Blondie24 Playing at the Edge of AI

Blondie24 | ScienceDirect

Blondie24 tells the story of a computer that taught itself to play checkers [西洋跳棋程序] far better than its creators ever could by using a program that emulated the basic principles of Darwinian evolution--random variation and natural selection-- to discover on its own how to excel at the game.

Unlike Deep Blue, the celebrated chess machine that beat Garry Kasparov, the former world champion chess player, this evolutionary program didn't have access to strategies employed by human grand masters, or to databases of moves for the endgame moves, or to other human expertise about the game of chekers. With only the most rudimentary information programmed into its "brain," Blondie24 (the program's Internet username) created its own means of evaluating the complex, changing patterns of pieces that make up a checkers game by evolving artificial neural networks---mathematical models that loosely describe how a brain works.

Published 2002,我使劲揉了揉眼,2002,不是2022.

其他策略——策略和技巧 (xqbase.com)

另一种类型的机器学习用来自动调整评价函数的权重,这属于遗传算法(Genetic Algorithms)的范畴,把遗传算法作用在评价理论上,就自然地得到了最适合的程序。以下就是这种算法的工作原理,你产生一个评价函数(典型的评价函数是线性的),每一项都有一个权重……这就是上世纪50年代Arthur Samuel在他著名的西洋跳棋程序里用的方法。

深蓝(Deep Blue)的小组用了大量特级大师的对局,然后让程序吻合这些局面。他们对程序和特级大师走出一样着法的频率作出统计,然后修改了权重再作尝试。如果程序能解决更多的局面,那么他们就保留这个修改,否则就重新修改。

神经网络(Neural Networks)能够发明一些评价模式,而不需要有人告诉它,它能对局面作出评价,并且结合到棋类程序中。神经网络由很多输入口,一些隐含的层次结构,以及一个出口组成。这些连接着入口和出口的层次,实际上由很多结点组成,结点有连接上一层的入口和连接下一层的出口。每个结点都可以和其他层有连接,每个连接是有强有弱的。如果同时改变连接及其强度,你就能得到一些评价模式,并确定他们的权重。

把有评分的局面告诉神经网络,它就会得到训练,而在结合了遗传算法后,神经网络也会自我学习。有个西洋跳棋程序Blondie24,就是用这种方法实现了自我学习的,并且下得还不错。我认为这才是真正的学习,它和人类的思考很接近。Blondie的作者把他们的程序吹嘘得很过分,但对于目前大多数研究者来说,事实上就是如此,你必须大造声势来获得资金。

posted @ 2022-12-12 23:08  Fun_with_Words  阅读(34)  评论(0编辑  收藏  举报









 张牌。