近10年来，人工智能技术得到了科技与工业界的极大的重视，预示着人类文明将进入智能时代。但是，作为智能时代基础的“智能科学”还远未成型。本文从电磁物理信息感知技术的独特视角，讨论智能科学如何发展的一些见解，指出人类智能与外在世界互为对偶问题、相互不可分割的根本属性，因此按人工智能所应对的对象及关联学科分为数学、物理、心理、意识4个阶段。其中第1阶段解决智能形成的通用学习算法的数学理论，第2阶段发展应对物理世界的物理智能。以此为基础，第3阶段发展应对智能涉及社会群体的高阶智能，第4阶段研究自由意识的本质和人工智能能否形成意识的超智能问题。结合笔者电磁信息感知专业领域，提出向物理智能发展的微波视觉新概念、相关内涵以及关键技术的建议，以笔者团队在这一方向的前期工作为例，讨论了以物理智能为基础的智能科学的研究与发展。
In the past decade, artificial intelligence has attracted a great attention from both academics and industry, suggesting the transition of the human civilization into a new intelligence era. However, the "intelligent science" has not yet developed into the science foundation of the intelligent age. This paper, from a specific viewpoint of electromagnetic information sensing, discusses how the intelligent science might be developed in future, and points out that the human intelligence and the external world form a pair of inseparable dual problems, as the fundamental nature of the intelligent science. Therefore, the development of the artificial intelligence goes through four stages according to the objects and the related disciplines, namely, the mathematical intelligence, the physical intelligence, the psychological intelligence, and the free consciousness. In the first stage, the mathematical theory of a general learning algorithm is developed to form the basis of the intelligence. In the second stage, the physical intelligence is developed to deal with the physical world. Based on this, the third stage of development responds to high-order intelligence involving social groups, while in the fourth stage, the nature of the free consciousness and the question whether the artificial intelligence can finally become a conscious super-intelligence are studied. Within the author's own research field of the electromagnetic information sensing, a new concept of the microwave vision is proposed along the direction of the physical intelligence to explain its key ideas and essential problems and to paint the roadmap of its technological development. Taking the previous studies of the authors' team in this direction as an example, we further discuss the research and development of the physical intelligence as one of the specific directions of a general intelligent science.
[1] Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786):504-507.
[2] Silver D, Huang A, Maddison C J, et al. Mastering the game of Go with deep neural networks and tree search[J]. Nature, 2016, 529(7587):484-489.
[3] Krizhevsky A, Sutskever I, Hinton G. ImageNet classification with deep convolutional neural networks[J]. Advances in Neu-ral Information Processing Systems, 2012, 25(2):1097-1105.
[4] LeCun Y, Bengio Y, Hinton G. Deep learning[J]. Nature, 2015, 521(7553):436-444.
[5] Fukushima K. Cognitron:A self-organizing multilayered neural network[J]. Biological Cybernetics, 1975, 20(3-4):121-136.
[6] Jarrett K, Kavukcuoglu K, Ranzato M, et al. What is the best multi-stage architecture for object recognition?[C]//IEEE, Inter-national Conference on Computer Vision. Piscataway NJ:IEEE, 2010:2146-2153.
[7] Glorot X, Bengio Y. Understanding the difficulty of training deep feedforward neural networks[J]. Journal of Machine Learn-ing Research, 2010, 9:249-256.
[8] Glorot X, Bordes A, Bengio Y. Deep sparse rectifier neural net-works[J]. Journal of Machine Learning Research, 2012, 15:315-323.
[9] Choromanska A, Henaff M, Mathieu M, et al. The loss surface of multilayer networks[J]. Arxiv.org, 2014, arXiv:1412.0233
[10] Pearl J. Theoretical impediments to machine learning with sev-en sparks from the causal revolution[J/OL].[2018-03-20]. http://www.cse-lab.ethz.ch/wp-content/uploads/2014/09/Judea_Pearl_GREAT.pdf.
[11] Sabour S, Frosst N, Hinton G E. Dynamic routing between capsules[J/OL].[2018-03-20]. http://papers.nips.cc/paper/6975-dynamic-routing-between-capsules.pdf.
[12] LeCun Y. Predictive learning[EB/OL].[2018-03-20]. https://www.youtube.com/watch?v=Qi1Yry33TQE.
[13] Ali Rahimi's talk at NIPS[EB/OL].[2018-03-20]. https://www.youtube.com/watch?v=Ount2Y4qxQo.
[14] 周翔. Tomaso Poggio:深度学习需要从炼金术走向化学[EB/OL]. (2018-01-28)[2018-03-20]. http://www.sohu.com/a/219443091_697750. Zhou Xiang. Tomaso Poggio:Deep learning from alchemy to chemistry[EB/OL]. (2018-01-28)[2018-03-20]. http://www.so-hu.com/a/219443091_697750.
[15] 朱崧纯. 浅谈人工智能:现状、任务、构架与统一[EB/OL]. (2017-11-02)[2018-03-20]. https://mp.weixin.qq.com/s/-wSYLu-XvOrsST8_KEUa-Q.
[16] Chen S, Wang H, Xu F, et al. Target classification using the deep convolutional networks for SAR images[J]. IEEE Trans-actions on Geoscience & Remote Sensing, 2016, 54(8):4806-4817.
[17] Zhou Y, Wang H, Xu F, et al. Polarimetric SAR image classi-fication using deep convolutional neural networks[J]. IEEE Geoscience & Remote Sensing Letters, 2017, 13(12):1935-1939.
[18] Zhang Z, Wang H, Xu F, et al. Complex-valued convolutional neural network and its application in polarimetric SAR image classification[J]. IEEE Transactions on Geoscience & Remote Sensing, 2017, 55(12):7177-7188.
[19] Song Q, Xu F, Jin Y Q. Radar image colorization:Converting single-polarization to fully polarimetric using deep neural net-works[J]. arXiv.org, 2017, arXiv:1707.07225.
[20] YueD, Xu F, Jin Y Q. Deep despeckling neural network for SAR images[J]. International Journal of Remote Sensing, 2018, in press.
[21] Song Q, Xu F. Zero-shot learning of SAR target feature space with deep generative neural networks[J]. IEEE Geoscience & Remote Sensing Letters, 2017, 14(12):2245-2249.
[22] Xu F, Jin Y Q, Moreira A. A preliminary study on SAR ad-vanced information retrieval and scene reconstruction[J]. IEEE Geoscience & Remote Sensing Letters, 2016, 13(10):1443-1447.
[23] 徐丰, 王海鹏, 金亚秋. 深度学习在SAR目标识别与地物分类中的应用[J]. 雷达学报, 2017, 6(2):136-148. Xu Feng, Wang Haipeng, Jin Yaqiu. Deep learning as applied in sar target recognition and terrain classification[J]. Journal of Radars, 2017, 6(2):136-148.
