed to control it. So we might need to experience the first, let’s say, disaster in order to be forced to define the rules that would govern the robot’s behaviour. We just hope that this experience of the first disaster is not it’s not a really bad move. But personally, I see that even though there are tries of setting these rules up from the beginning, I believe that this is a very impractical task to solve.

If we, if we take a step back and see what the last an Algorithm is, alright? And try to see, to distinguish between what an algorithm can do for people, but also what people can do with the use of algorithms. So with a regular alchemist’s, these two, let’s say, tasks are pretty much distinguished. So what the algorithm can do for people, it’s one thing. What people can do with this algorithm, it’s a separate thing. So for a computer science to write an algorithm, he will ask you of the rules that you want that algorithm to obey too, right? So let’s say you want an algorithm that when given the volume in leaders, it will give you back the volume in gallons. Okay? So the software developer will ask for the rule that governs this relationship between leaders and guns. And he will use this rule to write this algorithm so everyone would know the outcome, what the outcome would be, provided that the algorithm complies to those rules and it works correctly. Okay? Now this is what I would call a regular algorithm. If now the use of this algorithm can get people in trouble, then it’s not the algorithm that isn’t fault, but it’s the use of the algorithm that is at fault. So let’s say, for instance, that this piece of software that we wrote is used in an airplane and there is a switch in the airplane that says, okay, when you turn the switch on, I can display you the fuel in gallons. When you switch down, I can display the fuel that the plane has in its tank in litres. And suppose now that the switch is in the wrong position and the PILER asks for gallons instead of litres and takes off. Well, chances are that he will have to make an emergency landing somewhere else before his original destination. So what do you do? In this case? You have an algorithm that is functioned correctly, but it’s use made him a function. So in this case, the regulator step in, the industry steps in and you train the pilots correctly. You make sure that the after checks if the plane has enough fuel before you take off. You put regulations to force the ground fueling service to make sure that the plane has enough fuel to reach so we know how to handle, let’s say, all the regulatory issues that we have to handle in a regular code unquote alchemist. Now, what happens with the AI algorithm? Well, let’s take for example, a machine learning algorithm. In this case, you design the algorithm with some basic parameters and you tell the algorithm what data you have. You just fit the algorithm bunch of data and then you run it. The difference with this type of algorithm is that the system changed itself based on the parameters and the data that you gave. So you cannot really predict the outcome of the algorithms. If you had enough time for a human to go through the data that you gave it to and go through the parameters and run the calculations line by line on a piece of paper, then yes, we would predict, let’s say we would make the same predictions that the algorithm did. But it’s impossible for a human to run through that large amount of data with the parameters that you gave the algorithm to work on. So at the end of the day, the result that comes out of an AI algorithm could be considered unpredictable. We could have a result that we didn’t expect. And that’s where the ethics within the outcomes need to come. Because imagine now that you’re the Ministry of Health of a country and you ask an AI company to come up with an outcome that looks at the demographics of the population and it looks at the number of available organ donors. And we want the algorithm to suggest who will get a heart transplant, for instance, and who will not get a heart transplant. So you put in things like age, things like probability of having, let’s say, high probability of accepting this trust plan, and you put in a bunch of parameters, and then you give all the countries demographic data to this algorithm and you let it run. So the ethical questions now are required to be answered by the people who wrote the algorithm and not by the people who just use it. Whereas in a regular algorithm, we can give the responsibility of making the decision to the people that use the software. We roll in the AI version of the machine learning version of the algorithm. Well, basically, the algorithm is making the decision for us. So that is why we need to start thinking about putting ethics rules embedded in the algorithms themselves. And as we said earlier, this is an easy thing to do, because how do you decode all these ethical answers into a mathematical equation? And this algorithmic term and space is a new space. It’s being researched now at an academy, at an academic level. It’s reaching, touching a little bit the industry, but it’s definitely not there yet. So he still needs time. Yeah. Thank you very much for this input about the relevance of algorithmics.

Now we’ll move to specific questions about artificial intelligence and social challenges. So could you tell us more about assisted, augmented and autonomous intelligence? Yes. Okay, well, as we said earlier, if we take a robot, for instance, that does a specific job right now, the robot that works in the car assembly line is not really considered as a robot with an AI algorithm in it. We’ve seen these robots working in the industry for many years now before even this new wave of AI methods were becoming popular. So what happens in that space? Well, that is a task, that is a repeatable task, that is a task that is easily recreatible and that is accepted by the industry as a harmless piece of equipment. Now, the advantages for this having such a machine is that, well, we remove this high intensive labour activity from humans and we let machines do the boring and the unhealthy tasks and we allow humans to work in more productive and safer environments. Now, the disadvantage is that, of course, OK, this robot that we have in that assembly line has basically taken over the prominent opportunity of the people that we are able and willing to do these tasks. So things are, let’s say, accepted as far as the use of a machine doing a repetitive task of manual labour within a specific industry. Now, the problem becomes more complicated now when you have a robot or a machine that autonomously makes decisions and functions on a more, let’s say, code and code intelligence manner. So imagine now that you don’t have an assembly line and you have a robot which has the intelligence to perform the tasks that the regular robot performs today, but also decides that the design of the actual car needs to change. So let’s say that the robot now has the ability to design the car in a more aerodynamic way, let’s say, and it decides on its own, autonomously, that the shape of the car needs to be changed. And of course the robot will have the ability now to make the design change and at the same time build the car. Okay, the advantage would be that perhaps they come up with a better design and offer better fuel efficiency or whatever. But the disadvantage is that first of all, we displaced some more jobs now from the workforce. We perhaps took out the designer now, but at the same time, we’re not sure if these changes that this new robot now suggests will actually be safe. So yes, we might have a better efficiency, fuel efficiency, but will it be safer or safe or as safe for the driver and the passengers? So, again, everything needs to have a balance. There are advantages, there are disadvantages and that’s why there are risks and threats and challenges. But at the same time, there are opportunities.

So you already mentioned some opportunities and some risks and threats. In your opinion, what are the most relevant social concerns? I think personally, what I’m really worried about is things that come to it affects the security of people. So things like the job market and employment. I think eventually mankind will find its way and if we are careful enough, we can find answers to these problems because technology will always be there. We’ve been through this again and I’m hopeful that we’ll find answer to those questions one way or another. When it comes to things like safety, human people’s safety is one of the hardest problems to solve. And that’s where I think we should be paying more attention to. For instance, what happens if we have a virus in a car and that car now drives my mother every day to the grocery store? What happens if that piece of software is taken by a terrorist organization? What happens if the software is at floor by design, it hasn’t been tested properly? And then something happens during this trip of my mother’s trip to the grocery store and then we have a crash. The same type of problem that affects the safety and security of people. Of course, it’s multiplied by 1000 when we get into the use of machines in warfare. So right now we have drones that are driven by a pile of joystick that is sitting on the other side of the world. And this drone now is being flown by this pilot. And the pilot makes the decision of where to drop its bonds. What happens now when this drone is given the okay to decide on its own? How do you pass these ethical questions to the drone’s software? It’s the same problems that warfare was challenged every time a new web phone becomes available to the masses. So these types of decisions that affect the security and the welfare of people, I think these are the major areas that need to be looked at. And it’s probably the hottest areas that we can finance to.

And you already presented some examples in different areas. Could you give us maybe a few more examples related to facial recognition, justice and social network challenges? Yeah, the facial recognition, it’s a problem that’s been found, found and raised repeatedly the past few years. You could say that it started directly as they need to identify solutions in the security space, but it turns out that now it’s all over the different industries, around the technology. Let’s say industry. Let’s say, for example, that we train a system to use a camera that is sitting at the mall’s entrance, and we want to see if the person that is getting into the mall has a COVID or not, right? So COVID 19 is a very current issue. And let’s say that we have some kind of system that looks at the face and if we found the algorithms that I can identify along with other sensors and data can identify if this person might have coffee. So let’s say that we do identify a person that has coffee. A good next step would be, okay, let’s see, let’s use AI facial recognition and let’s find who this person is. Let’s go to Facebook, let’s go to whatever other social platform, let’s identify this person and then let’s see the pictures that this person has posted in the different social media platforms and see who this person’s friends are. And perhaps we need to reach them in order to see if they got together and they could have. They are suspects now of a COVID-19 positive test result. So where do we stop this? Right? Where do we put an end to this? If we have rules that have to do with personal data and security of data, well, perhaps these rules need to be pushed aside because of a pandemic. So do we let the algorithms do all that work for us without any control, offer help? But the fact that it’s algorithms can make decisions on its own creates this uncertainty for us. And that’s why people rightfully are sceptical about it.

Could you now tell us a bit more about the status and perspective of some regulations? For example, OECD or EU UNESCO GPAI. Yeah, like I said, I mean there are initiatives and people, smart people, are spending time on placing a framework around the use of AI. There’s also research being done right now for areas of ethics that are not being tackled by government agencies. But as we said earlier, they are at a very early stage and I have a feeling that the industry will be once again ahead of the regulators. If you have a look at the research that is currently happening with companies in the areas of robotics, you will reach the same wow reaction when you see how robots are performing today. And I have a feeling that the industry is already ahead of the regulatory framework and it’s most likely going to be another scenario being repeated where something bad will need to happen in order for everybody to react positively. The way things are moving, I don’t see why should AI be treated in any different manner. I think the industry will be moving along at the incredibly fast rates and the regulators and governments will be a step behind. This is, I believe, is the current trend right now and I don’t see any major factor changing. Thank you very much. Is there anything you would like to add regarding this issue? No, I think pretty much the question of ethics within AI. One could say that it is a valid question. It is a question that is being raised by societies throughout the world. I also believe that the industry will go ahead and design and implement whatever it thinks it’s best for their creators. I don’t think there will be anyway of stopping that from happening. I just hope that at the end of the day, the results of these AI initiatives will be beneficial to mankind and not create all these issues that we are afraid of creating. Eventually mankind kind and humanity will find its way, but I think it’s going to be a bumpy road.

My name is Dr. Sabasana. I’m the head of experimental AI at home in Berlin, and my task is to conduct research towards the expendability of artificial intelligence.

And could you maybe tell us more about deep learning and what is the difference between machine learning and deep learning? Yes, deep learning again is a subtitle of machine learning and describes the use and training of machine learning algorithms which have a deep representation of information. And this usually describes deep neural networks. The depth in this deep learning is in the deep neural networks is that one usually stacks multiple layers of possible representation of data. You can imagine it as just layers of mathematical operations which are then learned in training. So you give the shape of the network and the function of the network is learned in an Iterative training process by providing an example data and the term deep and learning. The approach comes from the depth of the network. And what is the difference between machine learning, deep learning and traditional programming? Okay, as I said, deep learning is part of machine learning. The difference between machine learning and traditional programming is the following so consider you have some data and you know the rules how to process this data, right? Then you can implement your solutions. And this is a typical programming approach. You have your data, you know how to process it, you implement your programs and outcomes and answer. The approach to machine learning is thus you have a lot of data and you know the answers to this data, but you have no idea how to come up. Basically, you don’t have the rules, right? The task of machine learning is to train your machine to learn the rules which allow you to connect the data and produce the expected answers. And once you have this, you have a trained machine learning model which can receive new data, the data it has never seen before because it has learned the rules and should not have learned the data by heart and it can then produce answers. So in machine learning speed, we say the model should generalize, which means it should have learned general rules on how to handle this data you put in to provide the correct answers. Once you have such a model, you can plug it in as a set of rules in your programming task. For example, if the set of rules would be so complex that you never would explicitly be able to COVID it by writing the code manually.

And can you provide one or more examples of popular use of machine learning? I think one example which is used fairly often is optical character recognition which means the machines and the post office which read the target address of your letter you write this is usually not done by humans but it’s just sensory machine. The machine deciphers your handwriting then digitizes the address and feeds all this information into a database and then the letter gets directed to the target. Another approach would be facial recognition, for example. For example in digital video cameras, webcams also valiant systems. So the spectrum of applications in machine learning is quite useful. For example, what we are doing in our lab is we use machine learning for natural disaster prevention for example, where we track where we have, for example climate data or air pollution data of the last years, months and so on. And then we train a model which should be able to predict how temperature arrival and so on behaves. Given a lot of factors which cooker over the last days months could you provide one or few examples of popular use of deep learning? Pretty much everything which is quite complex and was apparently unsolvable about ten years ago we licensed deep learning and that is image recognition, for example uses deep learning because using the step of the deep networks this deep architecture allows the model to learn like cascade of different feature processing steps. Right? As a matter of fact deep neural networks are somewhat motivated by the visual cortex of the human brain which processes information in several steps beginning from just receiving color information to basically neurons triggering to simple shapes like edges and round shapes and so on neural networks actually do quite similar things. And by going from the most atomic to very complex features for example, from edges color gradients to neurons which are there or have learned to recognize the heads of lizards, for example this complex image information can be processed efficiently and quite fast. And this leads to current machine learning models in image recognition and outperforming humans for example especially if you factor in time.

And what are the opportunities and the positive aspects of machine learning for society? For one, it’s the potential to reach a state of automation which strips away tasks which are labour-intensive but boring and no one actually should do because it can be automated nicely. This, of course, increases efficiency. It reduces errors because the machine never gets tired. In a medical setting, for example, machines could be used to augment the decisions of trainee. Mr. Pathologist, for example. Mr. Pathologist is an especially interesting domain here because it is known that a historian has its highest value when he’s going into retirement because he has a lifelong basic period of learning behind himself or herself. And those almost retiree his are very much faster than the new guys who need to learn the trade, right? And by faster, I mean they intuitively look at one of those histopathology slides and immediately see what’s going on, why the newbie needs to scan every bit of the slides meticulously and take time and so on and so on. And there’s also a group in class acting from Andreas Holzinger. He’s actually training machine learning methods based on data annotations made by an expert, Mr. Pathologist, with the goal of encapsulating his life experience in his topology into machine learning model. So it can potentially be used as a training companion for starters in this domain.

And what are the most relevant risks related to ethics, for example, in your opinion? For one, of course, is the intended use case of machine learning. For example, do you want to use it for the general good? Do you want to improve of society? Do you want to improve environment? Or do you want to plug it into a cruise missile? This is the core difference here. And the next thing is these are the extreme ends of the spectrum. And then there’s a plethora of societal issues in between. For example, do you automate your credit worthiness estimation of a person and use machine learning for that? And then there’s a question what data did you use to train this model? And did you maybe model, maybe train unwanted correlations between some features and data and the outcome? For example, we had learned that some ethnicities, for whatever reason, for example, skin color, ethnicity, as I said, should not receive, I don’t know, financial aid because of this reason, right? The question is always what data do you feed in? What data do you want someone to use? There’s this principle of data spacity which means only use the data you need to solve the task because additional data might create confronting behaviour in the model. And this is of course one of the current big issues with the ongoing automation. With machine learning. On the other hand, there’s always a question if you use real data to train the machine learning models and don’t like what the model is doing because the model itself is objective, the trust can learn from the data you provide. The data is the model’s sole reality. Does it mean that you don’t like what the model is doing? Or do you not like reality? Right. And I’m thinking it might not always be the right way to fix and curate your data to get rid of certain behaviours of the model. I would see it as an indicator for change necessary in society producing this.

One question regarding Extendable artificial intelligence X-A-Y can you explain what it is like? Yes. So the target of Expandable AI is to shadow into the black box of machine learning. So current machine learning is so usually the best performing machine learning models are quite complex, which means the outside observer really even the developer really has an insight on what the model is actually learning. And with accessibility or X AI, we aim to gain back some transparency on what the model is doing. Yeah, this can be done in several ways. What we did in our lab is we developed a modified backdrop method which means if you feed in some data points into the model, the model gets transformed layer by layer to the network or the model. Basically it traverses through the model in the end and the end results in answer of the model. Right. And we can somewhat invert this process by for example, if the model receives a picture and tells me it’s a cat, I can start with the cat output and say yeah, but why? And then I can pick apart the partial decisions of the models layer by layer until I reach the input again and I can then obtain, we call it a heat map correctly. It’s basically been masking in the input space where the more things that information is and you can do this for any potential outcome. For example, if you have a dog output of the model and then you can do the same process with the dog output and then you might receive the answer why the model thinks there is no dog in the image or where the dog information is not right. This is a way to connect the model’s use of information as given by the data points to the model’s output. Yes, expandability is a quite young field.I would say the earliest serious steps and more complex models have been done in the 2010 and since then it has been evolving quite rapidly. So there’s a lot of work going on. We are working towards providing explanations which go beyond simple heat map visualizations which need a lot of interpretation at times, especially if the data is hard to understand and need to acknowledge. But our end goal is to model under the treatment of improved exploitability which you’re currently working on should be more or less self-explanatory right by not saying look at this part of the image there’s information which I think as a model speaks for cat. But the model should then inform the user. For example, that I think there’s a cat because I see this and that and that and that. Cat like features the model has, for example, learned as features to use during prediction making. Thank you very much. And what does explainable AI make possible? What can you achieve through that? For one, you can understand what the model is actually doing and you can gain understanding on a per sample basis per sample basis means in this case for each data point you put into the model you get receive feedback on the model reasoning based on this data. You can then of course use this to verify your model. But in some cases you might also end up with the information that the model is producing the right output for the wrong reasons. For example. And this might point you at faults in your training data where you have introduced some confounding information, some confounding features which the model then connects to the output of cats but which are absolutely not Cadillac just because it’s easier for the model. And then we have the problem again that the training gate of the model is the model’s sole reality. You just give it a couple of thousand images of cats and the model learns how to get from this data source to the cat. And if those images, for example, have been crowd from Flicker and they all have a copyright watermark because they are stock images or something, the model might pick up that stock images are cats. Right? This is the problem one of the problems we can identify with externality and this then to the option to improve the model, improve the data source and so on so we are basically way more informed machine learning developers than before extending.

Well, hello. My name is Adriana. I’m working at the university medical center here in Ljubljana in Slovenia. My role is more or less in the field of informatics. So, I am employed in the department of the informatics and we are responsible for the introduction for the new It technologies into the hospital.

What are the challenges or the risks or maybe the threats of artificial intelligence in the field of health care at the moment, as we see, the biggest challenge is the answer to the question does it pay off? The introduction of different systems and new technologies is namely very expensive. It is especially expensive for small countries because you have calculated per capita per number of persons. So for us, new technologies are bloody expensive. On the European level that would be much easier. It would be much easier to COVID the expenses at the European level because then the cost per person or the cost per treatment would be much lower. So this is one of the first challenges. The technology is very expensive. On the other hand, you have to know whenever the hospital managers decide about the introduction of new technology, they make like a short calculation is it better than the existing procedure? And if it is better, does it pay off? Is it better for 20%, for 50%, or only for 0.5%? If the improvement is very low and very small, then this new technology will probably not be introduced into the hospital because there are many costs that are also including the education of the staff that are included in these costs. So for me, the most important challenge is the costs of this new technology. For some people, for example, challenge and risk at the same time is represented by the fact of the nature of the data. Data can be easily and quickly transferred. The owner of the data, which is the patient, the owner of the data doesn’t have a full control of his own data because data manager, which in our case is the hospital, could transfer the patient’s data to other institutions. So or to share it within the European Union. For some people this is like a challenge under risk. But for me it is also a kind of advantage. We have seen in the case of Pandemics that the national healthcare systems should be much closer connected and that the authorities, like the European or national authorities, should have a good insight what is happening with a certain disease in a certain population. So personally, I’m not against the sharing of data in order to preserve the good healthcare of the nation. I don’t see many threats. Some people see threats. But you see, I work in the It department, so we embrace the technology. And again, I must stipulate that some norms and some guidelines that European Union has introduced are very well accepted. The European Union is actually trying to reduce the threats in the use of AI in healthcare. And from my point of view, this is like the requirement that the data managers must tell the patients which algorithms they use. And for patients this will be, from the very beginning kind of difficult to understand. And perhaps we shouldn’t expect from all the patients that they will understand everything that is related to artificial intelligence. But at least I think it would be very good to educate the patients and the Europeans that their data is their data and they are in control. What will be done with their data in the future?

How can AI artificial intelligence help in early detection of diseases? Well, the early detection of diseases is probably one of the most important features of the artificial intelligence. The reason for this is that the monitoring of the development or following the signs of the development of the disease cannot always be done by the doctor or by the certified nurse. So if we have artificial intelligence in place, together with different visual components that can actually follow the patient or the European citizen, then it would be easy to detect the early onset of the disease. Now, why is this important, we can say yes, you will get your diagnosis six months earlier than if you would have been waiting for the doctor. How does it benefit you? Well, if your diagnosis is set early, you can start your treatment earlier and there will be less complications. You will be younger and the younger body debuffs the illness much better than the older body. So if we catch the disease in its early onset and the patient is still relatively young, the prospects for the development are much better than in case when we wait until it is already almost too late. And how can artificial intelligence help in treatment and authority prescriptions? I think this is like one of the best fields where we can expect the benefits, namely treatment and following the therapy might be boring and dull. On the other hand, it is very important that the medicines are prescribed and given to the patient in the exact prescribed manner. So mistakes here can be very costly. And artificial intelligence is actually very good at detecting the mistakes or the processes and actions that are not exactly the average for treatment of a certain disease. So if you have like a fisher intelligent intelligence system that follows the state of the patient and his improval of his disease, and then you can adjust automatically the dosage and the type of the medicine that had been prescribed. Again, it would be very good for the patients to realize that a certain drug is not really good for him or her and that certain adverse effects occurred sometimes. Adverse effects, some people call them undesired effects, they can be very small and cannot be seen by a naked eye. Therefore it is very good to have like this artificial intelligence system that is actually working on the kind of alert systems and it says right now, this and that parameters are not doing well with your patients. Go and adjust the treatment in this and that direction. Then, of course, the doctor goes and sees the proposal from the artificial intelligence and he or she decides whether he or she will follow the advice from the artificial intelligence. From my point of view, this is actually really improving the quality of treatment of the individual patients. How can absolute intelligence help in education and training of staff at all levels? Education and training of staff is really highly organized and very hierarchical, which means that all nation systems have their own ways how they will educate the doctors, how they will give them their certificate or licenses. And of course, for example, in European Union we also have a system of, I think it is called mutual acknowledge of the certificates and licenses. So in this basic sense I am not aware that the artificial intelligence would really help to form a young doctor. However, once the doctors are in their working environment they can learn from certain functionalities from the artificial intelligence and most often that would mean that a strong visual component is included and that the doctors are actually in their certain field they can improve their knowledge in a certain field.

So my name is Kelly Zone. I’m a senior researcher at the Norwegian Computing Center, which is a research institute working on AI and statistical modeling and generally computer science. And my field of research is AI, and inparticular, everything that has to do with language, so called language technologies or natural language processing.

Now we will talk a bit more about machine learning. So could you tell us what is machine learning and how does it work? So, machine learning is a branch of AI, but which has become known as the dominant branch of artificial intelligence. And it’s really about building models fora particular task where things are not preprogramed, but are learned from data. So instead of writing down, specifying detailed rules about what the system should do in every situation, you provide some data, training data, because you’re trying to make the system learn from those. And the system will then automatically identify some patterns that are useful for the task at hand and automatically then use it on new data once it’s learned. Could you tell us how is machine learning related to artificial intelligence and also to big data? Yes. So when it comes to the relation to AI, it really has become the dominant way of solving complex AI tasks. Machine learning is to do that on large training sets, and that’s how it relates to big data, in the sense that you are learning model’s toper form different tasks using large data sets. So it’s also really directly related to big data.

And so you mentioned already machine translation. Can you provide maybe one or more examples of popular use of machine learning? Yes. So machine learning search, like in Google search, search engines, computer vision, speech recognition. It’s used there basically for any kind of classification or prediction. And most companies nowadays have some kind of system. There automate some of their decision or predictions. In robotics, it’s always first also a field where you have a lot of machine learning models for helping the robots decide what they should do and how they should do it. In the industry, machine learning and deep learning are so heavily used. Before, it was, you know, statistical models and machine learning models for the development of this. You mentioned already some advantages and opportunities for machine learning for society. Can you also mention maybe some more? One of the main advantages is probably that of automation. It’s automating tasks that may be repetitive and interesting yeah, something I forgot maybe to mention. But also, everything that you do with autonomous, semi-autonomous cars. It’s also another example where much of driving is repetitive, is routine. It’s about looking for patterns, basically. And computers are quite good at recognizing patterns and doing that systematically.24/7 while human drivers are notoriously bad at keeping track of what’s happening on the road for hours and hours on end. That’s another case where automating a process that is more or less routine and repetitive in most cases can be heavily beneficial. Of course, it’s a good case in point because much of what’s happening on the road is repetitive. But sometimes you have a complex situation, you have something that is unexpected that happens on the road. And machine learning is based on historical data. It’s learning from what has been seen in the past and trying to generalize from it. But the capacity to generalize is much weaker than what we have as human individuals. So while we can understand quickly a new situation that unfolds, it’s much harder for computer.

What will be the evolution of human mission interaction and what are the emerging trends? Well, one important trend is that early human machine interaction was really about the human adapting themselves to the kind of constraints of the interface they were dealing with. In the 70s you had very complete machines with a lot of buttons and the point was that the human had to learn how to operate the interface and the interface itself had zero adaptation. And nowadays the trend is really as much as possible to have the interface the machine trying to adapt to the human user and trying to talk their language. And what interests me as research in language technology is to have machines that talk a language that are able to interact in the communication medium that is most intuitive for us human beings which is natural spoken language because that’s what we’ve spent most of our life doing talking and listening to each other in spoken language. This is the most powerful one because we can express ideas and thoughts and basically anything in a completely effortless manner in a way that would be impossible with a rigid interface where we had to click on buttons. But of course it’s also hard because natural language is ambiguous, is vague, it’s sometimes contradictory and uncertain. It’s always a lot of interpretation going on in order to understand each other. But it’s all interesting because having a machine that understands spoken language means it’s a machine for which you don’t need any training. You know how to speak your mother tongue and it’s both very useful and user friendly. And in some cases if you think, for instance, that driving a car or cooking when your hands are in the middle of your cooking preparation, it’s also one of the only ways just to have a concrete interaction with the machine because you cannot expect the user in anonymous car to spend the time looking at the screen and touching buttons. So in many cases the future of human robot human machine interaction is really having systems that can interact with us in a way that is familiar to us either through language or through some visual language that is easy to understand. And as the last question so you mentioned already some of the opportunities and some of the challenges. Can you think of more positive effects and also some challenges of human mission interaction for society? Well, there’s always the thinking specifically of language technology. There is a challenge of unequal access and unequal resources and that goes beyond human machine interaction. But an important problem for much of today’s language technologies is, for instance, that languages have not the same degree of support, and it makes a big difference if you’re a speaker of American English or if you’re a speaker of some remote African languages that has no access. Where Google Translate does not work, you have no possibility to use the technology in any way. Sometimes you have not even a keyboard on your cell phone to express yourself in your mother tongue. And that’s, in the future, is an important challenge. Not to speak, of course, access to technology. I mean, if you don’t have a phone to begin with, you have even less of an access. But all of this how to ensure that this technology is actually providing and use as an equalizer instead of existing inequalities in the world, that’s an important challenge.