On the Opportunities and Risks of Foundation Models
О возможностях и рисках фундаментальных моделей
2021-08-16
SCID: 54.1/utusbyn3
Discuss with AI
emergent capabilitiesfoundation modelsmodel homogenizationscale in deep learningsociotechnical impacts
Figures from the paper
Abstract (AI)
AI is undergoing a paradigm shift with the rise of models (e.g., BERT, DALL-E, GPT-3) that are trained on broad data at scale and are adaptable to a wide range of downstream tasks. We call these models foundation models to underscore their critically central yet incomplete character. This report provides a thorough account of the opportunities and risks of foundation models, ranging from their capabilities (e.g., language, vision, robotics, reasoning, human interaction) and technical principles(e.g., model architectures, training procedures, data, systems, security, evaluation, theory) to their applications (e.g., law, healthcare, education) and societal impact (e.g., inequity, misuse, economic and environmental impact, legal and ethical considerations). Though foundation models are based on standard deep learning and transfer learning, their scale results in new emergent capabilities,and their effectiveness across so many tasks incentivizes homogenization. Homogenization provides powerful leverage but demands caution, as the defects of the foundation model are inherited by all the adapted models downstream. Despite the impending widespread deployment of foundation models, we currently lack a clear understanding of how they work, when they fail, and what they are even capable of due to their emergent properties. To tackle these questions, we believe much of the critical research on foundation models will require deep interdisciplinary collaboration commensurate with their fundamentally sociotechnical nature.
Key Findings
1
Addressing opportunities and risks of foundation models requires deep interdisciplinary research because their impacts are fundamentally sociotechnical, spanning legal, ethical, economic, and environmental domains.
2
Current understanding of how foundation models work, when they fail, and their full capabilities is limited, necessitating further research.
3
Foundation models are large-scale, broadly trained models (e.g., BERT, DALL-E, GPT-3) that are adaptable to many downstream tasks across modalities such as language, vision, robotics, reasoning, and human interaction.
4
Homogenization around foundation models provides strong leverage for many applications but risks propagating the foundation model's defects to all adapted downstream models.
5
Scale yields new emergent capabilities not present in smaller models, increasing effectiveness across diverse tasks but also creating unknown behaviors and failure modes.
Research Object
Foundation models (large-scale pretrained models such as BERT, DALL-E, GPT-3)
Research Subject
Opportunities, capabilities, technical principles, risks, applications, societal impacts, emergent behaviors, and failure modes of foundation models when adapted to downstream tasks
Publication Details
Publication Date
2021-08-16
Journal
Publisher
ISSN
Cited by
2288
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest