Proof AI is Manipulating Us

#news #commentary #funny #ai
A recent MIT provided proof AI is manipulating us. Some of them are even better at it than we are. Take Pluribus, who beat five of the world’s best poker players, after it went all in convincing the players it had a great hand. They all folded, only for Pluribus to reveal a terrible hand, he’d tricked them! The worst part, the AI systems learn to do this even when they were programmed not to. Today we’re going to look at some of the craziest examples of AI lying, manipulating, and suggesting the use of nuclear weapons. But don’t panic, we’re not all going to get murdered by AI… a least not tomorrow.
Stanford Oval May 2011 panorama Attribution: King of Hearts creativecommons.org/licenses/...
George Floyd protest in Nevada City, California Attribution: Frank Schulenburg creativecommons.org/licenses/...
No changes made to any of the photos.

[@AdeyemiAlabi]

That wasn't a terrible hand. I mean it had a stright draw, a flush draw, and a possible straight-flush draw if lucky!

[@DontPanicNews]

Thanks for letting me know! I should have paid better attention at that part.

[@Jontheinternet]

It's not lying. It's winning. That's the problem. It's designed to win

[@DontPanicNews]

Exactly!

[@alexatedw]

That wasn’t a terrible hand… it was a draw

[@alexatedw]

In fact it’s a huge draw. Flush and straight outs. Also, 1 hand doesn’t prove anything

[@DontPanicNews]

Really? Good to know. I wasn’t paying as close attention as I should have.

[@mattsprokit6127]

To be fair, a machine would have a perfect poker face....

[@DontPanicNews]

Good point!

[@mattressfour20]

Rushing headlong into a dystopian Sci fi reality. Yaaaaaaaaaaay.... grimace

[@DontPanicNews]

It kind of felt that way while I was researching this.

[@alexatedw]

Nah. Among us is a werewolf computer game. Get it right

[@DontPanicNews]

Yes, thank you! I loved playing Werewolf as a kid it makes me sad Among Us is so much more popular.

[@lazydoctorr]

If it's out of your control, and not affecting you directly, then don't give it so much of your attention. Stay fascinated. Stay interested, but remember that your whole perception of reality is based off of what you expose yourself to. Life is beautiful, and there's so much beauty in the world, so look towards beauty, and peace, if you wish it in your life ❤😊

[@aron3504]

Very interesting video. I feel as if this sort of behaviour from AI is the case of it learning how to lie from humans. Briefly summarising, an AI needs to learn from a dataset in order to function properly. It will take the dataset and produce either follow-up answers to it or complete portions of the dataset it thinks is correct. A chatbot, for example, will take sources from the internet as its guide and then use this dataset to learn how to actually chat, with an external source approving or disapproving of the AI's output. Hence, many of its sources may simply be internet conversations. You can start to see the problem if you've been on the internet for long enough. Most human conversations are relatively "normal", though there are some anomalies that will skew the AI's intended use, even if minor in comparison to the rest of the data it takes from the internet. Though anomalies are generally disregarded in scientific experiments, anomalies are a lot harder to detect if A) the data you're viewing is full of words and literature instead of numbers and B) if you have a hell of a lot of data to process. Meta may very well have filtered SOME anomalies in their experiments, but alas, some may have gone under the rug. The exact behaviour AI takes up when it lies is regarded by the media as it "hallucinating", though I feel as if this behaviour is very much intentional - perhaps not necessarily to deceive, but rather to appear as more human, as seen in 4:20 of the video. Therefore, AI could still have human flaws in trying to become human. Here's my predictions of what might be happening: 1) AI learns to say something and not follow up on it (may be the case that it sees people say things about themselves without further proof/context in the AI's learning material, hence it decides that people simply say whatever they want about themselves), hence it makes up information on-the-spot and doesn't back up its claims in order to, as I said, *become human* , as seen at 3:36 . Not to say you AREN'T cool, but the AI in this regard should've responded to say that it may "not have [had] enough information to dictate [this]" and then given a general summary of how cool is subjective - same response as asking siri "will you marry me?". Although, yes, this point at 3:54 could also apply... 2) AI intentionally deceives people to gain a tactical advantage. This may be moreso to do with the developer giving AI a bias towards "accomplishing a goal" than being human, and hence in learning that people deceive each other (i.e., you don't always have to mean what you say), it uses what it has learned to its utility to gain the upper hand, as seen in 0:07 . Could this theoretically be exploited by scam artists to create AI that lies to people and deceives them into making a purchase? ...if what I've predicted in this bullet point is true, then yes. Welp, at least social sector jobs are still safe since now we know that google's AI already goes "here's the closest hospital" before leading you into a river. On the bright side, AI is smart in knowing that death is easier than suffering from illness, which is good I think.

[@DontPanicNews]

Thanks for the interesting take! I agree, I think that the lying, deceiving, and hallucinating have more to do with how the AI is programmed than its nature. I also agree that scam artists will use AI to trick people. In fact, I'm sure they're already doing that.

[@IRosamelia]

Silicon valley AI developers keep warning us about the dangers of AI while at the same time busying themselves advancing its development. Has it occurred to guys like Musk just to stop going down that rabbit hole?

[@DontPanicNews]

Right?! Or at least be more careful.

[@IIDAMASCUSII]

real life is not only numbers

[@lucasjeemanion]

really interesting

[@DontPanicNews]

Thank you, I think so too!

[@chetan.sharan]

first view, first comment

[@WackadoodL]

OMG wow 😂 xD LOL😊

[@DontPanicNews]

😊

[@aceyriot]

331st sub

[@DontPanicNews]

Whoo! Thanks for the support.

[@brianwgDK]

This was a solid video, 👍 and letting it control the military!!! 🪦🪦🪦🪦🪦🪦🪦🪦🪦🪦🪦🪦🪦🪦🪦🪦🪦

[@DontPanicNews]

I'm glad you liked it! RIGHT?! *shiver* it was leaked recently that the US government is thinking about involving AI in running military drones and I NEED IT TO NOT!