Should we worry that artificial intelligence (AI) will act like the F**k-Up Fairy on cocaine?
AI is fast. But you can't really count on it:
In a recent Air Force experiment, AI algorithms generated attack plans about 400 times faster than human staff, a two-star general told reporters here at the Air Force Association’s Air, Space and Cyber conference. The catch? Not all the AI-generated plans would actually work.
And that was just the relatively straightforward task of matching available aircraft and weapons to a target list. We're not even talking about maneuvering.
If large language model AI is just "probabilistically linking words and sentences together without considering meaning" then at best this is giving conventional wisdom the AI stamp of approval.
When I contemplated inserting AI into command and control when it is corrupted by enemies, I wrote:
The speed of FUBAR could be simply awesome.
But even a working AI as it is now constituted can produce poor plans incredibly faster than even idiot humans.
Still, what if you had five AIs generate plans and only examine the plans that are nearly identical to the plans generated by all the other AIs? Would that weed out just the bad plans or would it knock out good plans, too?
If we can use "AI" to filter, organize, and manage the flood of information commanders must face, commanders could exploit information as the key to battlefield success.
NOTE: TDR Winter War of 2022 coverage continues here.
NOTE: You may also like to read my posts on Substack, at The Dignified Rant: Evolved. Go ahead and subscribe to it. It's the right thing to do!
NOTE: I made the image with Bing.
