生成式AI与未来工作：增强还是自动化？

Generative AIand the Future of Work Augmentation or Automation? ABOUT THE AUTHORS John Zysman\\The Berkeley Roundtable on the International Economy (BRIE)\\ Mark Nitzberg\\UC Berkeley Center for Human-Compatible AI ABOUT THE WEIZENBAUM INSTITUTE The Weizenbaum Institute is a joint project funded by the Federal Ministry of Educationand Research (BMBF) and the State of Berlin. It conducts interdisciplinary and basicresearch on the digital transformation of society and provides evidence- and value-based options for action in order to shape digitization in a sustainable, self-determinedand responsible manner. Weizenbaum Discussion Paper Generative AI and the Future of Work Augmentation or Automation? John Zysman, Mark Nitzberg \\Abstract1 This report examines the potential impact of Generative artificial intelligence (AI) systems, suchas ChatGPT, on the future of work and, by implication, on productivity. It argues that althoughGenerative AI is powerful, it has significant limitations and risks that require humans to remain“in the loop” not only to prevent systems from going off the rails, but to capture value. Ratherthan taking a deterministic view that artificial intelligence (AI) will inevitably destroy jobs, thearticle suggests that an analysis should start with how firms can strategically deploy these toolsto gain an advantage. It asks whether “augmentation” or “simplistic automation” lies ahead. Ourobjective is to move beyond hype and despair.2 The existing digital infrastructure has enabled AI to be adopted quickly. However, projectionsbased solely on automating existing tasks fail to capture the complex reorganizations that arelikely to happen. Firms in sectors such as professional services, materials, and pharmaceuticalsseem to have particular exposure to the use of Generative AI tools. Adaptations will vary acrosscontexts and depend greatly on who controls the decisions about deployment. Maintaining thecentrality of humans is likely to prove crucial—in training systems, curating data, and assessingoutputs. One question is which business strategies and public policies encourage that engage-ment and make it possible. Although AI regulation debates matter, promoting social prosperity depends heavily on direct-ly shaping the trajectory of the development and use of AI. This requires influencing the con-straints and the incentives that firms face, as well as the strategic mindsets of decision makers.Which groups are engaged in the discussions and debates is of vital importance. The article rec-ommends that, beyond the traditional policy proposals, an independent public-interest con-sultancy needs to be established in order to design creative business strategies that augmentworkers in a manner that will support, rather than hinder, social prosperity. Ultimately, avoid-ing a dystopian scenario might hinge on fostering new norms in which human capabilities re-main essential. \\Table of Contents 1Introduction: Automation or Augmentation?52A Baseline and Beyond83From AI Winter through Machine Learning13to Large Langauge Models4Assessing the Impact of Emerging Generative AI on Work225Is an AI Economy for Social Prosperity Possible?31Imprint36 1Introduction: Automation or Augmentation? This essay considers the impact of Generative artificial intelligence (hereafter, GenAI) on work,the organization of work, and, by implication, on productivity.3Its goal is to frame the debate,so as to go beyond the hype and despair.4Hype and despair have accompanied every digital era,from the transistor and microprocessors, to cloud computing and platforms, and now modernAI and GenAI. In writing it, we seek to present a balanced understanding of how the evolution ofthis latest technology revolution could be steered. GenAI is “a type of artificial intelligence technology that utilizes deep learning models to createvarious forms of content, such as text, images, and code, based on the data they were trainedon.”5This represents a significant leap in computing capability. The seeming speed of deploy-ment and adoption in itself generates a sense of urgency that society and the economy havereached a dramatic turning point. GenAI emerged abruptly in both the public eye and commercial environments not only be-cause of significant innovation, but because the essential infrastructure and technology forits deployment were already in place. Data, computing power, and networks were all at handwhen machine learning and transformer-based systems were developed and deployed. The ex-traordinary pace of experimentation with these new tools, if not yet full or diverse adoption,was possible because the complementary digital tools and infrastructure needed for the de-ployment of GenAI were at the ready. The several prior decades, perhaps best characterized asthe emergence of the “Platform Economy”, saw extraordinary expansion in all of the digital ca-pacities required for Generative AI to emerge: storage, computation, and informati