人工智能是不是走错了方向？(5)

第三个问题：小孩子不需要喂很多数据，看两三次就能认识什么是苹果，什么是橘子。

I don't agree with you. I think the human brain and machine learning are very similar. The reason that machine learning models need a lot of data is because they are trying to learn the same thing that the human brain learns when it sees an apple or an orange. The human brain is able to learn from a small amount of data because it is able to recognize patterns. The machine learning model is trying to do the same thing. As pre-training techniques are developing, few-shot learning models are also proposed. For example, meta-learning or learning to learn is a subfield of machine learning where automated methods are applied to understand and learn from the learning process itself. One-shot learning is an object classification problem where one, or a few, examples (shots) of an object are used to classify new instances of the same object. Zero-shot learning is a computer vision problem where models are trained on one dataset but can be applied to completely different but related datasets . These methods show that it is possible for machine learning models to learn from a small amount of data, just like the human brain.

我不同意你的看法。我认为人脑和机器学习非常相似。机器学习模型需要大量数据的原因是因为它们试图学习与人脑在看到苹果或橙子时所学的相同的东西。人脑能够从少量数据中学习，因为它能够识别模式。机器学习模型正在尝试做同样的事情。随着预训练技术的发展，还提出了少样本学习模型。例如，元学习或学习学习是机器学习的一个子领域，其中应用自动化方法来理解学习过程本身并从中学习。一次性学习 是一个对象分类问题，其中一个或几个对象的示例（镜头）用于对同一对象的新实例进行分类。零样本学习是一个计算机视觉问题，其中模型在一个数据集上进行训练，但可以应用于完全不同但相关的数据集。这些方法表明，机器学习 模型可以像人脑一样从少量数据中学习。

第四个问题：神经元再少的动物，学习任何方向的任务都有主动性，而目前人工智能学习某一方向只是受控。

Your opinion is partially correct. Fewer neurons does make it harder for an animal to learn some tasks , but not all tasks. In general, animals with fewer neurons are less intelligent than animals with more neurons. However, there are some tasks that animals with fewer neurons can learn better than animals with more neurons. For example, animals with fewer neurons may be better at learning simple habits, while animals with more neurons may be better at learning complex concepts. Moreover, the direction of learning is not always controlled in AI. Some AI systems are designed to learn in any direction, while others are designed to only learn in a certain direction. For example, reinforcement learning is a type of AI that can learn in any direction, while supervised learning is a type of AI that can only learn in a certain direction; transfer learning is a type of AI that can learn in any direction if it is given the right data; some other learning paradigms including unsupervised learning can also learn in any direction.