Francis Bacon popularized the idea that science is based on a process of induction by which repeated observations are, in some unspecified way, generalized to theories based on the assumption that the future resembles the past. This idea was criticized by Hume and others as untenable leading to the famous problem of induction. It wasn't until the work of Karl Popper that this problem was solved, by demonstrating that induction is not the basis for science and that the development of scientific knowledge is instead based on the same principles as biological evolution. Today, machine learning is also taught as being rooted in induction from big data. Solomonoff induction implemented in an idealized Bayesian agent (Hutter's AIXI) is widely discussed and touted as a framework for understanding AI algorithms, even though real-world attempts to implement something like AIXI immediately encounter fatal problems. In this paper, we contrast frameworks based on induction with Donald T. Campbell's universal Darwinism. We show that most AI algorithms in use today can be understood as using an evolutionary trial and error process searching over a solution space. In this work we argue that a universal Darwinian framework provides a better foundation for understanding AI systems. Moreover, at a more meta level the process of development of all AI algorithms can be understood under the framework of universal Darwinism.
Universal Induction with Varying Sets of Combinators
AIDEUS;National Research University of Information Technology, Mechanics and Optics(国立信息技术、机械与光学研究大学);Aix Marseille Université(艾克斯-马赛大学);CNRS(法国国家科学研究中心)
Optimising the attribute order in Fuzzy Rough Rule Induction
Department of Mathematics, Computer Science and Statistics, Ghent University(数学、计算机科学与统计学系,根特大学);Faculty of Economics, University of Catania(经济学院,卡塔尼娅大学);Institute of Computing Science, Poznań University of Technology(计算科学研究所,波兹南技术大学);Systems Research Institute, Polish Academy of Sciences(系统研究所,波兰科学院)
Assessing Adaptive World Models in Machines with Novel Games
MIT(麻省理工学院);Harvard University(哈佛大学);University of Cambridge(剑桥大学);Basis Research Institute(基础研究机构);Princeton University(普林斯顿大学);ARC Prize Foundation(ARC奖基金会);University of British Columbia(不列颠哥伦比亚大学)
本文提出用具有持续新颖性的新颖游戏构建基准,以评估 AI 智能体通过交互快速归纳世界模型并实现类人适应与泛化的能力。
Adversarial Examples and the Deeper Riddle of Induction: The Need for a Theory of Artifacts in Deep Learning
The University of Houston(休斯顿大学)
深度学习是目前人工智能中应用最广泛、最成功的技术。它有望推动科学发现的前沿超越目前的限制。然而,怀疑论者担心深层神经网络是黑匣子,并质疑如果人类无法理解它们,这些进步是否真的可以被视为科学进步。与此相关的是,这些系统还具有令人困惑的新漏洞:最明显的是对“对抗性示例”的敏感性。在本文中,我认为,对抗性的例子将成为哲学和各种科学辩论的热点。具体而言,关于对抗性示例的新发现挑战了共识观点,即网络对这些情况的判决是由训练集中的过度拟合特质噪声引起的,并且可能是检测人类无法感知的预测有用的“数据几何结构的内在特征”的结果(Ilyas et al.,2019年)。这些结果应该促使我们重新审视对哲学和科学交叉点上最深刻的谜题之一的回应:纳尔逊·古德曼的归纳法“新之谜”。具体来说,它们提出了一种可能性,即许多科学的进步将取决于对人类难以理解的有用特征的检测和操纵。然而,在我们评估这种可能性之前,我们必须确定哪些(如果有的话)这些难以理解的特征是真实的,但只适用于“外星人”的感知和认知,哪些是深度学习的独特伪像--因为像镜头耀斑或吉布斯现象这样的伪像对预测同样有用,但通常被视为科学理论化的障碍。因此,机器学习研究人员迫切需要为深度神经网络开发一种伪影理论,最后,我为这一研究领域勾画了一些初步方向。