The USG must therefore be ambitious with *both* frontier and narrow systems. The incoming Trump administration has the opportunity—and obligation—to do so. 9/9
Full article here foreignpolicy.com/2024/12/04/t...
The debate about the trajectory of frontier AI is remarkably divisive—ultimately no one knows if, when, or to what extent scaling frontier models will prove transformative in defense. 8/
With chip export controls and a pragmatic tech culture, China will likely emphasize narrow AI. If we over-fixate on a resource-intensive race to superintelligence that may not pan out, China may meanwhile build an arsenal of small, practical, and decisive AI systems. 7/
Focusing on frontier AI above all else could sideline narrow AI that more effectively addresses pressing national security needs—think hypersonic modeling, autonomous drone swarming, and the material science research critical to developing new weapons 6/
But if you're skeptical about the potential of the current frontier AI paradigm (as many investors and engineers increasingly are), such a strategy is wasteful, misguided, even dangerous—risking the loss of critical other AI leads over Beijing. 5/
*IF* you believe we’re on the cusp of superintelligence, this strategy makes sense: manufacturing a bunch of nobel-prize-level minds and beyond would eclipse other AI priorities. 4/
One official admitted that the doc was “rooted” in faith in frontier progress projections. Lots on accelerating gargantuan frontier training runs, little on boosting other important forms of AI. 3/
The National Security Memorandum on AI , released just before the election, was the clearest indication that the USG approach was becoming dangerously unbalanced in favor of frontier AI. 2/
For AI in national security, 🇺🇸's growing fixation on frontier models was becoming dangerously unbalanced. Trump must reprioritize "narrow" AI defense systems, or risk losing out to 🇨🇳. My latest with @rubyscanlon.bsky.social in @foreignpolicy.com 1/9
foreignpolicy.com/2024/12/04/t...