On the Opportunities and Risks of Foundation Models

О возможностях и рисках фундаментальных моделей
Christopher D. Manning, Peter Henderson, Deepak Narayanan, Matei Zaharia, Li Fei-Fei, Ranjay Krishna, Michael S. Bernstein, William Yang Wang, Chelsea Finn, Dorsa Sadigh, Jeannette Bohg, Omar Khattab, Suvir Mirchandani, Michihiro Yasunaga, Lucia Zheng, Neel Guha, Daniel E. Ho, Mark Krass, Dan Jurafsky, Julian Nyarko, Erik Brynjolfsson, Zanele Munyikwa, Jure Leskovec, Thomas Icard, Rishi Bommasani, Tatsunori Hashimoto, Percy Liang, Suraj Nair, Eric Mitchell, Bo-Han Wu, Tianyi Zhang, Kyle Hsu, Ali Ahmad Malik, Jia-Jun Wu, Russ B. Altman, Krishnan Srinivasan, Christopher Potts, Armin W. Thomas, Stefano Ermon, Pang Wei Koh, Siddharth Karamcheti, Noah D. Goodman, Rohan Taori, Antoine Bosselut, Xikun Zhang, Emma Brunskill, John Etchemendy, Juan Carlos Niebles, Kathleen Creel, Geoff Keeling, Pratyusha Kalluri, Dallas Card, Ehsan Adeli, Esin Durmus, Faisal Ladhak, Yuhuai Wu, Sang Michael Xie, Joon-Sung Park, Tong Lee, John Hewitt, Florian Tramèr, Jiaxuan You, Yuhui Zhang, Tengyu Ma, Chris Piech, Yusuf Roohani, Keshav Santhanam, Ananya Kumar, Kaitlyn Zhou, Chris Donahue, Karan Goel, Christopher Ré, Rodrigo Castellon, Shyamal Buch, Benjamin T. Newman, Xuechen Li, Hongyu Ren, Xiang Lisa Li, Michael Zhang, Annie Chen, Aditi Raghunathan, Saahil Jain, Drew A. Hudson, Simran Arora, Sydney von Arx, Niladri S. Chatterji, Jared Quincy Davis, Demszky, Dora, Moussa Doumbouya, Kawin Ethayarajh, Trevor Gale, Lauren Gillespie, Shelby Grossman, Jenny Hong, Jing Huang, Fereshte Khani, Rohith Kuditipudi, Mina Lee, Isabelle Levent, Avanika Narayan, Allen Nie, Hamed Nilforoshan, Giray Ogut, Laurel Orr, Isabel Papadimitriou, Eva Portelance, Rob Reich, Frieda Rong, Camilo Ruiz, Jack Ryan, Shiori Sagawa, Andy Shih, Alex Tamkin, Rose E. Wang
2021-08-16

emergent capabilitiesfoundation modelsmodel homogenizationscale in deep learningsociotechnical impacts
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.
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.

Foundation models (large-scale pretrained models such as BERT, DALL-E, GPT-3)

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
Access Type
Author Information
Authors
Christopher D. Manning
Peter Henderson
Deepak Narayanan
Matei Zaharia
Li Fei-Fei
Ranjay Krishna
Michael S. Bernstein
William Yang Wang
Chelsea Finn
Dorsa Sadigh
Jeannette Bohg
Omar Khattab
Suvir Mirchandani
Michihiro Yasunaga
Lucia Zheng
Neel Guha
Daniel E. Ho
Mark Krass
Dan Jurafsky
Julian Nyarko
Erik Brynjolfsson
Zanele Munyikwa
Jure Leskovec
Thomas Icard
Rishi Bommasani
Tatsunori Hashimoto
Percy Liang
Suraj Nair
Eric Mitchell
Bo-Han Wu
Tianyi Zhang
Kyle Hsu
Ali Ahmad Malik
Jia-Jun Wu
Russ B. Altman
Krishnan Srinivasan
Christopher Potts
Armin W. Thomas
Stefano Ermon
Pang Wei Koh
Siddharth Karamcheti
Noah D. Goodman
Rohan Taori
Antoine Bosselut
Xikun Zhang
Emma Brunskill
John Etchemendy
Juan Carlos Niebles
Kathleen Creel
Geoff Keeling
Pratyusha Kalluri
Dallas Card
Ehsan Adeli
Esin Durmus
Faisal Ladhak
Yuhuai Wu
Sang Michael Xie
Joon-Sung Park
Tong Lee
John Hewitt
Florian Tramèr
Jiaxuan You
Yuhui Zhang
Tengyu Ma
Chris Piech
Yusuf Roohani
Keshav Santhanam
Ananya Kumar
Kaitlyn Zhou
Chris Donahue
Karan Goel
Christopher Ré
Rodrigo Castellon
Shyamal Buch
Benjamin T. Newman
Xuechen Li
Hongyu Ren
Xiang Lisa Li
Michael Zhang
Annie Chen
Aditi Raghunathan
Saahil Jain
Drew A. Hudson
Simran Arora
Sydney von Arx
Niladri S. Chatterji
Jared Quincy Davis
Demszky, Dora
Moussa Doumbouya
Kawin Ethayarajh
Trevor Gale
Lauren Gillespie
Shelby Grossman
Jenny Hong
Jing Huang
Fereshte Khani
Rohith Kuditipudi
Mina Lee
Isabelle Levent
Avanika Narayan
Allen Nie
Hamed Nilforoshan
Giray Ogut
Laurel Orr
Isabel Papadimitriou
Eva Portelance
Rob Reich
Frieda Rong
Camilo Ruiz
Jack Ryan
Shiori Sagawa
Andy Shih
Alex Tamkin
Rose E. Wang
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat
Make a presentation
100%