AI与教育的思考：幻觉与跃迁之间 Comments and Reflections on AI and Education

下面是我针对网友在一次对话节目中就有关AI和教育的问题发表的观点所做出的一些评论和想法。

当人们谈论人工智能（AI）时，首先想到的往往是“写作文”这一应用场景。确实，这是目前AI的一个重要应用方面，但它远不是全部，甚至只是AI众多应用中非常小的一个领域。更值得关注的是，许多人对AI的理解存在偏差。他们往往误以为AI已经具备逻辑推理能力，而事实上，AI目前并不具备真正的推理功能，它只是在“表示层”看起来像会推理，本质上仍处于“伪推理”的阶段。

在我看来，青少年在成长发育的过程中，更应及早地与AI共同成长。回想过去，有段时间批评学生“五谷不分”，于是让他们到农村去学农。可问题是，为什么一定要非得分清五谷？这反映出教育思维的某种局限。

中国从小学到大学，甚至博士教育，普遍采用“批发式”模式。这种模式让教师难以真正了解自己的学生，结果导致我们今天难以分辨一篇文章究竟是学生写的还是AI生成的。这不是AI的问题，而是教育方式的问题。

就像学中文的学生，若懂些书法，自然更好，能写毛笔字当然加分；但即使不会，也不妨碍他写文章。写字和写文章毕竟是两回事。在信息爆炸、知识膨胀的今天，一个人的知识结构和能力结构所能涵盖的范围是极其有限的。因此，在AI时代，对孩子的培养内容应当大胆更新和有选择地舍弃。对此，我非常赞同那位于老师的观点。

现在，ChatGPT已经逐渐成为用户的“外脑”。我们不再需要将所有知识都储存在自己的大脑中。未来的人机关系将发生革命性的突破。由此可见，评价AI不能局限于某一具体的应用场景——这是人类进化至今最为震撼的一次智力跃迁。

技术进步的总体趋势，是在解放人类的同时，也伴随着某些生物能力的退化。例如，尽管百米世界纪录不断被打破，但这主要得益于竞技机制对小概率事件的捕获，而人类整体的平均奔跑速度实际上是在下降的。再比如，人类寿命的显著延长，主要是由于工业化以来，饥荒减少、医疗和营养改善带来的结果，而非人类生理机能的增强。恰恰相反，人类的生物体机能比过去更脆弱，一场疫情就能造成上千万人的死亡。技术过去主要解放的是人的四肢，如今开始“动手”于人的大脑了。

我非常赞同卢老师在发言中所强调的“爱的教育”。这种思维方式正体现了一种“out of the box thinking”，即跳出传统框架的思维。

然而令人遗憾的是，以Geoffrey Hinton为代表的“连接主义”AI派系，似乎并不承认人类具有自由意志。他们相信，大语言模型的发展方向终将完全取代人脑。对此我无力全面否定，但我始终倾向于相信：人类总有一些东西是机器无法替代的。

正如我前面所提到的，人们对AI的另一种误解，是误以为它具有人类的情感能力。这其实是一种幻觉。AI在表现层面上似乎拥有情感，但它本身既无感情，也无自由意志，更没有自我价值观。它所表现出的能力，不过是一种“魔术效果”。讽刺的是，现实中并非所有人都具备这些能力，于是便常出现“真人不如机器智能”的现象。

The following are some of my comments and reflections made in response to a friend’s views during a discussion program on the topic of AI and education.

When people talk about artificial intelligence (AI), the first thing that often comes to mind is its application in writing essays. Indeed, this is currently one of AI’s major use cases. However, it is far from the whole picture—in fact, it is just a very small part of AI’s many potential applications. What’s more concerning is that many people have a fundamental misunderstanding of AI. They often mistakenly believe that AI already possesses logical reasoning ability, whereas in reality, it does not. At best, AI gives the appearance of reasoning at the representational level, but it essentially remains in a “pseudo-reasoning” stage.

In my view, it is especially important for young people to grow up alongside AI during their developmental years. This reminds me of a time when students were criticized for not being able to distinguish the “five grains” and were therefore sent to the countryside to learn farming. But really, why must one be able to distinguish the five grains? This reflects a certain limitation in educational thinking.

In China, education from primary school through university—sometimes even at the doctoral level—tends to follow a “mass production” model. This makes it difficult for teachers to truly understand their students, resulting in a situation where today, it’s hard to tell whether an essay was written by a student or generated by AI. This is not the fault of AI, but rather a problem with the educational approach itself.

It’s like students learning Chinese: it’s great if they understand calligraphy, and even better if they can write with a brush. But it doesn’t really matter if they can’t—writing characters and writing an article are two different things. In an era of information explosion and knowledge inflation, an individual’s cognitive and skill structure can only cover a very limited scope. Therefore, in the age of AI, we must boldly rethink and selectively discard certain traditional educational content when it comes to raising children. In this regard, I fully agree with Professor Yu’s view.

Today, ChatGPT is becoming an external brain for users. There’s no longer a need to store everything in our own minds. The future of human-machine relationships will undergo revolutionary breakthroughs. Hence, AI should not be judged solely based on specific application scenarios—this is the most astonishing intellectual leap in human evolution so far.

The general trend of technological progress is that while it liberates humanity, it also leads to a regression in some biological abilities. For example, although the 100-meter sprint record continues to be broken, this is largely the result of competitive mechanisms capturing rare statistical outliers, while the average running speed of humans has actually declined. Similarly, the significant increase in human lifespan is mainly due to reductions in famine and improvements in healthcare and nutrition since industrialization—not because of stronger human physiology. In fact, our biological resilience is weaker than before: a single pandemic can kill tens of millions. In the past, technology primarily liberated the human limbs; now it has begun to “work on” the brain.

I strongly agree with Professor Lu’s emphasis on “education with love.” This way of thinking truly represents “out of the box” thinking—breaking away from traditional cognitive constraints.

Unfortunately, the “connectionist” school of AI, represented by Geoffrey Hinton, appears to deny the existence of human free will. They believe that the development of large language models will eventually completely replace the human brain. I don’t have the means to refute this view entirely, but I tend to believe that there will always be something within humanity that machines cannot replicate.

As I mentioned earlier, one common misunderstanding about AI is the belief that it possesses genuine emotional capacity. This is a human illusion. AI may appear to have emotions at the representational level, but it actually has no feelings, no free will, and certainly no values of its own. What it demonstrates is merely a kind of “magical effect.” Ironically, not all humans possess these traits either, which is why we often see situations where real people seem less intelligent than machines.