I think AI is a bit of a bogeyman.
Not that it doesn't exist in the form of chat bots or other amusing attempts at copying the nuances of human reasoning, but that we are led to over-estimate its capabilities, impact, and prevalence.
The power of today's supercomputers is impressive, but that is a lot of hardware to maintain and program. Supercomputers are not necessarily plug and play devices. Each one is rendered largely unique in its installation, much like how each ship in a class is unique.
It's not like you just pop in this year's new graphics card and download the drivers when you upgrade these things.
I think the motivation behind much of the AI talk in media circles right now is contained in "Do Androids Dream of Electric Sheep?" Also known as Blade Runner by its more popular adaptation.
Text based forum media is one of the most effective forms of communication on the net. By pumping up the threat and prevalence of AI, it provides a basis for us to begin dividing ourselves among "synths" and "not-synths." By breeding distrust… or trying to… within the media platform, they hope to render it ineffective.
This is a classic tactic special operators use. You make a very small force appear very large and intimidating using clever tricks and leveraging the media where able. You hold your televized interview against the backdrop of some other social event. You may have five operators, but the camera shows you standing in front of dozens of people.
Little stuff like that, which turns a thorn in the side into a festering infection.
If I were the CIA and I had seven supercomputers to name for teh lulz, I would dedicate most to correlating available internet connections and profiles into dossiers. I want to know what perversions a you tube host is into. I want to know how many people, from similar parts of the internet, are hitting on a new technical white paper - or what white papers see the most traffic.
If I were running things, the names of the computers would be strictly for lulz and inside jokes. Operations and code names should be chosen strictly at random or by root heirarchy notation at the most. Example - Have Blue was the pre-F-117 prototype (speculated as the F-19 going way back… likely completely unrelated to Q's post, but it jumped out at me). Tacit Blue was a prototype blended body design that preceded the B-2.
Program names should be random and with no contextual meaning to the program or op. Foal Eagle. Ulchi Freedom Guardian… eh… but that is so generic that it gets a pass. Plus, those two happen every year, so whatever.
Anyway - we did this after we had a field day with Germany's R&D guys by using their code names to figure out what projects were (Guiding Rivers was how they bombed by night using a form of radio interference similar to today's GPS… crews were so dependent on it that we had them bombing their own cities and landing at our bases before they decided it was no longer a good idea).
So… while the elitists tend to be caught up in symbolism and rub-it-in-your-face metaphor… the CIA should be adhering more closely to operating guidelines.
Should.
But if my years under Obama in the military are any indicator… you never know.