The Importance of Effective Altruism in AI Security
A couple of days ago, a US AI policy expert told me the following:
“At this point, I regret to say that if you’re not looking for the EA [effective altruism] influence, you are missing the story.”
Well, I regret to say that, at least in part, I missed the story last week.
My article on why top AI labs and think tanks are concerned about securing LLM model weights seemed timely and straightforward. However, I failed to highlight the important context surrounding the effective altruism (EA) community’s connections within the AI security field.
The thread of connections between EA and AI security policy circles is finely woven, and it is ironic that I, like other reporters, overlooked it. As a journalist covering the AI landscape, I have been trying to understand how effective altruism evolved from an intellectual project to a highly influential group. Their primary concern revolves around preventing a future AI catastrophe from destroying humanity.
However, critics argue that this focus on existential risks detracts from addressing current AI risks, such as bias, misinformation, high-risk applications, and traditional cybersecurity.
Despite my knowledge of Anthropic’s connections to the EA movement, I failed to explore the EA rabbit hole in my previous piece. An article published by Politico the day after mine unveiled an important missing link. RAND Corporation researchers played a crucial role in shaping the White House’s requirements in the Executive Order on AI. RAND also received significant funding from Open Philanthropy, an EA group financed by Facebook co-founder Dustin Moskovits.
This article raised awareness of the prominent role effective altruists, including RAND CEO Jason Matheny and senior information scientist Jeff Alstott, played in policy circles. Their ties to the Biden administration and their contributions to RAND’s research and analysis further solidify the EA’s influence within AI security and policy.
The Views of AI Security Experts
I interviewed several key players in the AI security community who shed light on the significance of the EA’s involvement. Sella Nevo and Dan Lahav, co-authors of a report from RAND Corporation called “Securing Artificial Intelligence Model Weights,” emphasized the evolving nature of AI security. They highlighted that AI models could become of significant national security importance within the next two years. The authors warned against the potential misuse of AI models by malicious actors for developing biological weapons.
Jason Clinton, chief information security officer at Anthropic, expressed his primary concern about securing the model weights for Anthropic’s LLM. He highlighted the alarming threat posed by opportunistic criminals, terrorist groups, or nation-state operations accessing the weights of sophisticated and powerful LLMs. According to Clinton, compromising the entire file would mean compromising the entire neural network, which is a cause for concern in the AI security community.
These viewpoints underline the need for strong cybersecurity measures to protect AI model weights and prevent unauthorized access by those with malicious intent. Understanding the connections between effective altruism and AI security is crucial in shaping policies and regulations that will impact the development and implementation of AI technologies in the coming decades.
While the ideological agendas within the AI policy sphere may be underestimated, it is essential to acknowledge how beliefs and affiliations influence decision-making. The transparency and accountability of Big Tech companies and policy leaders are of utmost importance in ensuring the responsible and ethical development of AI.
In conclusion, effective altruism’s influence within the AI security landscape cannot be ignored. The community’s dedication to addressing AI risks and preventing future catastrophes is commendable. However, it is crucial to strike a balance between addressing both existential and immediate AI risks while fostering transparency and accountability in AI development and policy-making.
