Before diving deeper into the topic of “artificial intelligence”, it is worth taking a look at an explanation of the term. What is AI? Do we perhaps experience AI in our everyday lives, or is it just a dream that provides material for science fiction novels? Is it perhaps the case that AI is only a topic in research?
Short answer – of course, it’s not just a research topic and hasn’t been for several years. If you search for the term AI, you will find it everywhere. But if you compare the many definitions and explanations offered, a certain disagreement cannot be wiped off the table. The topic is not brand new, but it is more relevant than ever. We humans associate many positive but just as many negative emotions with AI. That’s why it’s always the subject of controversial discussions. In the meantime, we can no longer resist coming into contact with AI (topics) on a daily basis. A definition of my own would understandably be as difficult for me as it is for the experts, but perhaps a few examples will facilitate access to the topic. Here you can ask yourself the question whether these examples are AI or intelligent behavior.
Gaming AI
If you like to play on the computer (or console) and there are no human opponents available, you play against an “AI”. So there is something artificial in the computer that plays the same strategy game with me. You can observe the AI as it builds buildings, trades or attacks opponents, sometimes well and sometimes not so well coordinated, and makes contracts. In many cases, this wild doing can be interpreted as quite intelligent behavior.
So is this really artificial intelligence? Or is it “just” a process carved in code that is always repeated in a similar way, taking into account a few predefined rules? The different AI strength can be imitated quite easily by granting a stronger AI more (construction) resources or faster construction times. Among humans this would be called cheating. Analogously, a weak AI can be created by a corresponding malus. Of course, I want to point out here that it has nothing to do with the types of strong and weak AI you find in the literature. My example is aimed at game strength and not at capabilities per se. I am not a game programmer and therefore I simplified a bit when it comes to the computer opponent’s abilities. Surely there are more sophisticated ways to implement “intelligent” behavior, sometimes stronger, sometimes weaker. For example, one could vary with the provision of computing resources, as in the next example. But first back to the initial question. Can we really speak of an AI in the case of our computer opponent?
First I was convinced, but now I’m not so sure. In my opinion the answer depends on the one hand of course on the definition and on the other hand on the implementation of the computer opponent in the code. If the “intelligence” of the AI results only from a few actions with a fixed sequence, which are triggered when certain threshold values are met, then one can justifiably doubt the designation AI.
Chess AI
Another example quickly comes to mind for some when they hear/read “Deep Blue”. That’s the name of IBM’s chess-playing computer, which in 1997 defeated world chess champion Garri Kasparov in 5 games by 3.5 to 2.5. What was going on here? Did Mr. Kasparov have a bad day or could Deep Blue simply calculate faster? We have here a game with 64 squares and 16 pieces per color, with the pieces getting less over time. White has 20 options at the beginning of the game (16 pawn moves plus 4 knight moves) to open the game. This half-move by White is followed by the half-move by Black (both half-moves together make one move). Black has the same 20 options to choose from in response. So there are 400 different starting positions before the second move. After that, it quickly becomes confusing, because each possibility again allows other moves of the opponent. Good chess players now try to calculate as many half-moves as possible in advance, and then select the most promising of them. As far as I know, chess programs do exactly the same. The playing level thus depends on the depth of calculation, how many half-moves the human or the machine can calculate in advance. But shouldn’t the human be hopelessly inferior every time? But why was Kasparov able to win 2 games? The background could also be interesting for non-chess players.
Garri Kasparov (*1963; World Champion from 1985-2000): “If it’s a simple variation without a lot of branching, that can be 15 moves. But usually I operate with four to five moves.”[1]
Vladimir Kramnik (*1975; World Champion from 2000-2007): “When I see a certain game situation, I eliminate 99.9 percent of all possible moves – because they would be inappropriate in this position. So I can concentrate on the two, three or four logical moves and their consequences. Humans [can] sometimes even calculate further [than computers] because we are highly selective. That’s a small advantage for us humans. You can’t use it in every position, but sometimes you can. And then we can think 20 moves ahead. But that really happens very, very rarely.” 1
Gaming / Chess AI yes or no?
Are we facing an AI here? After all, the computer opponent can be assumed to behave as intelligently as a grandmaster. Otherwise, Deep Blue would probably not have been able to achieve such a triumph. But wait – the word intelligence comes up again! It seems to be of importance what intelligence really is. Is perhaps not intelligent behavior equated / confused with pure computational power here? Actually nothing else happens here than to try out all possibilities by brute forcing and selecting (on the basis of certain evaluation criteria) the best half-move of it – namely the one which checkmates my opponent as surely as possible. Ok, so you can beat the computer by simply memorizing all possible moves and using the human advantage of selection. This way, bad branches in the decision tree can be cut off right away, whereas the computer has to consider these moves as well.
Humans can thus save enormous computing resources. Where this leads to, however, and that even supercomputers have big problems with it, is discussed by Stefan Klein in his interesting paper “Wie berechenbar ist das Schachspiel? Eine kurze Betrachtung im Hinblick auf die finitistische Erkenntnistheorie von Dr. Alfred Gierer”. [2]. At this point, only one thing. He writes that after two moves there are already 72,084 possible positions. For the number of possible positions after the 3rd move there are only vague estimates (between 9,12×106 and 9,14×106).2
Already for the first 40 moves the estimates are approximately 10115 bis 10120 different game sequences.[3] That is an incredibly large number. So it seems that neither man nor machine is capable of completely calculating a game from start to the end in their heads, or on the board. Apparently both have the possibility to win the game. The machine can calculate more half-moves in advance in most cases, and the human has the ability to select. If man could calculate the same number of moves as the machine, he would never be denied any intelligence because of this increased ability. In addition, he would still have the advantage of selecting, which has been prevented from the machine so far. So one can say that the machine is able to play chess better than the human being in the vast majority of cases (since there are too few grandmasters). In this case, the machine is usually better at an activity that requires human intelligence. After each half-move, both sides begin anew to calculate the winning probabilities of their half-moves. Both sides are able to adapt to changing circumstances and to incorporate this change into their own actions. Is the chess playing computer an AI for me – My answer: I am willing to answer with YES.
If you are interested in some historical background about computer chess I would like to refer you to the site of Mathias Grontzki[4] where you can also find an interesting story about a very special Turk who defeated the great emperor Napoleon three times.
(Un)real life
My last example is the following.
A woman is blindfolded after an operation and has her chauffeur picked up from the agreed location. When asked where to go, she answers “home, please”. During the long car ride through the night, she has dozed off and is awakened by an emergency braking. “What was going on?” – The answer from the chauffeur came promptly: “I saw that deer was approaching and braking was the right decision in my view”. A little later, the woman heard the familiar squeak of the electric garage door and the chauffeur confirmed her guess by saying, “We’re home.” Getting from the car to the front door is easy out of habit, even with the current limitation. Upon reaching the front door, it is already open. Once in the hallway, the chauffeur’s concerned voice asks about her well-being and whether she would prefer to process the shock of the car ride with a warm bath and relaxing music or on the couch with a cup of tea?
Now you might think, nice story and well to the one who can fall back on a chauffeur in such a situation. But what if only the woman got out of the car and she was also alone in the house? Now the thought carousel starts, something can not be right. Making a call blindly is quite easy via voice input, being picked up from the “agreed location” can also be imagined via location sharing or other solutions. One has certainly heard of self-driving cars. But here it comes. The right reaction in case of a near accident and that the chauffeur wants to have seen the deer at night and then still had enough time to make the right decision. Totally unrealistic? Why shouldn’t cars be able to be equipped with special sensors, with the help of which autonomous cars can anticipate/predict the movement of animals. I can already imagine that. For the automatically opened entrance door there are again several solutions like cell phone app or camera recognition. But now it gets really exciting. How does the chauffeur get from the car into the house? Perhaps this example is artificial, but how far away are we from this scenario? There are many things in the smart home sector that can be controlled via a corresponding app – there are no limits on your imagination. Anything that can be detected by voice, image or via other sensors can be used. Just imagine that these smart assistants from the well-known companies become a bit more intelligent and can be paired with all these apps and you would simply take the smart assistant with you on your cell phone. Paired with the car, it has access to everything that’s going on in the car and in the house, it accesses and controls the apps there.
(Un)real AI?
At the end, it is again valid to consider how one should think about this chauffeur. Can we call it an AI, or did it behave intelligently, or can it do something better than a human? Because of the additional sensors, the car was better equipped at night and thus triggered a human-like reaction. If a human had been equipped with similar sensors, it would have had a comparable reaction given enough time. The system has processed/stored past information from the car and, based on this, reacts in the house as if it had really been there.
[1] https://www.gutefrage.net/frage/wie-viele-zuege-im-voraus-denken-professionelle-schachspieler
[2] http://www.sfbux.de/wp-content/uploads/artikel/berechenbarkeit.pdf
knowMLedge.com 