更多需要被解决的问题
在蒙特利尔举行的2015年NIPS大会的“Brains,Minds and Machines”(大脑、思维和机器)座谈会,以及2016年巴塞罗那NIPS大会的“Bits and Brains”(比特与大脑)研讨会期间,戴米斯·哈萨比斯和我在关于人工智能未来,以及下一个优先研究重点的问题上进行了激烈的辩论。人工智能中还有许多开放的问题需要解决。最重要的就是因果关系问题,它提供了最高层次的人类推理,以及行动的意向性问题,这两者都预示着一种精神理论。我之前提到,我们所创造的深度学习系统都不能依靠自己独立生存。这些系统的自主性只有在它们包含了迄今为止一直被忽视的,类似于大脑其他部分的功能时才有可能实现,例如管理摄食、繁殖、调节激素、稳定内脏的下丘脑,以及帮助我们根据运动预测误差来调整运动的小脑。这些都是在所有脊椎动物中发现的古老结构,对生存起着至关重要的作用。
哈瓦·西格尔曼(Hava Siegelmann)是马萨诸塞大学阿默斯特分校的计算机科学家,她表明模拟计算(analog computing)是超图灵(super-Turing),也就是说,拥有能够超越数字计算机的计算能力。[25] 可以根据环境进行调整和学习的神经网络也有超图灵计算能力,而普通网络从训练集中学习,然后其结构就被固定下来,在操作时不再从它们的实际经验中学习,这一点和图灵机是一样的。但是,我们的大脑必须持续适应不断变化的条件,这就使我们具备了超图灵能力。我们如何做到这一点并同时拥有以前的知识和技能,是一个尚未解决的问题。
哈瓦是DARPA终身学习项目的项目经理。她的“终身学习计划”正在资助一些高级研究,这些研究旨在为自治系统中的终身学习创建新的集成架构。
- 虽然我们这位传奇发明家命运未卜,但是一旦国王意识到自己被欺耍了,他很可能会因为自己的无礼而被发落。
- Tesauro and Sejnowski, “A Parallel Network That Learns to Play Backgammon.”
- R. Sutton, “Learning to Predict by the Methods of Temporal Differences,” Machine Learning 3, no. 1 (1988): 9–44.
- 参阅Richard Bellman Adaptive Control Processes:AGuided Tour (Princeton:Princeton University Press. 1961), 51–59.
- G. Tesauro, “Temporal Difference Learning and TD-Gammon.” Communications of the ACM 38, no. 3 (1995): 58–68.
- J. Garcia, D. J. Kimeldorf, and R. A. Koelling, “Conditioned Aversion to Saccharin Resulting from Exposure to Gamma Radiation,” Science 122. no. 3160 (1955):157– 158.
- P. R. Montague, P. Dayan, and T. J. Sejnowski, “A Framework for Mesencephalic Dopamine Systems Based on Predictive Hebbian Learning,” Journal of Neuroscience 16, no. 5 (1996): 1936–1947.
- W. Schultz, P. Dayan, and P. R. Montague, “A Neural Substrate of Prediction and Reward,” Science 275, no. 5306 (1997): 1593–1599.
- P. N. Tobler, J. P. O’Doherty, R. J. Dolan, and W. Schultz, “Human Neural Learning Depends on Reward Prediction Errors in the Blocking Paradigm,” Journal of Neurophysiology 95, no. 1 (2006):301–310.
- M. Hammer and R. Menzel, “Learning and Memory in the Honeybee,” Journal of Neuroscience 15, no. 3 (1995): 1617–1630.
- L. A. Real, “Animal Choice Behavior and the Evolution of Cognitive Architecture,”Science 253, no. 5023 (1991): 980–986.
- P. R. Montague, P. Dayan, C. Person, and T. J. Sejnowski, “Bee Foraging in Uncertain Environments Using Predictive Hebbian Learning,” Nature 377, no.6551 (1995): 725–728.
- Y. Aso and G. M. Rubin, “Dopaminergic Neurons Write And Update Memories With Cell-Type-Specific Rules,” in L. Luo (ed.), eLife.5 (2016):e16135.doi: 10.7554/eLife.16135.
- W. Mischel and E. B. Ebbesen, “Attention in Delay of Gratification,” Journal of Personality and Social Psychology 16, no. 2 (1970): 329–337.
- V. Mnih, K. Kavukcuoglu, D. Silver, A. A. Rusu, J. Veness, M. G. Bellemare, et al.,“Human-Level Control through Deep Reinforcement Learning,” Nature 518, no.7540 (2015): 529–533.
- Simon Haykin, Cognitive Dynamic System: Perception-Action Cycle, Radar, and Radio (New York: Cambridge University Press, 2012).
- S. Haykin, J. M. Fuster, D. Findlay, and S. Feng, “Cognitive Risk Control for Physical Systems,” IEEE Access 5 (2017): 14664–14679.
- G. Reddy, A. Celani, T. J. Sejnowski, and M. Vergassola, “Learning to Soar in Turbulent Environments,” Proceedings of the National Academy of Sciences of the United States of America 113, no. 33 (2016): E4877–E4884.
- G. Reddy, J. W. Ng, A. Celani, T. J. Sejnowski, and M. Vergassola, “Soaring LikeaBird via Reinforcement Learning in the Field,” submitted for publication.
- Kenji Doya and Terrence J. Sejnowski, “A Novel Reinforcement Model of Birdsong Vocalization Learning,” in Gerald Tesauro, David S. Touretzky, and Todd K.Leen, eds., Advances in Neural Information Processing Systems 7 (Cambridge,MA: MIT Press, 1995), 101–108.
- A. J. Doupe and P. K. Kuhl, “Birdsong and Human Speech: Common Themes and Mechanisms,” Annual Review of Neuroscience 22 (1999): 567–631.
- G. Turrigiano, “Too Many Cooks? Intrinsic and Synaptic Homeostatic Mechanisms in Cortical Circuit Refinement,” Annual Review of Neuroscience 34 (2011):89–103.
- L. Wiskott and T. J. Sejnowski, “Constrained Optimization for Neural Map Formation: A Unifying Framework for Weight Growth and Normalization,” Neural Computation 10, no. 3 (1998):671–716.
- A. J. Bell, “Self-Organization in Real Neurons: Anti-Hebb in ‘Channel Space’?”Advances in Neural Information Processing Systems 4 (1991): 59–66: M. Siegel,E. Marder, and L. F. Abbott,“Activity-Dependent Current Distributions in Model Neurons,” Proceedings of the National Academy of Sciences of the United States of America 91, no. 24 (1994): 11308–11312.
- H. T. Siegelmann, “Computation Beyond the Turing Limit,” Science 238 (1995):632–637.
本书评论