用户注册 登录
珍珠湾全球网 返回首页

岳东晓 -- 珍珠湾全球网 ... http://ydx.zzwave.com [收藏] [复制] [分享] [RSS] 岳东晓 -- 珍珠湾全球网

日志

人工智能赶上牛顿?

热度 1已有 7405 次阅读2022-3-9 04:21 |个人分类:科普|系统分类:教育

刚看到一篇文章介绍,有人构造了一个神经网络,输入太阳系各星体多年的位置与速度数据,网络得出了万有引力规律。机器学习真的要超过牛顿的天赋了吗?

我看还差得远。要知道,从地面上观察,你是不知道星体的位置与速度的。地球在绕太阳公转、地球还在自转,我们从小学的这些东西只告诉了最后的结论,怎么得出这些结论的,并没有讲。

从地面观察,你只看到一个个光点,不知道大小,距离,速度,而且好像它们都在绕着我们转。现在叫一个学过物理、数学的人,让他夜观天象,看能否独立发现行星轨道。我认为可能性几乎为零。

怎么验证人工智能的水平呢?应该输入若干年的星空照片,看人工智能能发现什么 --- 这已经是作弊了,古人是没有照相机可以准确记载星空的变化的。

即使输入星空照片,人工智能能发现行星轨道,也还是不能证明人工智能达到牛顿水平。神经网络只是用一个有非常多参数的非线性函数来进行数据拟合。而胡克与牛顿直接猜测出了平方反比律,然后牛顿倒过来进行了证明,在这样的作用力下轨道是开普勒之前算出的椭圆轨道,而且牛顿只用了最简单的数学(相当于现在的初中数学)。


路过

鸡蛋
1

鲜花

支持

雷人

难过

搞笑

刚表态过的朋友 (1 人)

 

发表评论 评论 (2 个评论)

回复 Wavelet 2022-3-11 12:53
A weird feeling about its stuff - GNN may have to clarify whether there is certain degree of dependency on its potential pre-existing knowledge of the system (solar?) that GNN is trying to learn from / analyze. Were such existence of knowledge to be found, it can essentially destroy the "learning" claim - a husband suspected a cheating spouse then investigated a potential affair v.s. the unknowing poor man picked up clues of infidelity from seemingly unrelated daily routines without divine enlightenment.

I would be very cautious about ML reaching the level of the most brilliant mind of mankind - I have seen tens millions of dollars spent on ML at my workplace with focus on much less grandeur topics with barely anything useful as a result. Put it nicely - it is not about learning Go and playing against pros. It is about inventing Go without knowing anything about gaming.
回复 岳东晓 2022-3-11 14:11
Wavelet: A weird feeling about its stuff - GNN may have to clarify whether there is certain degree of dependency on its potential pre-existing knowledge of the ...
  

facelist

您需要登录后才可以评论 登录 | 用户注册

Archiver|手机版|珍珠湾全球网

GMT+8, 2024-10-13 03:09 , Processed in 0.039045 second(s), 9 queries , Apc On.

Powered by Discuz! X2.5

回顶部