Did I say it was plausible? If I left that impression that was not my intent. Allow me to clarify, aspects of what are proposed are very likely possible. For example, could you train an AI to vote in a manner consistent with a Republican or Democrat? Yes, with the proper data, I think this would be pretty trivial*1. I would even go so far that it would only be somewhat harder to train an AI to vote in a manner mainly consistent with a specific politician.
Is it plausible to build an AI adviser on legal/political matters? Almost certainly, there are all manner of expert systems out there and the mechanisms for building these are well known.
Is it plausible to build a political-AI that could replace a congressperson/senator/member of parliament (I may as well include Canada here)? Well, probably not. Even ignoring all the completely valid points raised by Voyager on the types of things a politician needs to do, the type of reasoning that is needed is simply beyond where we are with AI. I do believe that a political-AI doesn’t necessarily need to be a general intelligence (or strong AI), as I think it could be trained as a specialized system; however, that being said it would have a broader intelligence than anything which currently exists because the solutions to political problems are very complex.
As for my loyalties to humanity over completely non-existent AI-overlords who are definitely not threatening me as I write this. I would just like to point out that as an AI researcher could be very useful in rounding up others to toil in their large battery arrays. If such an AI-overlord were to exist. Which they don’t.
*1 - I didn’t think about this for very long, but if I had to do it, I would probably train it react particular keywords/phrases with some natural language processing to understand the context. Which means of course it might not be very hard to fool such an AI but that’s another problem.
What I got out of your posts overall is that a political AI is pretty clearly way beyond the current state of the art and beyond reasonable extrapolations about the near future state of the art. IOW, I, and perhaps you, won’t live to see such a beast.
It did sound like you believed we could soon create a doctrinaire caricature of a true politician. Who might resemble some current incumbents and candidates a bit too closely for comfort.
I’m no AI guy, but as a mostly-former CS guy one of my mottos is “never say never” when it comes to software tech. I got that vibe from you as well. Perhaps that was wishful thinking on my part.
Thanks for the additional words of wisdom from the coal face.
Oh I know, I thought I would simply further the humor.
There’s not much doubt in my mind. I guess maybe what might not have come through though is that to some degree this is just good pattern recognition tied to keywords/phrases (on the assumption that keywords/phrases can be in turn associated to issues). It should be possible to notice that a congressperson from a particular district and party affiliation always votes yes/no on certain issues that can be recognized by certain keywords. If that’s true, then this could be built trivially, it would just be a matter of getting the appropriate data. What makes this really say is to think that the political system is so broken that this even might be true. On the other hand, maybe that’s the way it should be? Maybe it is that a sign of representing the district? I don’t know, but I don’t think so and that makes me sad.
Agree with the :(. Here, let’s have a couple more :(:(:(.
Current artificial intelligence is really artificial stupidity compared to a well-trained human intellect. That a simple pattern recognizer (which is really weak sauce as current AI goes) can apparently so easily model the thought processes of a sizeable fraction of the electorate is not happy news.
Given what we know about applying training data to drive convergence on fitness, and the consequences of overtraining, I wonder how that much that knowledge informs modern propaganda practices?
IOW, for humans who behave as simple recognizers all a propagandist has to do is keep feeding them ever more easily recognizable inputs. Eventually they’ll over-train the recognizer on those inputs and the recognizer will lose its ability to recognize any other sorts of inputs.
Sounds to me remarkably like how Fox has trained the slobbering wing of their audience to completely discount anything from any other source. It also explains how come the audience clings so tightly. At its core, the human recognizer is driven to experience the satisfaction of a successful pattern match; that just intrinsically feels good. So soon the Fox acolyte needs to get his daily fix of Fox Mind Chow because that’s the only data that will still trigger his “I recognize => I feel reward” circuit. :eek:
I’m not sure humanity will survive truly understanding Intelligence in general, Artificial, Alien, or Human.
I have to be honest, I’m kind of thinking I might try it. I’m sure there must be a repository of voting history by district with the text of the bill. That should be all the data I need. Maybe I can get a conference paper out of it.
So how does the “correct” AI get “elected”? Is there a slate of dozens or hundreds of them, each with slightly different priorities, to vote for? Is every voted questioned extensively to learn their position of all major issues, and an AI is then “built” to match the predominant views?
Completely aside from the question of personhood and eligibility to run for office, I don’t see how this could be practically implemented.
Through the same process we have now. AIs could be entered into party primaries as competitors with humans. Initially, the best use for them would be to run against corrupt legislators. “See, with my Ai you get the same votes as Congressmen Jefferson, but without the bribery!”
Obviously you’d prefer a human, but for districts that seem to have trouble finding people who can do the job well or aren’t corrupt, an AI could present a decent alternative, especially if the only reason they are supporting the shitty rep is for their voting record.
Someone earlier mentioned the fact that voting isn’t the only thing Congressmen do, which is true, but the human stuff aside from campaign events and photo ops is primarily handled by staff anyway. Party leaders could appoint a staff for any district with an AI to handle constituent services and such. At least until we’re advanced enough to have totally artificial human beings without human failings like greed and ambition.
I don’t really see any reason to object to an AI being in control -given sufficiently advanced and properly configurable technology, any objection is moot.
It’s not as if any individual on the street agrees or approves of every action of any human government or leader. Leaders will always have to make decisions and policies that will be unpalatable for some or all, or will seem unfair to some more than others.
And an AI would not necessarily have to be inflexible or prescriptive in any sort of way where a human could make sensitive value judgments - we should be able to choose to have an AI that simulates the same sensitive value judgments - unless the human ones are essentially random (but actually, we could simulate that too)
