The technology news was awash with one topic in 2023: ChatGPT. The prototype of chatbot which uses artificial intelligence to formulate responses, surprises with its ability to create texts that fit into different situations. From essays with the potential to pass exams, such as the first phase of the Brazilian Bar Association (OAB) test, to the drop in the value of shares of companies that invest in similar technologies responsible for making mistakes when giving answers during market-oriented presentations , the news brought about by artificial intelligence technologies is no longer of interest only to researchers and professionals in the field and fuels euphoric and terrifying speculations about their effects on the future.
“It is very difficult to pinpoint what should happen. I think we need to wait a little to be more sensible and have a clear idea of its effects.” The evaluation is by Ofer Arieli, professor at the Academic College of Tel-Aviv, in Israel, and one of the guests at the São Paulo School of Advanced Science in Contemporary Logic, Rationality and Information – SP LogIC. PhD in Computer Science from Tel-Aviv University, he works with applications of logic in artificial intelligence systems and states that the biggest challenge in the area is to develop systems capable of making their lines of reasoning explicit. “This is one of the most prominent themes in research in the area,” he comments.
In the interview given to Journal of Unicamp, carried out with the support of Walter Carnielli, professor at Unicamp and member of the advisory board of SP LogIC, Arieli talks about the contributions of the so-called theory of argumentation, the topic of his course at the School, he comments on the diffusion of logic among new generations of researchers and addresses the importance of the topic for the challenges of the contemporary world.
Unicamp Newspaper - How important is logic for artificial intelligence? Is it applied only to the development of systems or are there ethical and philosophical issues in which logic is involved?
Offer Arieli: One of the main goals of artificial intelligence (AI) is to imitate or follow the common sense reasoning model. In this sense, logic has a primary function in shaping this reasoning. This is the essence of artificial intelligence, the ability to understand, reason and imitate the way we think, make decisions, etc. Basically, this is the main role of logic in artificial intelligence. Of course, there are several other roles, such as the philosophical and mathematical roles, and all of these parts are involved in automated reasoning. For example, how do we revise our thoughts, or how do we draw conclusions from inconsistent or incomplete information? These are all everyday tasks. So, if we want to develop good artificial intelligence systems, which are rational, we need to be based on some type of logic. There are now other disciplines based on machine learning, but one of the problems with these systems is that they are not able to justify their conclusions. On the other hand, systems based on logic are capable of justifying them. That's what I work with, developing systems capable of drawing conclusions the way humans do and that are also capable of justifying their conclusions.
JU - In your course at SP LogIC, you covered the theory of argumentation. Could you explain what ideas are involved in this theory? Does it connect, in any way, with the concerns you point out regarding AI systems?
Offer Arieli: The theory of argumentation encompasses patterns with which we organize our debates, dialogues, persuasion strategies and also defeasible reasoning, using arguments and counter-arguments. Traditionally, I work in the context of paraconsistent logic, which means working with systems that can reason about inconsistent information. In this context, I consider two things necessary: first, most paraconsistent logic systems are monotonic, that is, as you extend your premises, you necessarily also extend your conclusions. This property is very useful in the mathematical context, because, if you arrive at certain proofs, you can use them to expand your knowledge. However, in common sense reasoning, this does not apply, because once a fact is learned, this knowledge would not be revised. In this sense, argumentation offers a good perspective for this type of reasoning, because it involves arguments, counter-arguments and new information. For me, it is a good theoretical perspective, which allows, on the one hand, to deal with inconsistencies and, on the other, to be able to review information and change thinking.
Another advantage of argumentation theory is that it is graph-based, which gives us a way to check what is happening. With it, you can visualize, to a certain extent, the cores of inconsistency. This is not a purely abstract method. That's a good thing, it's not just logic. It also combines other means to reach rational conclusions, such as graphs and other resources.
JU - Do you notice a growth in the study of paraconsistency in Israel? Are there influences from the Brazilian school of logic?
Offer Arieli: The Brazilian school of paraconsistency is, perhaps, one of the most prestigious in the world. Not only is one of the school's founders, professor Newton da Costa, Brazilian, many contemporary researchers are also Brazilian, such as professors (Walter) Canielli, Ítala D'Ottaviano, João Marcos and many others. Many Brazilians are involved in this area and this is very inspiring. In Israel, there are also spaces where logic is carried forward, but in a more modest way. Sure, we're a small country, but still, there are a number of researchers investigating paraconsistency and logic more broadly. Even though today it is a little difficult to attract researchers to this area of logic. Computer scientists are most interested in machine learning applications, the big field at the moment. So it's more difficult to convince students to work with us, because it requires a lot of study, effort, and we don't always see the practical results of this. In addition to being less profitable.
JU - Is this because it is a more abstract area?
Offer Arieli: Exactly. Still, we managed to work well. We have talented students who will be the next generation of Israeli logicians. It is a vibrant area, with many published works. I think there is great potential for growth, because inconsistencies are everywhere and they need to be dealt with rationally. I'm sure that paraconsistency will always be related to relevant issues, and will be for a long time.
JU - In our interview with professor Fernando Zalamea, also a guest at SP LogIC, he highlights the importance of logic to establish connections with different areas of science. You mentioned the difficulty of engaging students in this area. Do you believe that this is also due to a possible difficulty in bringing together different knowledge in the same area?
Offer Arieli: Yes, certainly. I can say, from my own experience, that logic is an interdisciplinary field. I came from the computing field, but today I work with more researchers from philosophy departments than from computing itself. Logic is at the heart of many disciplines. I think it really is a bridge between them. To carry out relevant research today it is necessary to bring together people from different disciplines, such as philosophy, mathematics, information sciences, and I believe that logic, as the heart of mathematics, is the bridge that can unite researchers from different areas.
JU - Have you been following the news about new artificial intelligence technologies, especially ChatGPT? It seems like people are very concerned about the implications of these features. What is your perspective on this scenario?
Offer Arieli: I think it's a type of evolution, something that's been happening for a few years. What we didn't expect is the impact caused by the software. There are several implications, not only in mathematics and artificial intelligence, but also in literature, in the arts, in everything. For me it's a mystery what will happen, because we really don't know. There are even researchers who believe that resources like ChatGPT can fulfill computer science functions itself, so the area can also change. Our department is holding many discussions about this, not only among mathematicians, but among professionals in the field and other people who do not know exactly what should happen or what risks and challenges it brings. Everything happens so quickly, from one week to the next progress is rapid. It is very difficult to pinpoint what should happen. I think we need to wait a little to be more sensible and have a clear idea of its effects, to get to the essence of the software and its implications. One thing that worries me is that we can't explain how ChatGPT reaches the conclusions it presents. This is a little scary, because when you can't trace the line by which the system reached a conclusion, identify the way in which the reasoning occurred, we can't predict what will happen in the future. I don't think I have a good answer at the moment, because we are at a very vague stage. I've always worked with things in which I could justify the conclusions of my studies, but this is something new and I don't know exactly how to deal with it.
JU - Does this mean that this is a system that does not present the sources on which it was based to deliver the results, does not explain its line of reasoning?
Offer Arieli: Yes.
JU - Is this a point where logic can help?
Offer Arieli: Yes. This is one of the reasons why the development of explainable artificial intelligence is one of the most prominent research topics in the area. This is because it is necessary to have AI systems that justify their results. Logic is one of the solutions, but, of course, there are also technology-specific problems. The main one is efficiency, because creating systems that justify themselves takes a lot of time and a lot of computational effort. This is the main problem and the main reason why people try to create systems that replace logic-based systems, because they are very computationally demanding. There are complex problems, especially in relation to the language used. I hope that in the future we can somehow bridge the shortcomings of logical systems and their positive properties, so that we become capable of integrating these two types of systems, logical and machine learning. But we are still far from that. Nobody really knows how to do this. But these are the main challenges.
JU - Do you believe that logic can contribute to our daily lives? How can we expand the use of logic in our decisions?
Offer Arieli: I think this is the most difficult question so far (laughs). I believe that the greatest difficulty in this is that our politicians have not acted very rationally. This happens in Brazil. It also occurs in Israel. I wish logic was more substantial in its actions and decisions, but unfortunately, that's not the case. We must continue to try to expand the logic and convince people of its importance, but I have become more skeptical about this. Real life has shown that interests matter more than rationality. We can't lose hope, but this is a big problem.
JU - And how can we deal with paraconsistency without it being used to justify deliberately wrong decisions or injustices?
Offer Arieli: I will cite an example about this. It is known that legal systems have many contradictions and one of the challenges of this is isolating the contradictions and highlighting them. For this, paraconsistent logic is very useful, because you can identify the point where the contradictions in your system are. This also occurs in medical systems and in several other fields. In this aspect, being able to deal with contradictions in a rational but non-trivial way is a necessity for any system. Maybe people don't know the term "paraconsistency," but we do it all the time. We always have to make decisions and draw conclusions in the midst of contradictions, with insufficient information. It is very important to be able to formulate responses and decisions under these conditions.
JU - In other words, it is not just about dealing with contradictions, but about identifying them and subjecting them to rational parameters?
Offer Arieli: Yes. And once you identify the source of these contradictions, you need to decide how to act, decide whether you maintain the existence of these contradictions and still make rational decisions. Or if you decide to resolve these contradictions by reviewing existing information. There are many perspectives and methodologies that allow us to deal with these situations.
