At a lunch with someone from a big chip company (Duke’s, nothing fancy), they told me how the previous administration was masterful at using leaks from unnamed sources to shape its narrative on why the US government needed to ration chips and control the global GPU market. I was reminded this by a piece yesterday in the NY Times, which looked very much like the previous administration is still trying to exert control.
The story had a number of red flags. It was a concatenation of worries, backed by hints from both named and unnamed sources (unnamed sources are always a warning sign that you may be being steered). Start with the Persian Gulf (always a source of unease), add the threat of AI (a good, if nonsensical, source of fear) and a battle between business greed and “national security,” (both a red flag and a cliché) and cite unnamed “national security officials who worry that the technology could be misused by the Emiratis.” A good story, well-constructed, but still dubious
The alternative, letting US companies supply the UAE with GPUs (it’s in a race with the Saudis to become the Gulf’s computational center) makes more sense, especially since the other choice is for the UAE to buy from Chinese companies, who make a perfectly adequate chip and are improving rapidly enough to compete. Huawei’s Ascend is good enough and getting better.
But let’s examine the charge of potential misuse, since it’s not clear what that misuse would be. The story hints darkly at “the potential to reshape an arms race among nations, and countries eager to develop AI” (a bit hyperbolic, since AI is a software tool, not a weapon). Discussion of AI often begins with an unexamined assertion that if AI is not tightly controlled by governmental and academic mandarins protecting humanity. AI has become an obsession for a risk-averse elite, part of the menagerie of hypothetical risks that is often presented as analysis.
Would the Emiratis design exotic weaponry or unleash cyberattacks on an unsuspecting world. They have never done so in the past and have shown no signs of changing. We are told that the Gulfies might let their AI capabilities be used for some larger global surveillance architecture. The only hostile country with the necessary resources to contemplate a new global surveillance system is China and China does not need the UAE’s help. Presumably the Emiratis could rent processing time to China or Iran for sinister uses, but while the UAE’s relations with Iran have improved, they do not go so far as to help Iran build better weapons. China (and any other country) does not perform sensitive miliary or intelligence-related applications outside its borders on devices controlled by others – the risk of compromise is too great, a consideration that applies across the board and which the story does not to consider.
Ther are also dark suggestions (from experts both named and unnamed) that letting the UAE have GPUs could ultimately contribute to unbounded risk after Artificial General Intelligence arrives. When we examine this risk and consider the number and difficulty of the steps needed to get from data center to Terminator, it is best seen as imaginary. Successful US companies are a source of strength and security. It is no longer the 1980s when we could deny technology to others, and the best way to strengthen security, pace unnamed experts, is to let American companies build the global computational infrastructure of which AI is a part.