一
当下，大模型表现出了智能涌现的能力，很多人认为，5年内，可以涌现出来一个有自由意志的版本。
OpenAI的首席科学家伊利亚表示：ChatGPT可能拥有自我意识。被誉为AI教母的李飞飞则表示，大模型不存在主观感觉能力，多少亿参数都不行。
很多人可能觉得，那么多大咖都在争论，你凭什么来说。
且听我慢慢说来。 数字化转型网（www.szhzxw.cn）
或者有人会说，人可以感知肚子饿，机器人也可以知道自己电量不足，扫地机器人都会在电量不足的时候，主动去充电。但这不是“主观感觉”吗？
他们争论的是感知，一种主观的感觉。讨论这个问题，或许真需要一点“佛性”。
你闭上眼，感受微风吹过，感受一下自己的自我意识，或者说“灵魂”，是不是在双眼之后，鼻子上面一点的地方？这就是所谓的“眼观鼻，鼻观心”。所谓“我思故我在”的感知。
对人工智能来说，可以安装传感器。如果有一个大模型操控的机械臂，上面有传感器，机械臂靠近火源，传感器会感觉到热，这个信号会传递一个热的Prompt，大模型会说，“我差点烫到了”，然后缩回手臂。这都是现实的技术。
但是，大模型真的感到了热和痛吗？它只是从它收集到的语料中发现，热对应这句话。
或者，大模型的服务器里有传感器，它会开玩笑，“你的这个问题，让我激动得CPU都烧热了”。但它真的有喜怒，真的能感受到CPU的热吗？ 数字化转型网（www.szhzxw.cn）
二
这是哲学史上的经典问题。
计算机之父，在上世纪50年代就研究过“机器能否思考”这个问题，并提出了图灵实验。图灵认为，如果一台机器在感知方面表现得跟人完全一样，那么，从本体论层面，去争论它到底有没有主观感知，就没有意义，我们就把它当做有主观感知就可以了。
简单地说，图灵认为，有没有无所谓，足够像就有了。这也是图灵实验的原理。
AI的过程，其实可以全部换为蒸汽机和机械开关，那么一个巨大的蒸汽机，咔咔转动的机械，会是活的吗？
一支笔无数稿纸，理论上都可以重现AI的过程。虽然这个过程时间会耗费很长很长。那么，这个过程会产生意识吗？对于这个问题，我前几年认为，不能。现在我的想法有了一些改变，工艺，可能很重要。
也就是说，一秒能写100亿个字的笔，和一秒只能写一个字的笔，可能的确会产生本质的不同。像星球一样大的，蒸汽机推动的，一秒只能运行1000次的计算机，和4纳米的一秒运行100亿次的计算机，会产生本质的不同。 数字化转型网（www.szhzxw.cn）
意识是达到某个复杂度的信息处理系统的涌现。自我、意义，都是复杂演化系统演化出来的适应性“幻觉”。只要系统足够复杂，的确可能涌现出意识。
但AI有意识和人类能观测、确定AI有意识是两回事。就像一个量子同时具有速度和位置，但人类不可能同时观测到。
AI是否会产生“自由意志”，比这个更基础的问题是，AI是否有“自我意识”，而“自我意识”的下一层是“意识”“感觉”。只有当AI有“感觉”之后，才有可能有“自我意识”“自由意志”。
类比的就是植物没有感觉，昆虫、爬行动物可能有“感觉”，高等哺乳动物有“自我意识”，而人类则讨论“自由意志”。意识是一切的基础。
其实，自由意志是什么，自我意识是什么，意识是什么，可能永远没有标准答案。别说证明机器有，一个人要证明另一个人有感觉都很难。 数字化转型网（www.szhzxw.cn）
这个问题看起来是，人有感觉，但要由人类到机器，从碳基到硅基去探索机器是否有感觉，遇到了困难，产生了争议。但其实，真正困难的地方，并不在于这个过程。而是人的感觉本身。
三
人有意识吗？一个人可以自知自己有意识，但怎么知道他人有意识？这不是一个不言自明的真理。
人判断另一人是否有意识，推导过程是：我能确定自己的存在，我有意识，我有大脑；他有大脑，他也有意识。这个推理过程，并不严密。因为人类不知道大脑是如何产生意识的。
人觉得，除了自己之外的人类都有感觉、有意识，这不是严密的逻辑，而是经验主义的推己及人。
抛开缸中之脑这些假设，一个人不难确定自己是存在的，但他怎么确定周围的人是有意识的呢？或者他接入的是一个巨大的虚拟游戏，自己的亲朋好友都是NPC，即电脑控制的角色。
反过来，在其他人意识中，也无法确定你是不是一个NPC。
这是因为意识具有不可观测性与私有性。
一个人永远无法知道另一个人是否真正有感知。讨论这个问题前，先讨论一个准备性的问题：你看到的红色和别人看到的红色一样吗？你眼中的红色，和我眼中的红色，是一样的吗？
由于生理性或病理性原因，一个人的两只眼睛，可能会有轻微的色差。色彩简单地说，深一些、浅一些。白色，可能黄一些、蓝一些。两只眼睛通向同一个大脑，这很容易被感知到。你现在就可以试试。
所以，不同眼睛看到的颜色是不一样的。那么，一个人看到的颜色色温更高，比如，一般人眼中6500K的白色，在他眼中实际上和一般人7000K的白色差不多。他永远无法知道这个差别，也无法告知别人这个差别，因为他没有办法把脑子里的东西抠出来，给别人看。
色盲、色弱，是通过比较的方式“外化”为可以表述的东西。比如，分不清绿色和红色，那么，绿色方块中的红字，你就不能分辨。但有些差异，是无法外化的。推到极端，你永远无法知道另一个脑中的红色，到底是什么样子的。 数字化转型网（www.szhzxw.cn）
打个比方，我们眼中的红色，在某个人眼中，很可能是绿色。但这不会影响他的生活。当他年幼时，老师指着一张图片告诉他，这是红色，红色这个词，这个感觉，就和他脑中的那个颜色建立了一一对应关系。
颜色深浅，是一个最简单的感知，都无法通过严密的逻辑对应，至于感知，这种东西，就更难了。人永远无法知道当一只鱼的感觉，哪怕是VR也只能模拟鱼的视角，而无法模拟鱼的感知。
现在流行所谓的生命上传。假如生命上传，传完了，那个被上传记忆塑造的人说，他有了你所有记忆、性格，他会觉得，自己已经被上传过来了，然后告诉工作人员，“你们可以杀死我原来的身体了”。
实验人员走过来说，我们要杀死你。原来的你，也就是现在在看这篇文章的你，看着另外一个你，比你更年轻的你，一个“他者”。你接受自己被杀死吗？
科幻小说中的瞬移门，如果是基于传递信息的，其实也是杀掉进去的这一个，在另一处组合一个。那么，你进不进？
实际上，一个人甚至都没办法证明，现在自己的自我意识，就是过去自己的自我意识。人都觉得自己的意识是稳定的，20年前的小孩，和现在的自己，是共有一个意识的。这种意识的稳定性，是虚假的吗？
今日之我是昨日之我吗？所有人类觉得自己一直存在，对意识的连贯性的肯定，其实源于身体的连贯性。其实这也只是一种经验，而非精确的逻辑。
早上，你醒过来，新的一天开始了。但这个时候的你，和那个刚刚被灌入意识的人，有什么区别呢？人无时无刻在进行新陈代谢。过去的身体会变为二氧化碳、尿素排出去，食物会变为新的身体。去年自己的“自我意识”可能已经死掉了，现在的自己，不过继承了去年自己的记忆而已。就像那个从生命上传设备走下来的那个人。 数字化转型网（www.szhzxw.cn）
四
这就是著名的忒修斯之船。
普鲁塔克记载的古希腊传说中， “忒修斯与雅典的年轻人们自克里特岛归还时所搭的30桨船被雅典的人留下来作为纪念碑，随着时间过去，木材也逐渐腐朽，而雅典的人便会更换新的木头来替代。最后，该船的每根木头都被换过了。因此，古希腊的哲学家们就开始问：‘这艘船还是原本的那艘忒修斯之船吗？如果是，但它已经没有最初的任何一根木头了；如果不是，那它是从什么时候不是的？’”。
哲学家Thomas Hobbes后来对此进行了延伸——如果用忒修斯之船上取下来的老部件来重新建造一艘新的船，那么两艘船中哪艘才是真正的忒修斯之船？
在证明今日之我是昨日之我之前，人无法证明他人有意识。在人类证明他人有意识之前，人无法证明AI是否有意识。
至于自我意识就更复杂了。一个自我意识和另一个自我意识的差别，是分子级别的，还是原子级别的，是组成两个意识的特定原子不同，还是在宇宙中的绝对位置？
打个比方，如果意识是一个杯子的高度和直径，哪怕精确到纳米，人类可以造出两个一模一样的杯子。但如果是指特定的一组原子，则这个杯子就是唯一的。如果是后者，那永远不可能有一模一样的杯子，就不可能有一模一样的意识。 数字化转型网（www.szhzxw.cn）
这就是意识的私有性，所谓意识上传，就是破解意识的私有性。但人类还搞不清楚意识的产生机制，更不用说分清楚意识的私有性，实现上传。
如果人类能证明、定义一个意识，意识就不再是私有的。这就是人类达成永生的这一天。但这一天可能永远无法到来，也就是说，AI的意识的确可能出现，但人类永远无法证明，就像人类也永远无法知道另一个人类是否有意识。
用一种科学的方法论说，无法感知、无法检测的东西，就是没有意义的。如果我们无法检验世界是不是虚拟的，那么虚拟和非虚拟这个问题，就没有意义，采纳最简洁的设定，在此基础上进行研究、活动、生产。
同类例子就是光速不变原理。单向光速是不可测量的，因为信息不可能比光速更快，人类只能测量往返光速。这就产生两个假设，第一，光速往返速度不同，去的时候快，返回的时候慢；第二，往返光速都是一样的。显然，第二个假设更简单，于是采纳最简单的那个：真空光速各向同性。
所以，如果现在要证明AI有意识，要有严密的逻辑。由于意识的无法观测性、私有性，是很难去找到证据的。 数字化转型网（www.szhzxw.cn）
争论AI是否有意识，就像去争论光速往返的速度是否一样，是一个没有意义的问题。
从经验角度看，图灵实验就够了，计算机之父已经解决了这个问题，不必再争了。
翻译：
Debating whether AI is conscious is pointless
One
Now, large models are showing the ability to emerge intelligently, and many believe that within five years, a version with free will could emerge.
Ilya, OpenAI’s chief scientist, said ChatGPT may be self-aware. Li Feifei, known as the godmother of AI, said that the large model does not have subjective feeling ability, and how many billions of parameters are not good.
Many people may think that so many big coffee are arguing, why do you say.
But listen to me slowly. 数字化转型网（www.szhzxw.cn）
Or some people will say that people can sense hunger, the robot can also know that its power is insufficient, and the sweeping robot will take the initiative to charge when the power is insufficient. But isn’t that “subjective feeling”?
What they’re arguing about is perception, a subjective feeling. To discuss this issue, perhaps we really need a little “Buddha-nature”.
Close your eyes, feel the breeze, feel your sense of self, or “soul,” behind your eyes, a little above your nose? This is the so-called “eye to nose, nose to mind”. The perception of “I think, therefore I am”.
For artificial intelligence, sensors can be installed. If you have a robot arm operated by a large model, with sensors on it, and the robot arm is close to the source of the fire, the sensor will sense heat, and this signal will transmit a hot Prompt, and the large model will say, “I almost burned it,” and withdraw the arm. These are realistic techniques. 数字化转型网（www.szhzxw.cn）
But did the big model really feel heat and pain? It simply found, from the corpus it had gathered, that heat corresponded to this sentence.
Or, a large model of the server has sensors, it will joke, “You have this question, I am excited about the CPU is burning up.” But is it really moody and can it really feel the heat of the CPU?
Two
This is a classic question in the history of philosophy.
The father of the computer, in the 1950s, studied the question “can machines think” and proposed the Turing experiment. Turing argued that if a machine behaves exactly like a human in terms of perception, then there is no point in arguing about whether it has subjective perception at the ontological level, and we should just treat it as if it has subjective perception.
In short, he believed that it did not matter if there was one, but if there was one. This is also the principle of Turing’s experiment. 数字化转型网（www.szhzxw.cn）
In fact, the process of AI can be all changed into steam engines and mechanical switches, so a huge steam engine, a rattling machine, will be alive?
A pen with countless pieces of paper can theoretically reproduce the AI process. Although this process can take a long, long time. So, does this process produce consciousness? For this question, I thought a few years ago, no. Now I’ve changed my mind a little bit, and the craft, it might be important.
That is to say, a pen that can write 10 billion words a second may indeed be substantially different from a pen that can write only one word a second. A computer the size of a planet, powered by a steam engine, that runs only 1,000 times a second, will be fundamentally different from a computer that runs 10 billion times a second at 4 nanometers.
Consciousness is the emergence of information processing systems that reach a certain complexity. Self and meaning are adaptive “illusions” evolved from complex evolutionary systems. If the system is complex enough, consciousness can indeed emerge.
But AI consciousness and humans being able to observe and determine AI consciousness are two different things. It’s like a quantum that has both velocity and position, but it’s impossible for humans to observe at the same time. 数字化转型网（www.szhzxw.cn）
The more fundamental question of whether AI will produce “free will” is whether AI has “self-awareness,” and the next level of “self-awareness” is “consciousness” and “feeling.” Only when AI has “feeling” can it have “self-awareness” and “free will”.
The analogy is that plants have no feelings, insects and reptiles may have “feelings,” higher mammals have “self-awareness,” and humans talk about “free will.” Consciousness is the foundation of everything.
In fact, there may never be a standard answer to what free will is, what self-awareness is, what consciousness is. It’s hard for a person to prove that another person has feelings, let alone that a machine does. 数字化转型网（www.szhzxw.cn）
The problem seems to be that people have feelings, but to explore whether machines have feelings from humans to machines, from carbon-based to silicon-based, has been difficult and controversial. But the real difficulty is not in the process. It’s the human feeling itself.
Three
Are people conscious? A person can know that he is conscious, but how can he know that someone else is conscious? This is not a self-evident truth.
The reasoning for determining whether another person is conscious is: I can be sure of my existence, I have consciousness, I have a brain; He’s got a brain. He’s conscious. This reasoning process is not rigorous. Because humans don’t know how the brain produces consciousness.
People think that all human beings besides themselves have feelings and consciousness, which is not strict logic, but empirical self-promotion. 数字化转型网（www.szhzxw.cn）
Aside from the assumptions of a brain in a jar, it is not difficult for a person to be sure that he exists, but how can he be sure that those around him are conscious? Or maybe he’s plugged into a huge virtual game where his friends and family are NPCS, computer-controlled characters.
In turn, in other people’s minds, it is impossible to determine whether you are an NPC or not.
This is because consciousness is unobservable and private.
One can never know if another person is truly sentient. Before we get to that, let’s ask a preliminary question: Do you see red the same way other people see red? Is the red in your eyes the same as the red in my eyes? 数字化转型网（www.szhzxw.cn）
Due to physiological or pathological reasons, a person may have a slight color difference in both eyes. The colors are simply darker and lighter. White, maybe a little yellow, a little blue. Both eyes lead to the same brain, which is easily perceived. You can try it now.
So, different eyes see different colors. Then, a person sees a color with a higher color temperature, for example, the average person’s 6500K of white is actually about the same as the average person’s 7000K of white. He could never know the difference, nor could he tell anyone else about it, because he could not extract what was in his head and show it to others. 数字化转型网（www.szhzxw.cn）
Color blindness and color weakness are “externalized” into something that can be expressed by means of comparison. For example, if you can’t tell the difference between green and red, then you can’t tell the red letter in the green square. But there are some differences that can’t be externalized. Taken to the extreme, you never know what the red in the other brain looks like.
For example, what we see as red is likely to be green in someone’s eyes. But it doesn’t affect his life. When he was young, his teacher pointed to a picture and told him that it was red, and the word red, the feeling, established a one-to-one correspondence with the color in his head.
The depth of color is a simple perception, which cannot be matched by strict logic, and perception is even more difficult. People can never know how to feel when a fish, even VR can only simulate the perspective of the fish, but can not simulate the perception of the fish.
There’s something called life uploading. If the life upload, the transmission is finished, the person who is shaped by the uploaded memory says that he has all your memories, personalities, he will feel that he has been uploaded, and then tell the staff, “You can kill my original body.”
The experimenter came and said, We’re going to kill you. The original you, who is reading this now, is looking at another you, a younger you, an “other.” Do you accept being killed?
The teleportation door in science fiction, if it is based on transmitting information, is actually to kill the one that goes in and combine one in another place. So, are you in or not?
In fact, one cannot even prove that one’s present self-awareness is one’s past self-awareness. People think that their consciousness is stable, and the child 20 years ago, and now they share the same consciousness. This stability of consciousness, is it false?
Am I who I was yesterday? All human beings feel that they have always existed, and the affirmation of the coherence of consciousness comes from the coherence of the body. In fact, this is just a kind of experience, not exact logic. 数字化转型网（www.szhzxw.cn）
In the morning, you wake up and your day begins. But what is the difference between you at this time and the person who has just been put into consciousness? People are metabolizing all the time. The old body will become carbon dioxide, urea will be discharged, and food will become the new body. Last year’s “self-consciousness” may have died, and the present self has only inherited the memory of last year. Like the guy who came down from the life upload device.
Four
This is the famous ship of Theseus.
According to the ancient Greek legend recorded by Plutarch, “The 30-oar boat in which Theseus and the young men of Athens returned from Crete was left as a monument by the Athenians, and as time went on, the wood decayed and the Athenians replaced it with new wood. In the end, every piece of wood on the ship was replaced. So the philosophers of ancient Greece began to ask, ‘Is this the same ship of Theseus?’ If it is, it no longer has any of the original logs; If not, when did it stop? ‘”.
The philosopher Thomas Hobbes later went so far as to ask which of the two ships would be the true ship of Theseus if a new ship were to be built from the old parts taken from the ship of Theseus?
One cannot prove that another is conscious until one can prove that I am today what I was yesterday. Until humans prove that others are conscious, humans cannot prove that AI is conscious.
Self-awareness is more complicated. Is the difference between one self-consciousness and another at the molecular or atomic level, the difference between the specific atoms that make up the two consciences, or the absolute position in the universe?
For example, if consciousness is the height and diameter of a cup, even down to the nanometer, humans can make two identical cups. But if you’re talking about a particular set of atoms, then this cup is unique. If it is the latter, then there can never be exactly the same cup, there can never be exactly the same consciousness. 数字化转型网（www.szhzxw.cn）
This is the privacy of consciousness, the so-called consciousness upload, is to crack the privacy of consciousness. However, humans still do not understand the mechanism of consciousness, let alone the separation of consciousness and the realization of uploading.
If man can prove and define a consciousness, consciousness will no longer be private. This is the day when human beings achieve immortality. But that day may never come, that is, AI consciousness may indeed emerge, but humans will never be able to prove it, just as humans will never know whether another human is conscious. 数字化转型网（www.szhzxw.cn）
According to a scientific methodology, what cannot be perceived and detected is meaningless. If we cannot test whether the world is virtual or not, then the question of virtual and non-virtual is meaningless, adopt the simplest setting, and conduct research, activity, and production on this basis.
A similar example is the constant speed of light principle. The speed of light in one direction is not measurable, because information cannot travel faster than the speed of light, and humans can only measure the speed of light in the round trip. This gives rise to two hypotheses. First, the speed of light goes round and round differently, faster on the way out and slower on the way back. Second, the speed of light is the same for each round trip. Obviously, the second hypothesis is simpler, so the simplest one is adopted: vacuum light speed isotropy.
Therefore, if we want to prove that AI has consciousness now, we need to have strict logic. Because of the unobservable, private nature of consciousness, it is difficult to find evidence.
Debating whether AI is conscious is like debating whether the speed of light travels round and back is a meaningless question. 数字化转型网（www.szhzxw.cn）
From an empirical point of view, the Turing experiment is enough, the father of the computer has solved the problem, there is no need to argue.
本文由数字化转型网（www.szhzxw.cn）转载而成，来源于秦朔朋友圈；作者：刘远举；编辑/翻译：数字化转型网宁檬树。
免责声明: 本网站(http://www.szhzxw.cn/)内容主要来自原创、合作媒体供稿和第三方投稿，凡在本网站出现的信息，均仅供参考。本网站将尽力确保所提供信息的准确性及可靠性，但不保证有关资料的准确性及可靠性，读者在使用前请进一步核实，并对任何自主决定的行为负责。本网站对有关资料所引致的错误、不确或遗漏，概不负任何法律责任。
本网站刊载的所有内容(包括但不仅限文字、图片、LOGO、音频、视频、软件、程序等) 版权归原作者所有。任何单位或个人认为本网站中的内容可能涉嫌侵犯其知识产权或存在不实内容时，请及时通知本站，予以删除。
