试论深度学习技术对人类社会持续发展所造成的风险

Deep Learning Poses Potential Threat to the Long-term Development of Human Civilization

深度学习机制的实质,便是通过大量数据训练的方法,使得人工智能系统能够在一个肤浅的层面上模拟人类在特定输入与特定输出之间建立的映射规律,与此同时,却对人类做出此类归类活动的宏观认知架构不闻不问。深度学习机制无法灵活地应对缺乏既有数据支持的新问题求解语境,因此,人类对于此类技术的过分依赖必然会导致人类社会应对新问题的应变能力不足。同时,此类依赖对于人类教育体制的侵蚀,也会在长远上使得作为数据标注活动得以可能的人力资源受到削弱,并由此使得人类的人文资源枯萎。与之相比较,目前尚且不那么主流的基于认知建模的通用人工智能技术,或许才是能够保证人类社会健康发展的新技术发展方向。

The very nature of deep learning technology is to build an Artificial Intelligence system that can mimic human behavior of recognizing a certain sort of inputs as a certain sort of outputs.But it can only mimic human behavior in a limited and superficial manner,as it does not address the global cognitive architecture underlying human behavior of recognition.And the training of any deep learning system is impossible without a significant amount of training data,whereas humans can respond to novel challenges by tolerating a small amount of relevant experience.The excessive reliance on deep learning may marginalize the necessity of keeping the traditional human education system running,and this will in turn undermine the human resources supporting the annotation of training data,without which many deep learning systems will stop functioning.And such reliance will also make social constructions more hidebound than it should be,since novel proposals lacking enough data-evidence will never be encouraged in such constructions.In contrast,Artificial General Intelligence based on modeling human cognitive architecture may exhibit a new technological approach that may elude most of the aforementioned problems.