Over the last year, those of us who follow China's AI governance have been carefully watching whether China would establish an AI Safety Institute (AISI) to match those in the UK, US, and globally. That institution has now emerged, and it tells us a lot about the state of debate
The China AI Safety and Development Association (CnAISDA) is China's self-described AISI counterpart. It exists primarily because China wants a seat at the table in international AI governance conversations. But the experts who are driving it––and who are really influential in
CnAISDA looks different from AISIs in the UK and US. Rather than building a new institution inside gov't, CnAISDA brings together a bunch of existing institutions with AI safety and governance expertise – Tsinghua is the most important of the bunch. But its structure looks pretty
Testing and evaluating risks for frontier models is the most important function of the UK and US AISIs. Some of CnAISDA's constituent institutions do work there (especially Shanghai AI Lab and CAICT), but it's nascent. CAC evals are still mostly about immediate domestic stability
Still, getting to this point has been a remarkable story. On a structural level, building an AI safety-focused org was a challenge amidst geopolitical competition and a Chinese ecosystem clearly focused on AI for development. But this is really a story about people. It's pretty
The next critical juncture is next month's World AI Conference in Shanghai. How seriously top Chinese leaders engage with frontier AI safety, and whether concrete commitments follow, will shine light on the level of CnAISDA’s domestic influence. https://carnegieendowment.org/...
@Scott_R_Singer Good news, hopefully international AI legislation efforts will follow. Maybe this time China would initiate them, rather than be "nudged" into them like with previous global threats?🙏
@Scott_R_Singer This is in-line with the insights in "The Most Dangerous Fiction: The Rhetoric and Reality of the AI Race" (https://papers.ssrn.com/sol3/p... International accords are hopeful if US steers away from a cold-war-style AGI race, learning from the past (Missile Gap, Atomic Bomb).
@Scott_R_Singer Didn’t the US one get Trumped, and the UK one renamed?
@Scott_R_Singer They have only one mission- AI must obey CCP . Else doesn't matter
@Scott_R_Singer @threadreaderapp unroll if you don't mind
@Scott_R_Singer And if it would, …would it be comparable in terms of internal policies to those in the west?
@Scott_R_Singer For UK, rebranded as AI security institute https://www.aisi.gov.uk