[HN Gopher] The AI Arms Race Isn't Inevitable
___________________________________________________________________
The AI Arms Race Isn't Inevitable
Author : billybuckwheat
Score : 14 points
Date : 2024-08-23 20:04 UTC (2 hours ago)
(HTM) web link (www.palladiummag.com)
(TXT) w3m dump (www.palladiummag.com)
| talldayo wrote:
| > Under this narrative, a nation possessing advanced AI
| capabilities could, theoretically, gain an insurmountable
| military and economic advantage.
|
| Sounds familiar. At the beginning of World War II, Nazi Germany
| thought their adoption of encryption was a game-changing
| application of industry that would provide an insurmountable
| military and information advantage. What they didn't know was
| that Poland had already cracked their encryption before the war
| had even begun. Everyone knows how the rest goes; Nazi Germany
| goes on emboldened by a broken protocol, and gets their
| communique leaked at every corner.
|
| It's not hard to see the same veins in think-pieces that advocate
| for "infrastructure investment" (eg. buy schools NVIDIA) and
| "public-private partnerships" to incorporate AI everywhere. We
| see this technology that _seems_ smart but is unproven in action;
| and our reaction is to entirely rely on it without proving the
| benefits. Advanced AI capability is just as they say - a
| narrative, a theory.
|
| > The result resembles a 21st-century Cold War, where escalating
| tension and mistrust make cooperation increasingly difficult and
| conflict more likely.
|
| The US has been in a Cold War with China for almost a decade now;
| our competition over AI is almost as irrelevant as the race to
| find Dodo Bird fossils. Free trade fucked the US over a long time
| ago, anyone that's paying attention should have expected this to
| boil over eventually. AI isn't even a straw on the proverbial
| camel's back, it's the Nvidia hardware that's a point of
| contention.
|
| > A more balanced approach is needed--one that acknowledges that
| legitimate security concerns can be better addressed through
| international cooperation rather than competition, as they have
| been in the past.
|
| What's really stretching my imagination, here, is that the threat
| of nukes is self-evident. The threat of advanced AI is not only
| theoretically irrelevent with current techniques, but
| demonstrably harmless. We've had advanced LLMs for _years_ now
| and nobody has summoned Yaldabaoth with ChatGPT 's secret
| apocrypha. We're going to be fine.
| Apocryphon wrote:
| As a corollary, the current chip manufacturing war seems a bit
| overblown as well. Faster chips for _what_ exactly? Do the
| current nanometer generations really make all that much a
| difference when it comes to drones or missiles?
| Terr_ wrote:
| I suspect the military need isn't so much "our toys must
| always be 10% better than anyone else's", but rather a
| question of logistics, ensuring that Y years from now you
| will be able to safely source an adequate amount of on-par
| components.
|
| The former is more-immediate and dramatic, so it gets trotted
| out when asking for funding.
| talldayo wrote:
| > Do the current nanometer generations really make all that
| much a difference when it comes to drones or missiles?
|
| For current-generation weapons; no, not really. I suspect the
| US wants to stop the export of Nvidia to prevent Russia and
| China from getting an all-inclusive SDK for computer vision
| and ML. That makes sense; Nvidia is partnered with Northrop
| Grumman and the rest of "the complex". Giving competitors
| parity with your flagship weaponry is a bad foot to start off
| future battles on.
|
| My best guess (just flying by the seat of my pants) is that
| the US wants to push their advantageous relationship with key
| NATO partners like ASML to drive a hard tech advantage in the
| future. Chips like STM-32s and 8088 clones are already mass-
| produced without license in China and sold to manufacturers
| wholesale. America can beat them (on paper) by leveraging the
| IP they have domestically to create superior alternatives
| locally, albeit at-cost. In high-margin markets like weaponry
| I think this will work; but for consumer products it's going
| to be a doozy to stop relying on unlicensed Chinese
| components and cheap labor.
|
| History shows this strategy has worked before. Without
| importing Russian engines, the PRC aerospace industry was
| utterly crippled for a good few decades. If the US can strand
| China on low-yield EUV nodes, there's potential to leapfrog
| Chinese compute capability by an order of magnitude. Both
| China and the US have committed to more than they're capable
| of with their domestic fab strategy, time will tell if either
| of them get anything from it.
___________________________________________________________________
(page generated 2024-08-23 23:00 UTC)