主题：【原创】围绕脑科学而发生的若干玄想 -- 鸿乾
家园 晓兵有个帖子对这个问题说的更详细 http://www.ccthere.com/article/3864887
家园 你能读懂？ 他帖子的易读性上已经超越人类极限了。
家园 智能的本质不是算法，而应该是在物理上的理解的意思是 就是Norvig与Chomasky所争论的核心观点。Chomasky认为不去理解智能产生的物理本质，一层层的从哲学到心理学，从生物到化学，大分子，小分子，原子，电子之间的相互作用如何在大脑这个规模上产生出智能来，而去单纯到依靠从大脑对外界沟通的行为上来获取大量实验数据，然后使用数据挖掘来排除噪音，开发模型去预测去除燥，最后把这个模型当作大脑中的智能，这不是他心中的正确的科学探索方法.
chomasky认为所有科学史上的发展都是找到了内在的原理才解决了问题的，从来没有说把问题的所有表现行为特征的数据做成数学模型，然后把这个模型当作科学原理去指导未来的进一步探索。比如牛顿的万有引力，伽利略的日心说等等，考数据挖掘是发现不了的。
这两个人是不是代表智能科学的最高水平不了解，但是他们两人对这个领域的了解代表了最高水平应该可以从他们自身的位置与关心的问题看的出来。当然许多人一门心思做工作，有观点但是不愿公开在这个问题上多花时间，这样的人也是很多的。 你这个连接中也提到chomasky现在别人都说”看到他绕着走“，所以他自己才要自比伽利略了。
有的最高水平的人，比如做深度学习的那个现在Hinton也加入google了，忙都忙不过来，哪有心跟人探讨这些问题。
家园 Hinton大牛 家园 the other 角度: 社会集體測度学 智能的本质 as a 社会集體认知系统, as opposed to that of a individual(but obviously we have to study individual 认知系统 as well);
for example: us 社会集體认知系统, how come that republican vp presidential candidate has been republican's top ticket? I listened to him 社会集體once on tv about his red neck econ logic, a joke;
obviously, there are many us systematic reasons, good and bad, tons of reasons there to explain things like this, but it also tells us that: politically, us 社会集體认知系统 pays huge cost in collecting and analyzing information for voters of a supposedly a democratic society of a global manager/leader, can AI provides a better model not to replace the current media/political campaign/brain washing(huge amount of money spent), but at least to help voters to be politically less stupid?
answer is: not, we have not even figured that out in economics, not to mention politics;
as an result of those "non-scientific" or 偽科學 nature of the current market economics and democratic politics almost globally, those "smarty" top social scientists took advantage of the current 带病上岗 system, and figured out how to brain wash and lead the "crowd" by family coaching, practicing/learning and a lot of luck, and become society's top leaders, and with 惯性, once top social scientists=4 ever, globally, and "real" scientists ended up as part of tool kit for those top social scientists, with a modest payroll, almost like Europe a couple of hundred years ago? when church leaders=top social scientists, at least co-managing the society, running the show for the mass;
tg's route is a little bit different, tg actually inherited and developed 馬列's whole series of models about 社会集體认知系统 and crowd's 觉悟系统 in particular, and that is part of reasons why tg is such a super effective social management organization of 1.5billion people, with all the "cost" discounted, tg inc is still highly profitable, low pe ratio stock, and everybody still loves it, even with the possible "double top" chart.
and the white can never figure that out, how come? as I said before, the current white never really understand their other white cousins 馬列's super models about 社会集體认知系统 and crowd's 觉悟系统, among other 馬列's models;
in terms of that kind of game, the white is too simple, too nave, way behind tg;
so unless white can figure out a new round technology break through based strong economic growth of many years, white will 自顧不暇 for long time, they will be too busy handling the class struggle between the crowd and elite in their society;
as a result, they will need tg's help from time to time in managing many international issues, and most importantly the Chinese market;
I think fb and goog have some ambitions in those social 智能 things, but as you said, fb guy is too young, not a bill gate's level; bill gate's math test result is actually later on a math publication; goog guys are all math guys, and they are over extended, kind of;
besides, we need break through in physics: 社会集體=强相互作用 system, a "full interaction" model, etc;
so, good for tg, many happy hours ahead
家园 发现你以前的一个帖子早就对这个话题有过评论 与这两个人的交锋涉及到的问题基本上是一样的
http://www.ccthere.com/article/3864887
放在一起读比较好。
家园 还是应该分层次来看这些问题 没有可能一下子全部解决，就必须要考虑分层分块。
”理解智能产生的物理本质，一层层的从哲学到心理学，从生物到化学，大分子，小分子，原子，电子之间的相互作用如何在大脑这个规模上产生出智能来“，这肯定是终极目标。但是，这个终极目标是短期内，绝对无法达到的。既然如此，目前可以达到的目标是什么？分层如何分，分块如何分？
我个人认为，目前正在积极发展的方向，即从神经生理和心理的角度，向更微观进行，以及向更宏观进行，是两个非常好的方向。
向更宏观的方向进行的，其实就是Darpa的那个著名的要求，要求搞出一个从数学上可以把控的算法来，使得在某些地方可以比较好的重复一部分脑的区域的功能，即使是很小的脑的区域也好。
还有一个方向，就是层次更高的，即各种统计方法。这种先获得结果的思路永远是有用的，也是第一考虑的。就用Chomsky例子，在没有搞清楚牛顿的规律前，各种各样的行星模型大行其道，包括非常精确的模型（例如希腊的那个行星计算机）。这些行星模型自然是有用的。用这样的比拟来看，事情就一目了然了。各种统计方法，其实就是那种行星模型，极端有用，但是并不提供穿透性的认识。
但是，另一方面，脑的认知，恐怕归根结底是各种统计方法的某种物理实现，只不过人不自知而已。从这个意义上看，统计方法又有其非常核心的价值。很可能就像人家说的，人脑的理解语言的部分其实就是一个很精密很快速的隐马尔科夫链的决策机。
家园 "所有科学史上的发展都是找到了内在的原理" chomasky认为所有科学史上的发展都是找到了内在的原理才解决了问题的, kind of one of the aspects of "the white core of the white logic", making sense to me, that one is a long article, I have not really read it yet, but your comment is pretty much on the money, a good summary
家园 科学家用电线连接两老鼠大脑共享信息 上个月早些时候，杜克大学的研究人员给老鼠装上红外线探测器，使其能感知红外光。在发表在《Scientific Report》的最新研究中，他们利用相同的技术，让两只实验鼠实现大脑信息共享，两只老鼠的通信距离甚至远达上千英里，他们实现了现实版的心灵感应。
研究人员首先训练一对老鼠完成简单任务：在控制杆上面的灯点亮时按下正确的控制杆，以能喝上一口水。然后，将两只老鼠放置在不同的封闭盒子内，在大脑处理运动信息的皮质植入微电极阵列，将两只老鼠的电极连在一起。
一只老鼠充当Encoder，另一只充当Decoder（如上图）。当Encoder按下正确的控制杆，其大脑活动产生的电刺激信号会传递到Decoder。Encoder能看到指示灯，但 Decoder没有，因此为了喝水Decoder只能依赖于Encoder通过大脑接口传来的线索。在实验中，Decoder的成功率约为70%。
- 违规禁，【补充】可申诉或道歉来解。偏要看
家园 感谢：作者获得通宝一枚。 希望早日进入实用阶段。今天看到google自驾车仍然有问题没有解决，比如下雪天看不清地面的线，交通灯坏了时由交警手动指挥交通也没办法。因此，这个自驾车只能是个部分自驾的逐渐发展进入实用，完全像google车这样的自驾还不能上市。
家园 once 红外=4 ever 红外 1.
first, analogy part:
1)老鼠brain能譜@around 红外
2)貓brain能譜>(much wider) 老鼠brain能譜
3)貓 can see through 老鼠,量子化正交分解老鼠brain, and 老鼠 can also see 貓, trying really hard guessing 貓 brain, with very low success rate, and eventually becoming emotional, giving up, being victimized again by 貓 with several more rounds of game down the road
2
now, with internet, goog,fb, bidu, etc, 老鼠 can share their 红外 data with each other, but 老鼠 are still largely played around by 貓 , losing money to 貓, but with a little "progress" for 老鼠 : 老鼠 tell each other, hey, folks, we have tried really hard, and we did a really good team work in this game , etc.
3.
now, goog/fb come in: hey, you 老鼠 folks, we know why you guys always lose money to cat, basically telling 老鼠 the story @above 1 & 2 part.
老鼠 to goog/fb: wait a minute, what the hell is going on here with you nerd guys, what is "量子化正交分解" etc, us老鼠 brain model is evolution based, we can't jump from 红外 to 量子化正交分解 overnight, not so fast, we just can't do that.
goog/fb to 老鼠: don't worry, we will make our AI user friendly/video based etc, with non-body contact, privicy protected technology, and with our global information real time processing network, we will upgrade your brain at least into near 民科 level, and with all these stuff, we hope you 老鼠 guys can play a much more fair game with cat folks, and we are working hard to make cat brain能譜 and their associated "sys admin" model become transparent in our coming platform, etc
4
老鼠 to goog/fb: sounds good, but how are you going to make money? Do we have to pay you and how do we know if you are not "united with cat" behind scences?
goog/fb to 老鼠: we are still working on the financial part of the model, but basically and like our current business model, cat folks have to pay us a fat advertising fees to get into the game, and you 老鼠 may or may not have to pay for AI 1.0 version, but may have to pay a little for AI of higher versions, etc
the bottom line is: in this AI assisted global information processing network, "value" will be traded much more based on value, much less on 忽悠.
5.
yes, "once 红外=4 ever 红外", may still be the outcome for most of the games, but 老鼠 in general now have a much better chance to beat 貓, and a few of 老鼠 will jump out of their conventional "红外=4 ever 红外" world, and into a a brave new world.
家园 在生物脑活动中，完全可能有超越现在科学理解的作用 在这个回复中有所讨论：链接出处
而河友晓兵的这个回复也非常有趣："小姑娘害羞":时间可达到1/1000纳秒链接出处
再讲一次，人脑比现有的电脑要具备高明很多的鲁棒性，那就是说，扰动很难产生实质性的信息处理上的区别。这是现在的电脑做不到的。但是，不管怎么鲁棒，还是存在一个小扰动就产生天上地下的差别的时刻和地方。
而晓兵的介绍，说明生物体甚至有价值地利用这些扰动，来完成很多看似不太可能的行动。这些都说明，在生物脑活动中，完全可能有超越现在科学理解的作用。而且，这些作用，可能在某个关键时刻被放大了出来，加入到时序记忆流中，从而很大的改造了时序记忆流。
家园 "现在的电脑做不到的:, gd intuition "再讲一次，人脑比现有的电脑要具备高明很多的鲁棒性，那就是说，扰动很难产生实质性的信息处理上的区别。这是现在的电脑做不到的。但是，不管怎么鲁棒，还是存在一个小扰动就产生天上地下的差别的时刻和地方"
1. very good intuition, I like that.
but to get further excited, you have to love theoretical physics they way you love "有花戴的人家的闺女", 闺女 here copied from "聊天大厅 自动刷新 可完整聊"
2. "扰动很难产生实质性的信息处理上的区别。这是现在的电脑做不到的"
because in most case of today's world, we are dealing with a world of "non sr macro with very little qm if any", so basically, a conventional heat equation takes care of everything, and it can perform heat equation opertions almost at light speed, and "put down" everybody else into the Gauss-Markov Heat Bath, "heatmblow up" their brain on their way into the heat bath, and they could never understand before and after their heat bath, "how come?"
3. for any meaningful further discussion on this subject, you have to get some
"有花戴的人家的闺女" come to the party, and the girl better to have some kind of theoretical physics brain, or we can't even get party started
------------------
Buchanan科学时评之带着时间赌博(下)
http://www.ccthere.com/alist/3830412
------------------
wiki:
For most of the last century, it was recognized that Fourier equation (and its more general Fick's law of diffusion) is in contradiction with the theory of relativity,[5] for at least one reason: it admits infinite speed of propagation of heat signals within the continuum field. For example, consider a pulse of heat at the origin; then according to Fourier equation, it is felt (i.e. temperature changes) at infinity, instantaneously. The speed of information propagation is faster than the speed of light in vacuum, which is physically inadmissible within the framework of relativity.
家园 孙昌璞 信息处理本质上是一个物理过程 1
http://power.itp.ac.cn/~suncp/researchc.html
2
ustc website, 段路明 郭光灿 量子信息讲座
本帖一共被 1 帖 引用 (帖内工具实现)