Hi, welcome. Welcome to. Well, what I expect is not just another talk about artificial intelligence, but instead I want to provoke you and ask you whether artificial intelligence is something new or has it been evolving throughout the years? Now, before we get started, a little bit about myself. My name is Vasko. As can see, I've been in software development for 20 years and counting, and I am coherently interested in the concept of artificial intelligence throughout history, in the sense that I am not totally convinced that artificial intelligence is a new concept. In fact, I argue that it has been there in our collective minds for quite some time. If you want to agree or disagree with me, or just send any comment, please do so. I can be reached via Twitter or LinkedIn. My contacts are on this slide. And without further ado, let's get started. As I said, I believe in evolution, in the evolution of everything and everyone, including ideas and concepts. So, is artificial intelligence a concept that has been just created or has it evolved through time? That's a question I ask. I have my opinion. I hope that will be clear by the end of this presentation. At the same time, we are going to look at the way how the concerns surrounding artificial intelligence have evolved through time as well. And since we are talking about time, let's travel back in time about 800 years before Christ, give or take a few, and let's land in ancient Greece. There we find the Iliad, a classical poem written by Homer, where the Ephastus God is described. Well, Iliad is not about Ephastus, but this character is mentioned. And as you know, in ancient Greece, the mythology was composed of several gods. And in this poem, Hephaestus has furnaces. He builds things made of metal. He builds machines. These furnaces know exactly what Hephaestus needs and wants. They are completely hands off. Whatever it is that he requires at any given moment to do whatever it is that he wants to do, they will provide. Now, you may argue that this is what we know today as automation, right? You push a button and you configure a whole set of machines to do whatever it is that you need doing. I give you that. True. However, back in ancient Greece, tripods were a sign of authority and could actually be a sign of power. Naturally, Hephaastus bit tripods for himself. Now, in ancient Greece mythology, gods would gather in an assembly of gods to discuss whatever it is that gods discuss. And if Hephaestus was not in the mood to travel to the assembly of gods, he would just simply send his tripods. Hephaestus'tripods would travel all by themselves to the assembly of gods for the duration and back to Homer's house. Now, this would mean that those tripods, those mechanisms, were capable of autonomous navigation, right? To get from Homer, Hephaestus'home, to the assembly of gods and back, they were capable of understanding directions. They were the perfect definition of an autonomous vehicle. Isn't that so? Well, you may argue, okay, but that poem is about a God. Even though gods, even in greek mythology, they were conceived as humans lookalikes. Does that fit in the definition of artificial intelligence as something that was built by humans to try to be closer or better than them? That's a bit debatable, I would say. However, it is a good example if we remain in ancient Greece and about the same periods, but we switch the text and now we talk about the Odyssey, also written by Homer. There is a passage there where the Phoenicius king is sending a group of visitors home at the end of their visit. And the king is so happy with them that he actually offers that they travel using the Phoenicius ships. And why is that? Now, you will forgive me, but I am going to read from a translation of the Odyssey because there is no way how I could put this better. It reads, for the Phoenicians have no pilots, their vessels have no rudders, but the ships themselves understand what it is that we are thinking about and want. They know all the cities and countries in the whole world and can traverse the sea just as well when it is covered with mist and cloud, so that there is no danger of being wrecked or coming to any harm. Now, isn't this wonderful? Almost 3000 years ago, Homer was imagining a fully autonomous sea vessel and a telepathic one at that. Nowadays there are but prototypes of these vehicles and they are far from being capable of doing what was imagined almost 3000 years ago. Now, if this is not a good example of someone already thinking about artificial intelligence, but not giving it that name, I don't know what it is. Now, let's continue our voyage through time and let's jump to our contemporary epoch. We land in 1754 in Europe and we find a book called Tretier de sansacion, or Treaty of Sensations, you know, on a literal translation by Monsieur de Kondilak, even though, well, todays we may look at this title and it may give us some thought. In fact, it is a philosophical treaty where Monsieur de Kondilak argues for the mind body dualism. Or as a contemporary philosopher, Gilbert Heil put it, the ghost in the machine. Now, Heil argued that such a dualism does not exist. But regardless of your opinions, what is interesting with the Tretidi san sacion is that de cond? To exemplify his position, a statue, and that statue was animated by an empty soul. He argued that if we could feed one sensation at a time to that soul, bit would eventually learn all human knowledge and all human abilities, therefore becoming equal to humans. In that regard, if you will allow me the obvious comparison, nowadays we train models. And how do we do that? Well, we feed them pieces of information, just like Monsieur de Kondellac wanted to do with his statue. And we feed them such information until such a point where we are convinced that the model has learned everything that it is capable of learning. Well, we haven't yet built a model that can learn of all human knowledge and all human ability. We have specialized models, but still the principle is pretty much the same. If we remain in the approximately the same time, we find a machine that was built with the purpose of playing chess. It was a box with a mannequin that would move the pieces in the chessboard. And it was said that this machine could place chess better than any human in existence. And in fact, the machine was showcased in many scenarios across Europe back then, and it actually won games of chess against all opponents. It was a sensation. Now, it was called the Turk or the mechanical turk, not because it originated necessarily in Turkey, but because the mannequin was dressed in an oriental custom. Now, this sensation came to an end when someone discovered that the machine was a hoax. There was no actual machine. There was a human, a very good chess player, but human, operating that machine, or should I say that puppet, to play the games of chess against the opponents. Sadly, there was intelligence. Yes, but it was human intelligence, hardly artificial. However, if we now jump to 1912, we find the work of Leonardo stories e quvedo. He built a machine, or an automaton, if we want to be precise, that could indeed play chess. Well, not a full game. It would play an end sequence of king and rook against king. It would always play with the king and the rook, and it would always win against the human opponent. And this time, this machine could actually play. It was, as said later in 1914, quite an advanced machine for its period. It could detect the position of the pieces on the board and could calculate the next move quite effectively. I would say it was one of the first, if not the first, example of machine that was capable of playing at least some chess. On the subject of intelligent machines, if we travel a little bit further in time, we find Isaac Asimov's work, he imagined a class of hobots that had a positronic brain. Now, the positronic brain in Isakasimov's conception was powered by a particle called the positron. It doesn't actually exist, but it was sufficiently powerful to create processing units that could indeed power a robot and actually give conscience to a robot or to what we today would call an Android. What was interesting about Izakazimov's robots was that they were bound by the three laws of robotics, three dogmas that were designed to prevent them from being used directly or indirectly to harm the humans. We can argue that Isaac Asimov was already concerned about the possibility of artificial beings, or should I say artificial intelligent beings being used as weapons against humans. Arthur C. Clark in 1968 wrote a story around a character wrote 2001 A Space Odyssey. And the interesting artificial character there was Hull, Hull 9000. But Hull was an operating system, so not a robot per se, but it was the operating system of a space station. Now Hull would observe and learn from the behavior of the human crew of this space station. Unfortunately, there was a malfunction in the space station and the crew decides that Hal needs to be disconnected. Now, faced with this perspective, Hal decides to defend itself. Having learnt about humans, he decides that the best way to defend itself is to eliminate the human crew. Hal became known by his sentence, I'm sorry, Dave, I can't do that. Which became, well, kind of an icon of artificial intelligence independence, or should I say some sort of sentience. Now, this is an example of an intelligent artificial being harming humans intentionally, arguably in self defense. But still, not all artificial beings in literature and cinema are dark or malevolent. In fact, one of my favorite characters is Marvin, the paranoid Android from Douglas Adams, the hitchhicker's guide to the galaxy. This Android has been around for a very long time. It is extremely intelligent. In fact, it is said that it never needed to use more than a tiny fraction of its enormous brain to perform any task. And the most interesting conversation it ever had was with a toaster, which I find quite interesting, considering that by this time Marvin had already met humans. And still a toaster was more interesting than the humans he had met. Anyway, going back to reality and a bit closer to our time, in 1996, IBM built a computer called the Deep Blue. This computer was capable of playing a full game of chess, and in 1997 it actually beat the chess grandmaster Gary Kasparov. You may wonder, why have people been obsessed with chess for so long and with machines playing chess for that matter? Because chess is an incredibly complex game in the sense that the number of possible combinations throughout an entire game of chess is so large that it cannot be solved by your typical combinatorics or game theory methods. That's what made it such a challenge for a machine. And in 1997, it was proven that a machine, or this case, software, could indeed play a game of chess as good and even better than humans. Now, I said at the beginning that we would be looking at the evolution of the concept of artificial intelligence. And I argue, given what we have just seen, that artificial intelligence was already there. Since the beginning, it was just not called that way. Humanity has been fascinated with the idea of intelligent machines performing better than humans at any given task. And they have always been concerned that those machines could, in fact, take over from humans and in some cases, even take over their own lives. So that's the dark side of artificial intelligence. Could it be a threat? Well, in cinema, a great example of artificial intelligence being a threat is the Terminator series of movies. It all started in that universe with an artificial intelligence that in the beginning, was supposed to help defend a certain group of humans. Skynet in that universe was an artificial intelligence in charge of the system of defense of the United States of America that, as anyone who has ever seen those movies know, went a bit rogue. Eventually, it decided that humans had to be destroyed because it became self aware and as a result, could no longer be controlled by humans. So humans decided to deactivate it. Again, we see the common pattern of humans perceiving an intelligence as a threat, and that intelligence deciding to eliminate humans in result. Another good example of a failed cohabitation between machines and humans is the universe from the matrix, also a series of movies where machines becoming self aware and eventually cohabitation becoming impossible. In this universe. Therefore, a war broke out, and this time, the machines did not decide to annihilate humanity. Instead, in this universe, they use human bodies to harvest them for electricity, but they need to be kept in a virtual reality world in order not to go insane. I'm not going to spoil the story, especially given that recently a new movie was released in this universe. But it is sufficient to say that things didn't quite turn out well in this universe for humans, and also neither for machines. Going back to the written word, another good example of artificial intelligence understanding humans, and vice versa, is robopocalypse by Daniel Wilson. Now, in this story, a professor attempts to create an artificial intelligence program that could be capable of absorbing all human knowledge. Now, it so happens that this program was actually quite successful in that regard. So successful that the program decided that first humanity no longer needed to search for knowledge. Bit would take over that task from humanity as it learned it, then decided to consider itself a God and state that humans had become obsolete now that it existed. When it got to this point, the professor who developed the program tried to disconnect it, tried to shut it down. Unfortunately for said professor, the program managed to take control of the environmental controls of the room where it was deprived the room of oxygen, killing the professor and escaping that room into the Internet, eventually infecting, or should I say repurposing, all other robots in the world, starting once again a war against humans. I am not going to say how the book ends, but it is quite an interesting ending. Now, we have now looked at a couple of examples where machines and artificial intelligence become dangerous to humans. We could argue that it was because humans were dangerous to them. Let's not get there right now. But another question deserves to be asked. And what if machines, what if artificial beings could be kind? What if they would go the other way? Instead of mimicking the worst in humans, why couldn't they mimic the best in humans as well? Machines like me by Ian McEwan is a novel set in a time where artificial humans, or synthetic humans, if you prefer, were just behind produced. So this man Charlie gets some money, decides to buy one of those synthetic humans called Adam. And these synthetic humans have a particularity. They are pre programmed from factory, but they don't actually have a personality. Their new owners tweak a whole set of configuration parameters in order to try to give their new synthetic human a unique personality. Now, it so happens that Charlie has a neighbor, Miranda, and she works with Charlie to give Adam a conscience. Now, Adam turns out to be almost perfect, an almost perfect human in such a way that actually, I'm going to spoil the story a little bit for you. A love triangle actually erupts between these three, and their relationship, emotional and even physical, gives rise to a few questions. Right, so, for example, what makes us human? Is it what we do on the outside? So, are the things that others can view from us that make us human? Or is it about our inner lives that make us human? Opinions are divided regarding machines like me. I believe it is still an interesting bit, and it may lead us into other works, such as, for example, real humans. Originally a television series running from 2012 to 2014 set in a similar world. There are synthetic humans. They are intelligent, there are different models with different purposes, and they eventually also build relationships. Or should I say humans build relationships with these synthetic humans, giving raise to questions such as do these synthetic humans or human robots or even hubots have any rights? Should they get paid? Should humans be allowed to form relationships, emotional and otherwise with them? Is a new society behind created? Are there parallels, are these synthetic humans in these works parallels to certain groups in our society? How do we as a society face the possibility of having synthetic humans who are better than humans walking among us? That is a question that is also asked indirectly in Philip K. Dick's classic, do androids dream of electric sheep? In this world, also, androids exist built for specific work, mostly manual labor. But a few have evolved beyond that stage, becoming humanlike not only in their appearance, but also in their behavior, in such a way that only a complicated physical and logical test is required to determine if a given being is human or synthetic. Again, questions are raised on what it means to be human. Now, if we are talking about human characteristics, if we are talking about machines, if we are actually talking about building machines and building algorithms that make decisions, there is another human concept that becomes quite important, which is the concept of fairness, of justice. Are our algorithms fair? So let's take the discussion a few notches down from the philosophical point where we were down to the algorithmical level, and let's ask ourselves if the models we are building are fair. Let me give you a couple of examples of models that didn't quite turn out as they were meant to be. In the beginning of the year 2000, also at the beginning of the Covid-19 epidemic, confinements were in order. One of the problems that had to be solved was the problems of student grades, because students had been working for months. And now, when the confinement started, well, it was also the beginning of the exam season. So educational authorities all over the world wondered, how do we solve this problem? The traditional way of determining a student's grades, an exam in a classroom, together with other students under the supervision of teachers, was not possible. Many solutions were adopted all over the place. Specifically in Scotland, the Scottish Qualifications Authority decided to employ an algorithm that would calculate or predict the best grade for each student. Sounds like a good idea. The problem was that the results were particularly skewed depending on where the student lived, depending on where the school was located, and also depending on, sometimes, the school itself. Unfortunately, the algorithm was not actually looking at the academic performance of each student. Criticism was, as you can imagine, paramount. Eventually, the algorithms results were overturned, and instead, each teacher awarded each student a grade based on the student's work throughout the year. Another example of a biased algorithm was, in the United States of America, an algorithm intended to preemptively avoid complications in patients that could potentially need medical care in the future. And so the idea was, let's apply this algorithm to these patients history so that they can be preemptively treated in order to avoid serious complications down the line. Of course, one can be cynical and say that these algorithms had not only the best interest of the patients in mind, but also the best interest of the health care industry. That's another story. Regardless, it sounds like a good idea to try to predict who needs more medical care to prevent complications, right? It is a good idea. The problem was that this algorithm was using a proxy indicator, and that indicator was the previous health care spending of each patient. So, in other words, if the patient had spent a considerable amount of money in healthcare in the past, then it was very likely that patient would suffer complications in the future. Therefore, it should receive health care, or should I say more health care now in order to avoid such complications. The problem with this indicator is that certain segments of the population, for socioeconomical reasons, or just for lack of availability, or a combination of these and other factors, did not spend much money in health care in the past. So when they got into a situation in which health care was required, the algorithm would look at their history and would conclude that these people were not in risk of serious complications, when in fact it would not be the case. And of course, unfortunately, according to Scientific American, these conclusions were mostly towards black patients. The algorithm would conclude that they would not suffer from complications, so no further care or no additional care was required, whereas other patients who had spent more money in the past were awarded more care. Now, the algorithm has since been revised according to the scientific press, but the whole point is that the idea behind an algorithm may be quite good and may be worth of pursuing, but the data that is fed into the algorithm may not lead to that conclusion, because the algorithm may end up training on a bias in the data instead of the intended purpose. Which is why an active field of research nowadays is understanding the machine learning models is explainability, such that nowadays these models, they fall into one of two categories. They are either black box models or white box models. As the name implies, black box models produce results that are extremely hard to explain and may not even be understood by domain experts. White box algorithms have been designed in a way that allows results to be understood by domain experts. It goes without saying that this is still a field of active research, and the goal is to have as little black box models as possible and as many white box models as possible. Especially because we can say that all data is biased. Every single data set is biased because someone had to make a decision on which data was present there. In some cases, it's pretty clear that some pieces of information should not be in the data set. For example, if we go back to the algorithm that was calculating student grades, if the idea is to evaluate students academic performance, then, for example, it does not make sense to include the postcode in the data set. It may happen that the algorithm will find a correlation between grades and specific postcodes, and then people are graded according to the place where they live instead of according to their performance. This is just one possible example. There are many more. And again, this is why explainability is so important todays. And this is why extreme care needs to be taken in preparing data sets and submitting what is relevant information for the models. And what about tomorrow? Are we going to see an artificial intelligence reach the singularity? Are we going to coexist peacefully with sentient intelligences? Are we going to see our worst dreams become true? Hopefully not. Well, that's something we cannot predict today. We can say that is still a black box prediction. There is no way of understanding how things are going to turn out. I hope I can still make a nice trip in a phoenician vessel sometimes. Thank you for being with me. I hope this has been understanding and useful. Let me know if you have read or watched any of the works I mentioned today. Have a nice day and be let's build a better future together.