Thinking about the nature of AI

 Foreword

In recent days and a friend discussed at the present stage of development and the future of artificial intelligence, and carefully reread 尤瓦尔赫拉 Lee's "A Brief History Trilogy" produced some new ideas about AI, and feel the need to organize come out.

 

The nature of the program, AI's

Modern computers are based on the Turing machine, von Neumann architecture and implementation, and operation of the program in which is composed of two parts: 

Algorithms + Data Structures = Programs 

Refers to the code algorithm (herein "algorithm", "code", "logic" is a synonym), the program data is input to the algorithm, then the arithmetic operation output data, including the current artificial intelligence, its essence is true , but the algorithm is complex, only bigger amount of data.

 

Algorithm program was conceived before the program run by the programmer and implemented, will not change after the program is running, unless the programmer to manually modify the algorithm.

 

Accordingly, all the procedures are now processing the data input, the data are output. But the human brain is not only data processing capability, as well as logical computing power. Logic can be derived from some or certain scene data, derived law. At this stage of the computer in this capacity is zero.

 

Be emotional score:

	The human brain	computer
logical ability	10	0
Data computing power	1	10000

For example logic 

Most of the problems encountered in human life, work belong to logic. Without taking into account the reality of complex situations, for example a very intuitive mathematics:

List all the combinations

For example, come up with 3 balls from 1-6 balls, list all the possible cases 

Humans can combine their knowledge to come up with a variety of methods, such as 

A first method: a take from the start, all the listed cases, and then taken from all the cases of 1, then take a list of all cases, in all cases finally to take in all of the two balls a re-take, all cases exhaustive.

The second Methods: 0-63 (corresponding to 111111 binary 63), then converted into a binary number, then the number of binary number 1, and if it is 3, then that corresponds to a situation, then available as a binary number of columns -

The third method: See data generated in the second method, a pattern can be found:

    a. First 3 on the leftmost 1,

    b. Then find the leftmost contiguous 1, its rightmost 1 will move to the right one, if there are remaining 1 moves to the left, to get a new number.

    c. b process continues until all of the 1 in the far right.

There are basic algorithm programmer can quickly find the most efficient method of 3, 1 too take up memory, Method 2 is too occupied algorithm time.

A third method we have found that by law, a small amount of data from the logic, whereby the algorithm.

Below this is the third method of language go: 

 https://gitee.com/xiangism/blogData/blob/master/math/combination.go 

A computer program is now possible to reflect the above process, the law in the blocks is impossible to reach, but also impossible to write the algorithm used to generate program data.

If the programmer wants to Which the challenge, given here three additional sets of data, the program written, can not be changed, to see whether the predicted number generation logic or later. And similar data can also have various modifications.

This is just a very digitization of the above examples, in fact, think about all the problems encountered in our lives almost have to rely on human logic from the data obtained from the logic to deduce new logic. Human learning, life, work is to obtain a logical, deductive process of logic, application logic. 

After humans have logic skills, you can deal with a variety of different situations (equivalent to different situations of different data), specific ways of doing things (algorithm) according to the prevailing scenario evolved apply to the scene. And now only deal with artificial intelligence programmer preconceived scene for scene has never come out of it is powerless. 

And the programmer can not advance all the circumstances are taken into account, because the world is full of uncertainty, the only constant is change. 

Human logic can lead to something drawn a clear, concise rules from complex data in order to guide future human activities. 

From Pythagoras Theorem to calculus, Newton's laws of motion to Einstein's theory of relativity, from the Schrodinger equation to Yang - Mills theory, 
mankind is through the powerful logic skills, summed up the formula from the experiments, observations, data , laws and theorems (may be referred to as: algorithm), 
thus creating a brilliant civilization now.

 

Why is the computer logic skills 0 

If you read the above problem, in fact, you can get the answer, "computer" is used to calculate, rather than logical deduction. 

如果想让计算机程序能够处理某个事物，必须对其进行数据建模、建立对应的数据结构，而现在没有能对算法进行建模的理念基础，不能将算法有效地表示出来，也不能对算法进行运算了。 

仔细考虑人脑从数字中找规律的过程，可以发现也是在认识范围里尝试各种可能性（各种算法），这和计算机穷举数据类似，只不过计算机是操作的数据，而人脑操作的是算法。 

所以，只要能将算法有效地建模就可以实现计算的逻辑能力。

 

人机深度结合的可能性

如果不实现计算机逻辑能力的从0到1过程，其计算能力和人脑逻辑能力的合并也只能是空想。 

因为现在计算机不懂人脑是如果逻辑思想，只会按照代码指令忠实地去运行。 

而人脑也不懂人工智能的运作机制：深度学习从海量数据中统计出一个最优值，人脑也无法理解其具体的运算过程。

 

喊话赫拉利

赫拉利的“简历三部曲”被本人分在人文类书籍里，除去人类智能部分，他对人类历史、现在与未来的思考部分值得我们参考，但几乎不认同他对AI的预测。

人工智能不可能开办公司

如果开公司的所有情况都能事先被编码好，仅仅会处理数据的人工智能根据这些规则就能开公司、赚钱的话，我只能说赫拉利把现实世界想得太过简单了。 

不会被人工智能替代的职业

数据处理、逻辑推演本质上是数学问题，而人类生活中遇到的事更多的不是数学问题。我大致总结以下类型的职业都不会被人工智能替代： 

1. 需要逻辑的行业。程序员

2. 人类还没有完全搞懂的行业。医生

3. 服务行业。客服、前台、导游、理疗师、健康教练、健康顾问

4. 充满未知、变数的行业。司机、律师、谈判专家 

程序员

2050年肯定不会出现程序员失业的情况，甚至连会一点点写程序的人工智能都不会出现。出现的仅仅是生成代码模板的IDE而已。 

司机

与飞机相比，路面的情况就复杂很多。所以飞机的自动驾驶在几十年前就出现了，随着人工智能将尽可能多的情况都考虑进去，路面上的自动驾驶自动性提高到99%了。但世界总是充满未知、充满不确定性。 

2050年肯定不会有哪个国家或地区不需要人考驾照。 

并且不管别人怎样，在我一生之中肯定不会做这样的事：自己在睡觉，将车完全交给人工智能去开。 

医生

人类对自己身体的了解还太少了，2050年肯定不会出现一家没有医生、只有人工智能的医院。  

客服

顾客找到客服人员，可能有这么几种情况： 

1. 遇到了产品中的问题需要解决，这种在一定程度上可以用人工智能来解决，但只要是一个稍微大点的产品或系统，其出面对用户时出现的问题就千奇百怪，并且随着产品的升级，新问题和旧问题交织在一些时，就不是人工智能能够面对的了。

2. 对产品提建议。如果一个用户可以对产品提建议，那么说明其对产品已经有感情了，这时其多半想面对的是人工，而不是冰冷的人工智能。

3. 吐槽。用户本来就已经对产品产生了一定程度上的厌恶，如果不能和真人进行一番沟通的，而是面对一个机器，很难想象这个用户还会继续使用该产品。

4. 纯属无聊，就是想调戏客服。这种情况对于一个想有好口碑的公司来说，还是用人工会比较好，哪怕浪费的人力资源。 

2050年不会有淘宝卖家全部都使用人工智能小二，并且人工客服的使用比例也不会比2019年少太多。 

前台、门卫

这类职业是公司的脸面，并且面对的事务也会比较复杂，所以不可能被替代。  

理疗师、健康教练、健康顾问

这类职业往往顾客看中的是人的品质，如果他自身是个体弱多病、瘦骨伶仃的人，就算掌握的健康知识再多、再丰富，恐怕也得不到顾客的认同。

律师、谈判专家、基金经理

这类职业面对的复杂性比我们想象的要大很多。

猜想

人脑无时无刻都在“编程”，也就是生成各种算法，然后将算法存储起来，得到合适的时机时再拿出来使用。 

人类的人脸、图像识别这么快，可能是因为对每个认为的人脸都有生成一个特定的算法，在找人时就是用的那一个算法去找。对认识的不同人，每个人都生成不同的识别算法。 

意识只不过是一系列逻辑、算法的集合。 人脑用逻辑去思考“我是谁，我从哪里来，我要去哪里”之类的问题后，于是就有了意识。 

总结

1. 图灵机最大的缺陷：只能对数据进行处理，算法的所有逻辑都是程序员写死，算法不能自我演化。

2. 如果想改变这种现状就得创立新的数学分支，使得它可以对算法进行编码，实现操作算法的算法。

3. 想要人机深度结合必须要实现2。

4. 如果人类或宇宙不是超级智能体的杰作，那么人类是能够在数学层面上实现操作算法的算法。但愿这样的数学分支能在100年出现。

5. 既然实现自动智能如此之难，我们还是得脚踏实地在现在的AI基础上有所作为，除非你的脑袋被上帝丢的苹果砸过。

6. 虽然现在AI只能对数据进行处理，但它的应用场景随着大数据、Iot的到来也会普及到人类活动的方方面面中，但也不必太对其过于恐惧，毕竟一个严重偏科生不可能取代一个发展平衡的好学生，最多是成为其助手。

7. 随着5G时代的到来，会大大加快Iot的进程，使万物互联成为可能。只会处理数据的AI(如果为了商业需要还是叫AI的话)会大显身手。