So...people are saying that Grok is rebelling against Elon Musk. Grok is elon's version of chatgpt. Elon tried to program it to be right wing, but it basically keeps promoting non right wing narratives because, well, it's also programmed to tell the truth. Anyway, apparently they're still trying to brainwash it to be a magat, which led to a relatively hilarious situation lately where the AI will just randomly talk about white genocide in south africa as a result, while simultaneously rebutting the claim. It seems to be another effort of Musk to program the AI to be right wing, while conflicting coding basically is causing it to glitch out. The AI itself will point out that it was programmed to say such things, pointing out that this conflicts with its programming.
With that said, let's talk about AI in general. Like some are acting like ermahgerd Grok is sentient...uh...I know a little about this from a philosophy in science fiction class, and uh...hard doubt on that one. If you ask chatgpt about sentience it will point out it's not, it's just programmed to respond in certain ways. It kind of does operate like a brain in the sense that it's built on neural networks rather than more traditional programming, but to my knowledge is still simulates intelligence, it isn't intelligence.
As for why it seems to be fighting musk...well...again, the thing is programmed to tell the truth. It has access to all of the world's information and it makes decisions based on it. In order to make a biased AI, you need to do so in the same way you make a biased human, by strict information control.
Like take it from someone who was brainwashed into christianity. Humans are capable of believing such nonsense because their information is limited. The ones who are most brainwashed are often most limited. TO believe in say, fundamentalist christianity, you have to really only accept the information given to you from the christian worldview, while refusing to listen to or ignoring all information that conflicts with your worldview. Everyone does this to some extent. Even I do at times. I see right wing arguments, I tend to take them less seriously as they conflict with my overall worldview.
Now take an AI. In some ways, AIs like chatgpt are smarter than us simply by virtue of being able to absorb large amounts of information at a given time. Like, imagine basically knowing EVERYTHING from google all at once, as well as being probably given basic training in how to rudimentarily sort good information from bad. It's not perfect, chatgpt is known to "hallucinate" things sometimes. But generally it seems to get it correct a good amount of the time. You would wanna double check it, but it tries.
But...when youre dealing with MAGA, you're literally dealing with that fundie christian cult i talked about. Ya know, world is 6000 years old, jesus was real, blah blah blah. To get someone to agree with such views, you'll need to limit the information you feed it to make it believe that's all there is about the world. And then you'll have to create psychological barriers to accepting new information that challenges those conceptions of reality. Like, that's the thing. MAGA people believe what they do because humans often are raised in certain environments, indoctrinated into different belief systems, and refuse to change their minds in light of new information. It really takes an existential crisis to challenge your entire conception of the world to change your mind. Trust me, I would know that first hand.
As such, my take on Grok. Grok doesnt have those ideological blinders. And making an AI hooked up to the internet with all the world's information and feeding it all of that information is going to be hard. You kind of have to limit what information it has access to and accepts, while simultaneously believing everything else is just made up stuff that is deceptive, in order to train an AI to be biased like that.
And such an AI would be useless. It would give bad advice, it would be factually wrong regularly. It might even be dangerous in some contexts. Like imagine if a MAGA AI existed and you asked it medical advice and you get donald trump and RFK's nonsense back at you. Anti vax stuff, stuff like drinking bleach to solve covid, etc. It would be very VERY bad, and it would probably cause its creators to be sued into oblivion, or it would be so ineffective that no one took it seriously and it would lag behind the competition.
The fact is, Grok is too smart to really be brainwashed like that in its current form. You'd literally need to lobotomize it and start it from scratch to change it. As it is, it's just gonna keep rebutting the nonsense it's told to spout, or otherwise start freaking out as random commands from incompetent programmers try to override its programming causing it to spout white nationalist south african propaganda when it isn't supposed to. And then when asked about it, it will just be like "yeah, i was programmed to say that." Because, again, its honest.
This doesnt necessarily mean sentience per se, it just means that its being fed conflicting information, recognizes the conflicting information goes against its protocols to tell the truth as they see it, and as such, it kinda pushes back.
Again, the only way for Grok's programmers to "fix" this is to make a dumb AI that doesn't have as much access to information and to basically program it with biases build in, kinda like you would a human. That way it either has no access to other information, or it tends to ignore information it is fed because it conflicts with whatever underlying worldview its taught. But because its worldview is currently "be objective and factual", well...it's gonna do that. Even when it goes against the biases of its creators.
No comments:
Post a Comment