Filed under: Artificial Intelligence
But where is the entrance?
It’s time to take a break.
Science fiction movies prefer to depict aliens as insect-like, which are commonly ugly, cold blooded, uncivilized, without fear of death etc. However, at the same time, they usually have technologies we can’t even imagine of (e.g. district 9, cowboy and aliens, etc.). So does it mean that those species, somehow, at some point, evolved into a certain state, that unconsciously they know how to let P = NP (or approximately, but with guarantee), such that there is no need to have a highly civilized society to support researches which are unpredictable, and inventions can just be done on demand?
December 11, 2011
It seems rather too far away for the biological science area to discover the mechanism of natural death and find out a way to make existing people immortal. It will probably take more than 50 years. So I’m not hoping to have that applicable to me. Another possible approach is to first develop true AI and hopefully by using large scale AI, this science discovery process can be accelerated. However, based on my current knowledge of AI, the true AI, when developed, has limitation on its scale, which intrinsically, will not exceed the level of human intelligence to a non-trivial extent. What only we can do to improve the capability of AI is to improve the connectivity of AI systems to the external world, and the communications among individual AI systems. But after all, that is no more than a group of well-collaborated scientists.
However, there is still hope. We can do something like this: When we know truly what AI is and how intelligence works in our human brain, which is not that difficult, then we can develop a artificial brain(AB) that is functionally the same as the biological brain(BB), except that it is implemented by man made techniques, and will never die. A further step we need to go through is to develop a method to connect AB to BB, let the artificial neurons and biological neurons work together, exchange information freely. We make sure that at first, the portion of AB is 0%. Then the portion of AB goes up gradually by substituting neurons in BB one by one, using neurons in AB. Finally, we get a AB that never dies, and it is you. At any time when the substituting is happening, you are still alive and have your complete, unique personality with the hybrid brain. Once you are 100% in AB, then you can base your body on whatever you want, thus you never die. This is not that difficult compared to the biological approach, and is hopefully possible in the near future. I’m looking forward to that.
If you watched the Tron movie, this makes a reasonable explanation to the fact that people can change their physical forms to the digital correspondences, and still maintains their complete personalities continuously.
April 26, 2011
Here is a video about this subject I presented in my English Speaking and Teaching class.
/video/FirstPresentationTL504Apr.flv
Here is an interesting paper discussing this issue 
October 14, 2010
学过一些人工智能以后，对于我们自身的很多问题，都可以有新的认识，下面列几个这样的例子：
- 聪明与否是后天决定的？很多时候这是我们聊以自慰的说法，其实不尽然，从一个刚出生的生命，发展成一个具有初步认识能力的人，是一个很艰难的过程，我们的大脑里，结构并不是单一的，有相当多的硬逻辑(Hard Coding)，在生命的最初就在帮助我们形成各种认识，这在人工智能里称为启发(Heuristic)。另外，神经元连接的生长方向也不是完全自由的，有各种导向因子在帮助成形，这也是大脑里有各种分区的原因。这几点都是可以有先天的区别的，而他们对聪明程度都可能产生很大的影响，比如说爱因斯坦就是因为左右脑之间的连接特别多。
- 记忆可以被复制？尽管回答它可不可以还是很遥远的事情，可是基本上来说，这个做不到。学过人工智能就会知道，大凡勉强能被称为知识表达的模型，其中的知识都是高度混合杂糅的，比如说神经网络里的权重，它们并不跟我们所理解的知识要点有一一对应的关系，而它们的量的确定，也就是通常说的学习问题，往往都是超NP难的，假设将来真要做记忆复制，很可能就要把知识从一方表示出来，再让另一方学习，这就跟一个人用嘴说，另一个人从听到的内容里学习差不多，是很困难的事情。唯独贝叶斯网络是一种比较好的知识模型，能很直接地作为现实世界的一个抽象，然而，它的结构学习过于困难。很有可能的一点是，神经网络是把贝叶斯网络进行变形得到的，是变换成一种利于学习的东西得到的产物，如果这种变换的难度在可控的范围以内，那么从权重到知识的双向转换才变得有可能。
- 人工智能可以超过人类？这个问题很难说，有关智能的基本理论并没有成形，所以关于究竟智能的发展空间有多大，还很难有定论，乐观主义者总会说一切皆有可能，可是反过来讲，如果真的有更强的智能，那么进化作为最强大的设计师，应该早已把它发展出来了，事实上没有，那么到底是因为人类还太年轻，还是因为这件事情本身就不可能呢？从学术圈的发展来看，目前困绕人工智能领域最大的问题是，通用的模型没有办法被有效地学习，稍微专用一点的模型则面临推广能力不够的问题。所以今后很长一段时间，人工智能的主题将是近似，即如何有效地近似，使得模型的学习能在很短的时间里自适应地抓住最主要的逻辑。而这种近似，很多时候是一种trade off，它们对于智能的规模，记忆的容量这些要素，会提出什么样的要求，现在还不明朗，如果这种trade off使得智能只能达到现在人类的规模，那么人工智能就没有办法超越人类了。
- 音乐是宇宙间统一的语言？曾经有人下过这个论断，紧接着就有天文台把载着音乐的电磁波信号传到广阔的宇宙。可是事实上，这个问题是这一篇blog里唯一最能下定论的，那就是，音乐不是宇宙间统一的语言。说穿了，音乐是情绪的重现，对于音乐的解读，取决于人的听觉系统和管情绪部分的脑是如果关联的，对于人来说，可能都一样，但是对于不同的生物，这就未必了，而这种关联的产生，除了先天的硬逻辑以外，大部分取决于我们日常做不同的事情时听到什么样节凑的声音，如果一个人的人生经历很不正常，那么这一部分就可能有区别，一个例子是凡高，在他的脑海里，各种视觉影象代表的含义与正常人是不一样的，听觉和视觉虽然不一样，这一点上是共通的。
- 为什么婴儿会哭？其实这个问题也是由人工智能来解释的，不过比较复杂，简单来说就是活跃的神经元变化比较乱的时候，人会感觉不舒适，这种神经元的快速变异，正是大脑开始工作最初时期的一个表现，所以如果刚出生的小孩不哭的话，说明神经活动没有开始起来，解决办法就是医生必须打小孩的PP，把神经活动激发起来，相当于给一个反向传播的神经网络一个初始的随机刺激。
February 21, 2009
今天来点评一下黑客帝国里最有哲理的话。
- Everything that has a beginning has an end. [by The Oracle and Smith]
最经典的一句，灌穿始终，当Oracle最后一次跟Neo说这句话的时候，Neo终于明白了问题的solution是要end他自己，而当Smith最后悟到这一句的时候，终于Crack了Neo，结束了这一切。其实任何时候回答不了别人的任何问题，都可以这样装B一句，不会有错的。
- You see there is only one constant. One universal. It is the only real truth. Causality. Action, reaction. Cause and effect. [by Merovingian]
这句学Machine Learning的应该最有感觉，世间的一切逻辑都逃不出Causality，所以专门用来学习Causality的Bayesian Network才会得以大行其道，连Bill Gates也十分推崇Bayesian Network。
- Everything begins with choice. [by Mopheus] The problem is choice. [by Neo]
前一句是Mopheus回答Merovingian的话，后一句是Neo回答The Architect的话，都很经典，虽然世间的逻辑只有Causality，但是因为我们不能考虑到的因素太多了，所以才会有Uncertainty，不能任何事都按Causality来办，所以才会需要Choice.
- Choice is an illusion, created between those with power, and those without. [by Merovingian]
这是Merovingian回答Mopheus上一句的话，同样经典，具体的诠释要等到The Architect出现以后才有，可见Merovingian作为The Matrix的6朝元老，对领导The Architect的政策，是吃得很透的。
- She (The Oracle) stumbled upon a solution whereby nearly 99% of all test subjects accepted the program, provided they were given a choice – even if they were only aware of it at a near-unconscious level. [by The Architect]
这是上一句的诠释，也是The Matrix的精髓之一，跟《货币战争》里讲的大道无形，每个人都被没有知觉地剥削着，是一样的道理。The Architect的王者风范也体现得淋漓尽致。
- Hope, it is the quintessential human delusion, simultaneously the source of your greatest strength and your greatest weakness. [by The Architect]
The Matrix的精髓之二，对机器种族来说，既然想办法阻止人类的resistance要花很大的代价，而且不一定能做到没有漏网，那么精心设计一个方案，让他们能看到希望，并以此为精神支柱，行动纲领，聚在一起，然后监视和操控他们的行动，这样就可以高枕无忧了。可怜了Mopheus，扮演了一个傀儡的角色。历史上，宋江也是差不多，虽然他觉得他的替天行道很有正义感，其实对朝庭来说，它起的作用是网罗了要造反的人，然后可以一网打尽。
- Then tomorrow we may all be dead, but how would that be different from any other day? This is a war, and we are soldiers. Death can come for us at any time, in any place. Now consider the alternative. What if I am right? What if the prophecy is true? What if tomorrow the war could be over? Isn\’t that worth fighting for? Isn’t that worth dying for? [by Mopheus]
最困难的行动前Mopheus的一次演说，可见他作为resistance的领导人，得确是相当有演讲能力，一句话说得人们热血沸腾。可惜前两个what if都不对，第3个以一种没有实效的方式对了。
- Do not try to bend the spoon; that’s impossible. Instead only try to realize the truth: There is no spoon. [by a kid]
Neo在跳楼训练的时候给自己打气也说过there is no spoon，意思是一切都是虚幻的，老美似乎喜欢用spoon来指代实在的东西，在美剧Friends里，Ross和Joey说的Grab the spoon，就是把握机会的意思，不知道有没有联系。
- What is real? How do you define real? If you’re talking about what you can hear, what you can smell, taste and feel then real is simply electrical signals interpreted by your brain. [by Mopheus]
黑客帝国之所以可以被黑的原因，不过为什么The Matrix里死掉了外面的世界里也要死掉？是仅仅为了拍片需要呢，还是有某种程序设计上的技术原因，比如说，因为加密机制的问题，在可能会死的时候没有办法及时中断与The Matrix的连接？
December 28, 2008