Media and Infrastructures of ArtificialIntelligence: Computer Vision, Transfer Learning, and Artificial Neural Networks asBlack Box

当前人工智能（AI）技术以人工神经网络（ANN）为主导。该项目的中心研究兴趣是探索媒体在其特定的基础设施组织中参与这些人工神经网络技术的生产和形成的方式。两个角度是中央这个研究重点：第一，该项目的目的是调查的话语和ANN专家的做法，他们对媒体的理解：什么明确和隐含的媒体概念（例如，在功能方面作为媒体的工作，组织和/或通信）的特点，这个特定的话语？其次，目的是研究人工神经网络的工业科学实践作为媒体的做法：媒体和基础设施如何表达自己在人工神经网络的研究/开发的具体实践，尤其是在比较和结合巩固，官方知识的话语？我的具体重点是计算机视觉领域和迁移学习的机器学习方法，包括它们的复杂纠缠。特别是在计算机视觉领域，如果人工智能系统不必为每个新任务进行训练，而是能够利用以前的学习经验（例如，识别相似但不同的对象），则会更有效。作为对计算机视觉和迁移学习观点的补充，该项目使用人工神经网络作为黑盒的共同特征作为出发点，探索如何在人工神经网络研究的话语和实践中协商假设的不透明性问题。因此，该项目还对旨在消除或减少这些不透明问题的战略感兴趣，特别是在媒体及其基础设施组织的作用方面。该项目在一个概念框架内解决了这些研究问题，该框架结合了话语分析，媒体民族志和演员网络理论（包括所谓的演员媒体理论）。最重要的是，该项目旨在阐明不同媒体及其基础设施组织的知识技术功能，作为人工神经网络技术的生产和形成的配置，特别是在其人类学方面。

Current technologies of Artificial Intelligence (AI) are dominated by Artificial Neural Networks (ANN). The project's central research interest is to explore the ways in which media in their specific infrastructural organization are involved in the production and formation of these ANN technologies. Two perspectives are central to this research focus: First, the project aims to investigate the discourses and practices of ANN experts with regard to their understanding of media: What explicit and implicit concepts of media (e.g. in functional terms as media of work, organization, and/or communication) characterize this particular discourse? Second, the aim is to examine the industrial-scientific practice of ANN as media practice: How do media and infrastructures articulate themselves in the concrete practice of ANN research/development, not least in comparison to and in conjunction with the consolidated, official knowledge of the discourse?My specific foci are the field of computer vision and the machine learning approach of transfer learning, including their complex entanglement. Especially in the field of computer vision, it is more effective if an AI system does not have to be trained for every new task, but is able to make use of previous learning experiences (for example, recognition of similar but different objects). Complementing the perspectives on computer vision and transfer learning, the project uses the common characterization of ANN as black box as the starting-point to explore how problems of supposed opacity are negotiated in the discourse and practices of ANN research. Accordingly, the project is also interested in strategies devised to eliminate or reduce these problems of opacity, especially with regard to the role of media and their infrastructural organization. The project addresses these research questions within a conceptual framework that combines discourse analysis, media ethnography, and Actor-Network Theory (including so-called Actor-Media Theory). Above all, the project seeks to shed light on the epistemic-technical functions of different media and their infrastructural organization as configurations for the production and formation of ANN technologies, especially in terms of their praxeological dimension.