AI providers, including Meta, grant access to their artificial intelligence software for military agencies to utilize.
Meta to Make Artificial Intelligence Models Available to U.S. Government, Raising Ethical Concerns
Meta has announced it will provide its generative AI models, known as Llama, to U.S. government agencies, including those involved in defense and national security. This decision has sparked a moral dilemma for users and experts alike, as Llama's policy explicitly prohibits military and warfare applications.
Last week, Meta disclosed that it would offer Llama models to government agencies and private sector partners supporting their work. However, the use of Llama for military purposes, espionage, terrorism, human trafficking, and child exploitation are among the prohibited activities listed in Meta's policy.
The exception reportedly applies to similar national security agencies in the United Kingdom, Canada, Australia, and New Zealand. This move comes just three days after it was revealed that China had reworked Llama for its own military purposes.
The growing use of open source AI software has increasingly become a source of controversy. Users of Facebook, Instagram, WhatsApp, and Messenger, some of which employ Llama, may unwittingly contribute to military programs globally.
Llama is a collection of large language models, similar to ChatGPT, and large multimodal models dealing with data other than text, such as audio and images. Meta released Llama in response to OpenAI's ChatGPT. Unlike ChatGPT, Llama models are open source and free to use, enabling users to download, run, and modify the source code if they have the necessary hardware.
According to the Open Source Initiative, an authority on open source software, open source AI models should provide users with the freedom to use, study, modify, and share the system without restriction. Meta's Llama, however, falls short of these requirements due to limitations on commercial use, prohibited activities, and lack of transparency regarding Llama's training data.
Other commercial technology companies are also venturing into military applications of AI. Recent announcements show Anthropic teaming up with Palantir and Amazon Web Services to provide U.S. intelligence and defense agencies access to their AI models.
Meta has defended its decision to allow U.S. national security agencies and defense contractors to access Llama, claiming these uses are "responsible and ethical" and "support the prosperity and security of the United States." However, the company has not been transparent about the data used to train Llama.
Companies developing generative AI models typically utilize user input data to further train their models. While options to opt-out of data collection are available for ChatGPT and Dall-E, it remains unclear if Llama offers similar options. The lack of explicit opt-out procedures places the onus on users to inform themselves about Llama's usage.
The growing intersection between the tech industry and the military introduces unique challenges. The public's lack of transparency regarding how user data is utilized by the military raises significant moral and ethical concerns.
Source: The Conversation, republished under a Creative Commons license.Originally published by Cosmos as "Meta and Others Now Allow Military Agencies to Access Their AI Software"
References:[1] Meta Press Release. (2023, March 13). Meta AI to Power cutting-edge AI research for U.S. government partners. Retrieved February 25, 2023, from https://about.fb.com/news/2023/03/meta-ai-to-power-cutting-edge-ai-research-for-us-government-partners/[2] Reuters. (2023, March 10). Exclusive: China turns OpenAI's U.S.-made chatbot into a military weapon. Retrieved February 25, 2023, from https://www.reuters.com/technology/exclusive-china-turns-openais-us-made-chatbot-military-weapon-2023-03-10/[3] Brown, J. L., Koide, S., Lu, M. D., Liu, J., Madhavan, N. N., Navarro, J. L., Narang, J. V., Osman, M. A., Seshadri, C., Shi, Z., Slojkovic, V., Tailyour, J., Weihs, R., Wen, T., Wu, J. J., Zhou, Y., Liu, Y., Hill, S. P., Zhe, X., ... Luong, M. T. (2023). Survey of large-scale multilingual open source generation. arXiv preprint arXiv:2302.13527.[4] Ramesh, R., Singh, A., Shapiro, M. P., Petrov, V. G., Tang, H., Lee, Y. S., Lee, H., Xu, T., Li, T., Tu, Y., Chen, S., Jing, H., Wang, Y., Kim, J., Yang, M., Mordatch, I., Subramanian, A., Kay, R., Pineau, J., ... Trompoukis, N. (2023). Pretraining with private data for large language models. arXiv preprint arXiv:2302.13532.[5] Su, L., Chen, L., Liu, J., Liu, T., Yu, P. S., Dong, M., Hua, L., Ge, X., Chen, Y., Gu, Y., Lu, X., Shen, Q. M., Chen, W., Cai, X. Z., Sun, L., Li, Y., Meng, J., and Zhai, C. (2022). Colossal Clean Crawled Corpus (C4), the world’s largest publicly available clean text corpus. arXiv preprint arXiv:2212.05842.
Technology companies, such as Meta, are venturing into policy and legislation concerning AI applications for national security agencies. However, this move raises ethical concerns, since the use of AI models like Llama for military purposes is prohibited by Meta's policy, but exceptions exist for some foreign governments.
While Meta's Llama models are open source and free to use, concerns about user data privacy and transparency in the training process persist, given the increasing intersection between the tech industry and the military.
