AI technologies – AI & ethics

QUESTION (F. ANANASSO)

Let’s start with the most basic question. Given your experience, Emanuela, what do you know about global strategies in the field of artificial intelligence? What is the context? What are the international models? I hear particularly in America. What is the international context?  

ANWER (E. GIRARDI)

Yes, thank you Fulvio, and good morning everyone. Certainly right now, let’s say that artificial intelligence technology is considered like a strategic technology for the development of future societies. And a number of Countries have started addressing artificial intelligence, investing a lot of resources and – above all – developing a national strategy for artificial intelligence. At the moment, there are about 50 national strategies for artificial intelligence around the world and some are in development, so they will be published in the next few years or at even this year, and so all the Countries have seen these technologies as really strategic because of the disruptive aspect AI has on society, already today on our everyday life and more and more on our life in the future. InI particular, though, let’s say that although there are 50 different strategies already worldwide, we can basically speak of two-three development models for artificial intelligence globally, i.e. the American model, the Chinese model and, in between, the European model has recently entered. At first glance, the American and Chinese models, appear rather different, but there are indeed great similarities, because in both models, in reality, the development of artificial intelligence technologies is led by large private groups that have, among other things, a presence all over the world — not only in China and the United States. The two big differences between these groups are that while in China we have a tendency towards government-centric control of these technologies, in the United States there is a kind of tendency instead towards deregulation, hence a kind of lack of regulation of these technologies. Europe started a bit late in terms of investment in the development and adoption of these technologies, and in particular it started in 2016 with an agreement in 2018 ‘married’ with a cooperation agreement between all member states for the development of a coordinated plan for the development of artificial intelligence technologies. Starting a bit late, it decided, however, to focus on the ethical vision and therefore primarily on the definition of ethical guidelines for the development of artificial intelligence technologies. And from this research group that the European Commission convened, the European vision of artificial intelligence came up, which is a vision that is defined as ‘human-centric’, i.e. putting humans at the centre and promoting the use of artificial intelligence technologies to improve people’s lives. The second aspect, however, is that it promotes the use of ‘trustworthy’ technologies, i.e. they are reliable. This is somewhat at the heart of the European vision. Trustworthy because if we think of a very simplified definition of artificial intelligence technologies, where and with which we can identify systems that analyse their surroundings, i.e. collect data in order to achieve a certain objective and do so autonomously, by displaying a behaviour that if it were carried out by a human being, we could define as intelligent behaviour. Now, in order for them, when they reach this goal, perform some action that is defined by human being, for the human being to accept the outcome of this action or decision – which is taken by an autonomous system – and therefore trust both the use of these systems and the actions or decisions that are taken by these systems, they must be reliable (‘trustworthy’). And what does that mean, reliable / trustworthy? Let’s say that as far as the European version and definition of reliable technology is concerned, it means three things. Basically, it means that they must comply with European laws, with the ethical values that are included in the European Charter of Rights and they must be safe from a technical robustness point of view, they should not harm people. So, if these systems meet the three requirements, are ethically sound, then the European Commission says that they can be developed and used within the European Community. Then the European Commission went ahead and developed a set of two important documents a white paper, a European data strategy, but actually the document at the moment that is more controversial but also more important is the one that was presented in April 2021 and is the so-called AI Act,  which is a set of documents that are the first proposal in the world to regulate the use of artificial intelligence systems. This document is very important because first of all it uses a risk-based approach, i.e. it divides all artificial intelligence systems into four categories based on risk. The first one is considered the unacceptable risk ones, so these are the systems that are banned within the European Community. Then there are high-risk, medium-risk and low-risk systems. With regard to high and medium risk systems, a whole series of requirements are defined, the so-called assessment lists, that is, a series of requirements or certifications that these systems must have in order to be used within the single market of the European Community. So these aspects are very important and there is a lot of debate within the EU Commission, since on the one hand, it is said that the objective of the European Commission is to encourage the development and use of artificial intelligence technologies that in some way certainly protect people and protect European citizens, but at the same time that there is a kind of legal certainty for companies to be able to develop these systems in a safe environment from the legislative point of view, so as not to have risks when they introduce artificial intelligence technologies in their products and services, in their market offer. Clearly, the biggest limitation being discussed at the moment is that over-regulation could in some way restrict innovation. And this is especially true if we consider the two systems we talked about earlier, the Chinese one and the American one, where actually there is much less regulation, and especially the American one compared to the European market that we are trying to create. And so it is very difficult on the one hand to balance the need to regulate these systems, which if they are used maliciously can actually cause damage, and on the other hand also to try to encourage the development of European innovation and to ensure that there is also a real development of artificial intelligence technologies, but not only artificial intelligence, also in fact all the supporting technologies. So I’m thinking of blockchain, I’m thinking of cloud technologies, I’m thinking of performance computing because it’s actually very important that these technologies can also be developed at European level and that what is called European technological sovereignty in the coordinated plan is therefore realised. Let’s also think, for example, to the Gaia X project, that aims at  the same objective,  that of achieving a kind of European technological sovereignty. 

QUESTION (F. ANANASSO)

So, if I understand correctly, beyond the American model and the Chinese model that with different aspects go in a different direction than ours, the European Commission one  is more human-centric and “trustworthy”, as you said. Then afterwards we’ll talk maybe about what’s going on in Italy. If I understand correctly ,the AI Act you’re talking about (the April 21, 2021 one) put various levels of risks. Is there anything you want to tell us on that, what are some relevant points? Besides the risk classification of the European proposal’s, are there relevant aspects? I’ve heard of 32 cases where one can intervene subjected to the fact that there are warrants of arrest. Is there something that can be said quickly about what are the essential points of the European proposal?

ANSWER (E. GIRARDI)

Yes. Actually, as far as the risks are concerned, what I can tell you are the aspects of artificial intelligence systems whose risk is considered unacceptable, for example social scoring, people remote biometric monitoring or systems that can somehow influence or manipulate the behaviour of people — especially vulnerable people. These features are considered unacceptable and are therefore prohibited at the moment within the European Community. In terms of high-risk systems, instead, in my opinion at the moment there are aspects still to be clarified, because we are talking about conformity assessment, a list of a set of criteria that will have to be met in order to be introduced and used within the European Community. But the problem is that there are also talks of a European certification, right now being not very clear. Among other things, we have to keep in mind that the value chain of artificial intelligence systems is extremely complex. If we think about the development of an artificial intelligence algorithm it goes, for example, from those who collect the data, those who do the training, those who develop the algorithm, those who then put it into use in their own systems within their own company and/or put it into the market, and then maybe the dataset gets changed — a re-update is needed and then one might need to change the system and in that case  monitor everything through use. We are really talking about the whole life cycle, the artificial intelligence system and then it makes all the different players in the supply chain constantly stay connected and constantly monitor the use of the artificial intelligence system that they have developed, trained and put into the market. So these aspects are still not well clarified because there is an aspect of supply chain that is very complex in my opinion. 

Another complex aspect concerns the fact that at the moment high-risk systems are defined, not in a very clear way, but they are defined nonetheless. But another aspect concerns the fact that the European Commission at the moment has, let’s say, ascribed to itself the right to change this list and then adding new systems, which is quite controversial because normally there should always be a division between the legislative power and the executive power, so in fact there are already many discussions about this, saying that it is not the European Commission that has the power to change an annex of a law. And so these aspects are quite controversial, for example I was reading the other day that the various CEN, CENELEC, … and most of the various standardization Bodies asked the reason for the above EU Commission position. 

There is also a very interesting aspect regarding just the definition of AI standards, which at the moment are still missing at the European level and so all the various standardization bodies are working on that. It’s one of the feedbacks they have requested since these days everybody is writing feedback to the European Commission, because 6 August 2021 was the last day to be able to submit feedbacks. A proposal for regulation is one of the feedbacks, which is just to delete Article 41, that is the fact that the European Commission can vary, let’s say, more or less as it likes (that’s not exactly the case, but it can vary discretionarily) this list of systems. So these are some of let’s say the most conflicting elements.

Another very conflicting element concerns the AI definition, since it is a very broad definition whereas it is actually supposed to be, let’s say, future-oriented and future proof. But from the other point of view at the moment systems, that I don’t know if they are AI, are also considered as AI systems, like advanced statistics systems, linear regression systems … that then make it – let’s say – much more complex the use of these systems that probably today are not artificial intelligence, so there are so many aspects of discussion. I think it will take two or three years before this regulation is applied in all member states.

QUESTION (F. ANANASSO)

Excellent, excellent, and without mentioning the newly established Institute for Artificial Intelligence in Turin, which has had initial funding, am I wrong? 

ANSWER (E. GIRARDI)

No, no, unfortunately you are wrong, It actually went a little differently, in the sense that yes, it was supposed to be in Turin, but now there is a bit of uncertainty in reality, since this decision, that had been taken by the previous government, was not followed up and then after various discussions, what was decided to do is a centre for automotive research and development that will be based in Turin and will also use artificial intelligence technologies, with funding, if I remember correctly, of 20 million euros. Therefore, at this stage, the Italian Institute for Artificial Intelligence has been put let’s say on hold, which in my opinion is a great pity because that would have allowed Italy to participate in the various international tables, let’s say with an institutional hat, because today, unfortunately, Italy is present at many international tables, but is represented only by individuals who are the excellences, not at institutional level, so it is, let’s say , a missed opportunity. 

Let us hope that this new task force will be able to propose it again.

QUESTION (F. ANANASSO)

Let’s say it’s a half a victory, let’s see it in a positive light. And look, we have seen the World, Europe and Italy, now, to conclude, there is a lot of need for training, obviously, because everyone is talking about artificial intelligence, that thing that a few really understand completely if one doesn’t go into it in some detail. Thus, I think we agree that training is crucial to understand what these technologies are and how to use them safely. Is there even the risk that without proper training – I am talking about the Italian system, since we are in this project of training for adults in artificial intelligence – there could be a sort of “digital divide” in the sector of artificial intelligence, compared to others? How do you see it, what is the importance and what do you think are the important things that should be done precisely to teach well and better the potential, pros and cons,  risks and whatever of artificial intelligence? 

ANSWER (E. GIRARDI)

In my opinion this is the key aspect and  the most important. Just think that at the beginning we were talking about over 50 countries that have already defined their national artificial intelligence strategy. Think that there are countries like Finland that have put as the number one priority of their artificial intelligence strategy the training, the education of their citizens in artificial intelligence technologies, so it’s really a key aspect. Also China, by the way, is investing so much precisely in the education and training of its citizens, since they  realised that this, let’s say, is the key to being able to actively participate to the society of the future. So in my opinion it is precisely the key aspect. That’s because if we think about us as workers or as citizens or as students, for example, then if we put on the workers’ shoes, the jobs of the future will certainly require the ability to use artificial intelligence technologies, and so this aspect is fundamental, training or upskilling is necessary to learn let’s say new job profiles or re-training or reskilling to really learning about totally new jobs. 

And this is, in my opinion, a key aspect, because if we think of a company that is introducing at this time perhaps robotic automation systems of processes and then somehow freeing up resources because there were tasks carried out by people that are now carried out by automated systems, therefore this time that is freed-up, these freed-up resources should, in my opinion, absolutely be reinvested in training for employees. This aspect is also important from an economic point of view, because it also means a redistribution of what are also the economic benefits of the introduction of automation systems in companies and therefore it is also a redistribution, let’s say, of what are the beneficial effects among the various stakeholders, i.e. investing in training, which should, in my opinion, be a right of all workers. 

The other aspect is that on these things it is really important to learn what are, let’s say, the digital skills, not only more basic, but advanced, to be able to know what the artificial intelligence technologies are and to be able to use them for one’s job, because all jobs will also use these technologies no matter what. In this regard it is very interesting, for example, an approach that concerns more, let’s say, the education, so the schools, like a very nice project developed at MIT in Boston, together with a project funded by Schwartzman, a private equity fund by Mr Schwartzman himself, founding a college called the Schwartzman College, which is a kind of bilingual school. That is, whatever one decides to study – philosophy, anthropology, medicine, law, … – he/she studies along with computational science and artificial intelligence, because the idea is that everybody in the future will need, no matter what, a solid base of computational science and the ability to use artificial intelligence technologies. These subjects should also be included as compulsory in our school, really from the time we are children, i.e. primary school, subjects which in theory are already there, but in reality, unfortunately, they are not taught  in depth. And so perhaps the most important thing is the training of trainers, so the project you are carrying out is also very important in my opinion, since it aims first to train trainers so that they can then clearly go and teach these subjects in schools. It is very important because if we think about, for example, the doctors who are now in hospitals, they are probably not able to use the new artificial intelligence systems and even the doctors who are now at university, who are not studying these technologies, when they go to work in hospitals will not be able to use them, and we have seen how useful these technologies were during the pandemic Covid-19, so if we could really train them to be able to use these technologies, it could really be a very useful tool in their work. 

And then there’s the other aspect to consider, that the jobs of the future will be different from presently, that is, the doctor of today will be very different from the doctor of tomorrow. Thus they really have to change as well the approach both to learn how to use these technologies and to have a mentality, a more fluid, more open mindset to be able to tackle even new jobs that probably don’t exist today and that will change a lot in their contents.