An IEEE Spectrum essay argues that AI for Good campaigns can miss underlying social problems and involve questionable data mining.
Publication
Licensing is neither feasible nor effective for addressing AI risks
This essay argues that licensing is neither feasible nor effective for addressing AI risks and says non proliferation benefits incumbents.
About Licensing is neither feasible nor effective for addressing AI risks
This essay by Sayash Kapoor and Arvind Narayanan, dated June 10, 2023, argues that licensing or non-proliferation of advanced AI models is neither feasible nor effective as a way to reduce AI risks. It examines enforcement, noting that licensing requirements can be ignored and that monitoring data centers would become harder as training costs fall. It also argues that non-proliferation would concentrate AI development among a small number of companies and could increase certain risks, such as shared security vulnerabilities and homogenized outcomes in areas like hiring.
The essay supports its points with specific examples, including a reported 44-fold drop in the cost of training an image recognition classifier between 2012 and 2019, and the authors' own estimate that the open-source Falcon model could be trained for less than USD 1 million. The excerpt available for this entry ends partway through its list of five AI risks, so the remaining arguments are not described here.
Key features
- Argues that licensing AI model training is hard to enforce
- Cites falling training costs for computer vision and language models
- Examines how non-proliferation could concentrate AI development among few companies
- Discusses AI risks including monoculture and outcome homogenization
- Responds to the Future of Life Institute's proposed six-month training pause
Best for: Readers interested in AI policy, regulation, and the economics of AI model development.
Good for
- Supports its arguments with specific cost and training examples
- Addresses the proposal's enforcement and market concentration effects in separate sections
Watch out for
- Presents the authors' own cost estimates, which are not independently verified in the text
- The available excerpt ends before the full list of risks is covered
Licensing is neither feasible nor effective for addressing AI risks at a glance
- Official website
- normaltech.ai/p/licensing-is-neither-feasible-…
- Category
- News & Media › Tech News
- Type
- Publication
- Platforms
- Web
- Language
- English
- Last checked
- Oct 6, 2026 · Online when checked
- Listed
- Oct 5, 2026
Similar publications
Other listings in Tech News with the most in common.
This essay considers modern language and diffusion models in the context of earlier AI research and hype cycles.
Publication
Mir Fantastiki covers science fiction and fantasy books, films, series, games, comics, and anime.
Publication
Passion Times publishes Hong Kong news and information across society, culture, arts, and radio programs.
Publication
This Microsoft article discusses adjusting expectations about AI and change.
Publication
TestingCatalog covers AI news, model releases, tools, leaks, and rumors from the artificial intelligence industry.
Publication
Popular in Tech News
Well-known listings from the same topic.
The File Drawerchris-said.ioSecListsseclists.orgMLOps Nowmlopsnow.comDataconomydataconomy.comAlien Dreams: An Emerging Art Scene - ML@B Blogmlberkeley.substack.comPipeline Conversationpodcast.zenml.io
Licensing is neither feasible nor effective for addressing AI risks FAQ
Who wrote this essay and when was it published?
Sayash Kapoor and Arvind Narayanan wrote it. The essay is dated June 10, 2023.
What position does the essay take on AI licensing?
It argues that licensing is neither feasible nor effective for addressing AI risks, and that non-proliferation would mainly benefit incumbent companies.
What cost estimate does the essay give for training an open-source model?
The authors estimate that Falcon, which required fewer than 400 GPUs and two months to train, can be trained for less than USD 1 million.
Is something wrong or out of date?
Tell us and an editor will check it.
Own this website?
Claim this listing
Claim the listing and suggest updates yourself. An editor checks every claim first.