Openness in the context of AI encompasses all principles or philosophies surrounding the ability and right to freely use, share, modify, and study the AI system or its components (weights, code, dataset, research paper, development framework, etc) without restriction of purpose and without discrimination against individuals or groups, similar to other philosophies of openness such as Open Source or Open Access.
This repo collects relevant papers or other content for research on openness in the field of AI and AI ethics. Given that the research on openness in AI is still relatively recent and lacks scientific literature, this repo also incorporates blog posts from reliable sources (such as university websites, recognized organizations in the field, etc).
If you want to contribute, feel free to send me a message or make a pull request!
- Definitions, degrees of openness and terminologies
- Costs, risks and benefits
- AI systems described as 'Open' or 'Open Source'
-
Deep Dive: AI
Open Source Initiative
since 2022 [Website] -
The Gradient of Generative AI Release: Methods and Considerations
I. Solaiman
Proceedings of the 2023 ACM conference on fairness, accountability, and transparency, 2023, [Paper] -
Opening up ChatGPT: Tracking openness, transparency, and accountability in instruction-tuned text generators
A. Liesenfeld, A. Lopez, and M. Dingemanse
Proceedings of the 5th International Conference on Conversational User Interfaces, Jul. 2023, [Paper] [Live Table] -
AI weights are not open ‘source.’
Sid Sijbrandij
Jun 27, 2023, [Blog Post] -
The Time Is Now to Develop Community Norms for the Release of Foundation Models.
Percy Liang, Rishi Bommasani, Kathleen Creel, and Rob Reich
May 17, 2022, [Blog Post] -
Beyond “Release” vs. “Not Release”
Girish Sastry
2021, [Blog Post]
-
Hazards from Increasingly Accessible Fine-Tuning of Downloadable Foundation Models
A. Chan, B. Bucknall, H. Bradley, and D. Krueger
arXiv, Dec. 22, 2023, [Paper] -
Open-Sourcing Highly Capable Foundation Models: An evaluation of risks, benefits, and alternative methods for pursuing open-source objectives.
E. Seger et al.
arXiv, Sep. 29, 2023, [Paper] -
Limits and Possibilities for ‘Ethical AI’ in Open Source: A Study of Deepfakes
D. G. Widder, D. Nafus, L. Dabbish, and J. Herbsleb
Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency, Jun. 2022, [Paper] -
The tension between openness and prudence in AI research
J. Whittlestone and A. Ovadya
arXiv, Jan. 13, 2020, [Paper] -
Open (For Business): Big Tech, Concentrated Power, and the Political Economy of Open AI
D. G. Widder, S. West, and M. Whittaker
Aug. 17, 2023, [Paper] -
Strategic Implications of Openness in AI Development
N. Bostrom
Global Policy, 2017, [Paper] -
The Offense-Defense Balance of Scientific Knowledge: Does Publishing AI Research Reduce Misuse?
T. Shevlane and A. Dafoe
Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, in AIES ’20, Feb. 2020, [Paper] -
How Open Source Machine Learning Software Shapes AI
M. Langenkamp
Thesis, Massachusetts Institute of Technology, Nov. 01, 2023, [Paper] -
Democratising AI: Multiple Meanings, Goals, and Methods
E. Seger, A. Ovadya, D. Siddarth, B. Garfinkel, and A. Dafoe
Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society, in AIES ’23, Aug. 2023, [Paper] -
Building open-source AI
Y. R. Shrestha, G. von Krogh, and S. Feuerriegel
Nature Computational Science, 2023, [Paper] -
On the Societal Impact of Open Foundation Models
S. Kapoor et al.
2024, [Paper] [Website]
- Llama 2 (Meta) [Paper] [Model Card] [Download weights] [Inference Code]
- Meta’s LLaMa 2 license is not Open Source, Open Source Initiative, July 20, 2023, [Blog Post]
- BLOOM and BLOOMZ (BigScience) [Paper] [Model Card] [Github]
- Alpaca (Stanford) [Website] [Model Card] [Github]
- Mistral [Website] [Dowload weights]