3 ms·

Military experts fear the Pentagon’s been sucked into the AI hype machine

13 points by jocker12 5y ago

nonameiguess 5y ago
Agree strongly with all of this, especially the part about requirements not being connected in any way with the actual needs of boots on the ground. My biggest frustration as a defense/intelligence contractor, and what seems like a great failure of government contracting at all as it pertains to technical project management, is the fact that we cannot legally interact with our end-users. It makes the push for "Agile" development fundamentally impossible as Agile development requires constant user feedback. But instead, we only get to interact with an acquisition office. Our customers are not the same as our users, and we develop to the needs of the customer, who may be well-meaning people with some experience and connection the ultimate end users of the systems, or may just be politically driven careerists trying like hell to paint their pet projects as a success to get a promotion. My wife was running QA for a data center provisioner, and at least some of the application provider tenants were people on the same floor of the same building, and they legally couldn't talk to each other. Could only interact via their respective acquisition offices. Everything is a time-delayed game of telephone, hoping your comments are even relayed accurately, if at all. AI hype is just a tiny, tiny slice of this problem. The fundamental problem is the technology we develop may or may not be what the actual users want or need, and the users have no means of directly buying and the developers have no means of directly selling, or of even jointly deciding on requirements and testing.
- kordlessagain 5y ago
  Software is hype at its core. Not everyone understands some problems can be solved with it and some can’t and why that’s a thing even, with different “hardnesses” of problem spaces. See Penrose tilings. Search can solve lots of problems for people however, and integrating Visual and semantic text search with machine learning networks is probably a good way to explore this frontier better by using humans as the fallback processor for non computable “calculations” (guesses explained in stories?). General AI is probably the most interesting outcome, but likely doesn’t have a clear path to “development” due to it being highly cross disciplinary, a bit fuzzy with its obvious use case (new friends) and the seeming extensional threat (bad friends).

13 points by jocker12 5y ago

Agree strongly with all of this, especially the part about requirements not being connected in any way with the actual needs of boots on the ground. My biggest frustration as a defense/intelligence contractor, and what seems like a great failure of government contracting at all as it pertains to technical project management, is the fact that we cannot legally interact with our end-users. It makes the push for "Agile" development fundamentally impossible as Agile development requires constant user feedback. But instead, we only get to interact with an acquisition office. Our customers are not the same as our users, and we develop to the needs of the customer, who may be well-meaning people with some experience and connection the ultimate end users of the systems, or may just be politically driven careerists trying like hell to paint their pet projects as a success to get a promotion. My wife was running QA for a data center provisioner, and at least some of the application provider tenants were people on the same floor of the same building, and they legally couldn't talk to each other. Could only interact via their respective acquisition offices. Everything is a time-delayed game of telephone, hoping your comments are even relayed accurately, if at all. AI hype is just a tiny, tiny slice of this problem. The fundamental problem is the technology we develop may or may not be what the actual users want or need, and the users have no means of directly buying and the developers have no means of directly selling, or of even jointly deciding on requirements and testing.

kordlessagain 5y ago
Software is hype at its core. Not everyone understands some problems can be solved with it and some can’t and why that’s a thing even, with different “hardnesses” of problem spaces. See Penrose tilings. Search can solve lots of problems for people however, and integrating Visual and semantic text search with machine learning networks is probably a good way to explore this frontier better by using humans as the fallback processor for non computable “calculations” (guesses explained in stories?). General AI is probably the most interesting outcome, but likely doesn’t have a clear path to “development” due to it being highly cross disciplinary, a bit fuzzy with its obvious use case (new friends) and the seeming extensional threat (bad friends).