Íà ðóññêîì: http://www.proza.ru/2010/11/21/1103
Even those who can visualize in their mind the work of the Artificial Intelligence (AI) in calculating the winning numbers in lottery (and most people can’t), often mix up between the AI methodic with the statistic one. It can be understood, since any intelligence is forced to rely on experience for further advancement, which means, in a way, to be based upon statistic data. But what is the difference?
Any idea of a statistic method relies on a data base with results gathered beforehand. The data needed might change, so can its processing, but the basic essence will remain the same. The AI works in a slightly different way, even with the principles at the base of those systems can vary. For example, let us look at one of the models, requiring the interaction with humans, which completely spares the use of the classic statistic method.
For this, let us look at the semantic net of neurons of the AI at its simplest form. It has an entrance, an exit and an internal mechanism. At the entrance, a signal is introduced: in the case of our example, it can be numbers and the man who gave them. The mechanism processes the input according to its database of numbers, which considers, for example, the amount of numbers given by different people, their matching and the “percent of luck” of each giver, compared to each number separately. At the output, a matrix of numbers with a percent value for each number. Of course, the example is simplified to the maximum- only to allow the comprehension of the full picture of one of many examples of AI work principles.
So, 10 thousand men during a year try to guess the numbers, transferring the data to such a mechanism. Then, a day comes when AI “decides” that it’s ready to send 10 thousand tickets. After receiving numbers from real people, it tests each man and each of his numbers separately. People with ID numbers 600, 700 and 5000 guess the fourth number with the high probability of 40% and they gave the numbers 25, 26, 26 accordingly. But the majority (60%) has bet on the fourth number with the variants 25, 26, 28. There are 5% who bet the 4th number to be 20, and those people don’t enter the number of “lucky ones”, their luck in guessing the 4th number doesn’t rise above 1%/
The AI doesn’t care how the “lucky ones” guess, but it “remembers” that they can be trusted more in the choice of specific numbers, so the weight of those numbers is also great. In this case with the 4th number, it is probable that the system will give “OK” to 26, since it’s present twice at the “professionals of the 4th number”, and also seen in the majority. But the number 20 won’t be ignored by the matrix either, just that the implied value of it will be so small that it can be ignored when the 10 thousand tickets are constructed.
For an even simpler example, let’s take a matrix for 3 numbers out of 4 possible. It can be like this:
..............place............
number......1......2......3
.....1....75%...0%...0%
.....2....0%...50%...50%
.....3....0%...50%...25%
.....4....0%...0%...75%
Let us imagine, that the system was partaken by only one man, who sent the numbers 1, 2,4. So the matrix at the output will be thus:
..............place............
number......1......2......3
.....1....100%...0%...0%
.....2....0%...100%...0%
.....3....0%...0%...0%
.....4....0%...0%...100%
This means that the number 3 won’t even be considered, since the “authority” of the only “teacher” of the system is impeccable. But the most interesting thing will happen when the AI will start learning on it’s own. To continue the example of the three numbers, we’ll assume that the man always bet 1, 2, 3, but the 4 was always drawn at the 3rd place. With each draw, the “trust” of the AI for the number the man places on the 3rd place will drop. This will happen until the system will bet “the opposite”, despite the “teacher”, since the probability of him NOT guessing is higher than his “luck”.
Back to the example with the matrix of 37, the probability of that turn of events is practically excluded, since the AI will “search” for people, who can be “trusted”, in percent value. The system will learn from their mistakes and correct itself with each draw. But, especially in the case of a lottery, the partial value of the majority can be a negative factor as well in the AI system, since the more people guess the prize; the lower is their share in it. That means that the AI can go “deliberately against” the majority, along with a small group of “lucky ones”, in order to get more in a case of a victory. In other words, for example, the sixth number can’t be guessed in most cases, and the system chooses the numbers from the lowest percent, through exclusion. That way, a matrix of sorts is form, in which every number has a place in the needed six, and also a partial value in each of the places. The system decides which of the numbers are to be chosen according to the demands of number of tickets and the partial value of the matrix.
Off course, the AI doesn’t have to work exactly by this system (I also remind that the system analyzed here is simplified to the maximum, in order to allow general understanding), and there are many principally different mechanisms and mathematic models of AI. But here’s something of interest: after working several years in the lottery business sphere and constantly conversing with different professionals in the given area, running into many techniques of lottery and defending a dissertation regarding neuron network AI… I’ve never run into successfully working developments regarding methods of AI operation in order to foretell the results of the lottery.
The author of the article is the head of Programming and Statistics department in Israels’ lottery company.
Method "Ranges": http://www.proza.ru/2010/11/14/569
Method "Grid": http://www.proza.ru/2010/11/25/896