Banning self-recursive improvement in AI models?
Some people are suggesting this, including Ezra Klein. I do not understand how it is supposed to work. Put aside the issue of foregone innovations, let us say I seek to access an AI from overseas. Am I allowed to visit their website? So many sites have AI behind them, including Chinese open source AI. […] The post Banning self-recursive improvement in AI models? appeared first on Marginal…
Various individuals, such as Ezra Klein, have proposed the idea of banning self-recursive improvement in AI models. However, the mechanics behind such a ban appear unclear. One could argue that accessing an AI from a foreign website is no different from visiting any other site, regardless of its AI capabilities. If Chinese open-source AI sites are accessible, why would American users be denied access to others?
Are all these sites to be banned, and if so, how can this be enforced? Is the government required to review the model's creation logs, a process that seems impractical? Moreover, what if a company like Anthropic licenses its intellectual property to a subsidiary in the Cayman Islands, allowing it to continue improving its AI? This raises questions about the future of multinational corporations operating across borders.
If AI models can only be developed in certain nations, it may favor foreign entities and potentially hinder American competitiveness. The concept of preserving "AI sovereignty" for Americans appears problematic when considering the practical implications of such a ban. Could this mean that regular companies can utilize AI for code generation, but the leading labs cannot?
If so, how should we resolve this equilibrium? Alternatively, the provision of code-writing functions might become unavailable to everyone. While some may view this idea as reasonable and less terrifying, it is crucial to consider the practical challenges involved in implementing such a ban. Even if one believes in engaging with China, it is important to acknowledge that competitors can advance more swiftly than anticipated.
Additionally, the possibility of securing compute resources from entities like Elon Musk further complicates the situation. Despite the challenges, it is essential to recognize that numerous decisions have been made in the past, which suggest that maintaining a balanced approach to AI development is the most viable path forward. Procrastinating on this issue may prove detrimental to all parties involved.
Written by urgent.news from Marginal Revolution's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.