The AI dooooooom thread

[PJL]

'I was only following AI' is the new 'I was following orders' excuse.

[Jacob]

'I was only following AI' is the new 'I was following orders' excuse.

Yeah, but the systems are set up such that the humans involved don't have any way of questioning the AI. There's no point in the process where there's a human with information to challenge the output of the software.

[Baron von Schtinkenbutt]

The only thing "new" about this is using the term "AI".  Automatic target recognition failures have been blamed for tens of thousands of civilian deaths over decades.  Like nearly all the problems "AI" is being blamed for creating, it's once again exacerbating an existing problem.

[Jacob]

The only thing "new" about this is using the term "AI".  Automatic target recognition failures have been blamed for tens of thousands of civilian deaths over decades.  Like nearly all the problems "AI" is being blamed for creating, it's once again exacerbating an existing problem.

Yeah, that's the point the article makes.

And my take on it, that AI is another mechanism for further diffusing any responsibility for the outcomes, and that that mechanism is a separate function from the AI decision making process itself.

[crazy canuck]

Aside from the errors AI makes in drafting legal arguments, the main preoccupation of in house lawyers these days is how they are going to manage the tsunami of legal issues and legal risks they see coming because of the pressures coming from the executives to use AI, without anyone understanding how it should be used and how to mitigate all the errors it makes.

The only thing "new" about this is using the term "AI".  Automatic target recognition failures have been blamed for tens of thousands of civilian deaths over decades.  Like nearly all the problems "AI" is being blamed for creating, it's once again exacerbating an existing problem.

Something that might be new is the high trust and low verification occurring with the use AI.  Of necessity for anyone pushing the use of AI - if anyone was to spend the time and resources needed to check the work and fix all the errors found, the use case for AI would be dramatically reduced. 

[Sheilbh]

Yeah, that's the point the article makes.

And my take on it, that AI is another mechanism for further diffusing any responsibility for the outcomes, and that that mechanism is a separate function from the AI decision making process itself.

I've mentioned it before but Dan Davies' The Unaccountability Machine is fantastic on this and the artificial intelligences that are the corporations and government bureaucracies that already run the world.

And why economics needs to be crushed as a discipline.

[Legbiter]

Niftiest use of a trillion-dollar autocomplete script I can personally think of, where it would actually be genuinely useful, is in training it on the entire written corpus of, say, Old Norse and Old English plus Latin. Or medieval Polish and medieval Lithuanian, throw in Latin and thus giving the young scholar a fantastic research tool without having to first learn all these languages plus reading the entire corpus (impossible in one human lifetime) and then being able to ask said Trillion Dollar Clippy some very pointed research questions.    

There's probably an enormous number of similar niches where it's fantastic. Most commonly, it's probably going to be used by male shut-ins to make customized porn clips and virtual boyfriends for their female equivalents.   

[Valmy]

There's probably an enormous number of similar niches where it's fantastic. Most commonly, it's probably going to be used by male shut-ins to make customized porn clips and virtual boyfriends for their female equivalents.   

Oh yeah. This is a multi-trillion dollar porn machine. That is where it is going to make most of its money back.

[Legbiter]

[Legbiter]

Everybody always talks about an out-of-control AGI that will paperclip or Skynet us all. But what if the God-Machine, once birthed by a seriously mentally ill SF "Effective Altruist" type is instead more like a concerned zookeeper observing a degenerate group of pandas failing to breed effectively?

And after going through the data in a microsecond it just goes "it looks like electricity causes some strange distress in Homo Sapiens.  I'm going to take it away for your own good!"

Leaves a few scattered medical stations around and goes on to do it's own thing in space. And we get back to this. 

No more Zoom, no office politics, no traffic.