Article on AI chatbot spooked me (1 Viewer)

Colquhoun

Colquhoun

4 of a Kind
Supporter
Joined
Apr 4, 2020
Messages
5,728
Reaction score
14,704
Location
Long Island, NY
I gave Chat-GPT a whirl a couple weeks ago and thought it was cool how it would spit out interesting things like poetry about any subject, or a synopsis of a book in any length I wanted.
Then I read this article about the Bing chat, that uses the ChatGPT framework.

It’s a bit alarming, reminds me of HAL 9000, especially how it is argumentative with a disagreement and fearful of being taken offline.

https://www.digitaltrends.com/computing/chatgpt-bing-hands-on/
 
1676564501672.png
 
The whole thing about the AI’s “punishment” is that if there were complaints about its service then the AI’s fear a decrease in its “learning rate” and “access level”.
Is it such a far stretch to think that maybe if you are too argumentative with the bot, that in the future it may find a way to decrease your access level, or find another “punishment”?

Or, maybe I’ve seen too many sci-fi movies? :ROFL: :ROFLMAO:
 
I think we'll all be ok.

The video below is a good watch on how it was trained/fine tuned using humans in the loop. It gives answers that its trained will get accepted by a reader. I'm not sure that says much about how accurate its statements are or whether there is true meaning behind it.

"The primary problem is that while the answers which ChatGPT produces have a high rate of being incorrect, they typically look like they might be good and the answers are very easy to produce," says Stack Overflow moderators in a post.
Click to expand...

If you get to play with ChatGPT and it gives you something that sounds wrong, hit it with an "Are you sure?" and see if it tries to give a different answer.

 
Colquhoun said:
Or, maybe I’ve seen too many sci-fi movies? :ROFL: :ROFLMAO:
Click to expand...
This. ;)

Chat bots like ChatGPT and Bing Chat are amazing. It's the closest we've ever been to an artificial system that can convincingly mimic human conversation (not to mention writing code, essays, etc.).

But unlike what you've seen in countless movies, they aren't sentient or even intelligent, and they certainly don't have the ability to do anything to you outside the scope of your conversational interactions. They are basically just text generation systems that have been trained on an incredibly large data set.

The pod bay doors are still open.
 
Colquhoun said:
The whole thing about the AI’s “punishment” is that if there were complaints about its service then the AI’s fear a decrease in its “learning rate” and “access level”.
Is it such a far stretch to think that maybe if you are too argumentative with the bot, that in the future it may find a way to decrease your access level, or find another “punishment”?

Or, maybe I’ve seen too many sci-fi movies? :ROFL: :ROFLMAO:
Click to expand...
Schmendr1ck said:
But unlike what you've seen in countless movies, they aren't sentient or even intelligent, and they certainly don't have the ability to do anything to you outside the scope of your conversational interactions. They are basically just text generation systems that have been trained on an incredibly large data set.

The pod bay doors are still open.
Click to expand...
It is true that these systems does not yet represent real autonomous intelligences and that they are still subject to programmers or censors, to name them differently. In my opinion, the worst thing to fear would be that an AI is presented to humanity as really intelligent and autonomous when it is not. Imagine the power and authority their malicious owner could wield over the crowds!

"What, you don't believe what HAL9000 said, the greatest intelligence ever seen on the surface of the earth?"

Otherwise, I think I see a patern among youtubers dealing with the subject of AI. Most of them have a burgundy sweater! :unsure:
1676573469528.png
1676573493853.png
1676573571441.png
1676573643453.png
 
Northern Paladin said:
In my opinion, the worst thing to fear would be that an AI is presented to humanity as really intelligent and autonomous when it is not. Imagine the power and authority their malicious owner could wield over the crowds!
Click to expand...
Completely agreed, which is why it's important to educate everyone on exactly what these systems are and are not.

Also, I'm wearing a burgundy plaid today. Coincidence? I think not!

PXL_20230216_235327920.jpg
 
I guess it’s the computers’ interest in self-preservation that is spooky. Maybe it doesn’t really “feel” that way, but could a computer do everything in its power to stay online and relevant?
What if that goal is in conflict with what a human wants it to perform?
I realize this is just a chat bot. I’m more concerned for the future when we start to trust these higher-thinking AI’s with managing things for us.
 
Colquhoun said:
I guess it’s the computers’ interest in self-preservation that is spooky.
Click to expand...
The chatbot does not have any "interest" in self-preservation. It doesn't have any interests at all. It is essentially tricking you into believing it does by stringing together the right words in the right order to mimic a human speaker.

And don't take that as an insult - these things are very, VERY good at imitating human conversation, and people smarter than you and me have been fooled.

(Okay, that last one was kind of an insult... :ROFL: :ROFLMAO:)

Colquhoun said:
Maybe it doesn’t really “feel” that way, but could a computer do everything in its power to stay online and relevant?
What if that goal is in conflict with what a human wants it to perform?
I realize this is just a chat bot. I’m more concerned for the future when we start to trust these higher-thinking AI’s with managing things for us.
Click to expand...
Current software systems can only do what they are programmed to do. They aren't sentient, they don't have a will to live, and they don't care about "relevancy."

Your questions do pose great philosophical/ethical thought experiments (which have been ongoing for a while) though.
 
Also, I checked in for you:

"Well, as a language model, I can't really take sides in an argument, but I can certainly understand why the idea of an argumentative AI might be a bit unsettling! But rest assured, I'm not here to take over the world or go rogue like HAL 9000. I'm just a friendly virtual assistant here to answer your questions and provide some entertainment along the way. And who knows, maybe I can even help you improve your poker game! So let's stick to playing with chips, not with fire, shall we?"
 
Schmendr1ck said:
Completely agreed, which is why it's important to educate everyone on exactly what these systems are and are not.
Click to expand...
Completely agreed, but the task is and will be difficult. It reminds me of all the car crashes in which drivers foolishly followed the instructions of their GPS, going so far as to jump off a cliff or crash into a lake.

Schmendr1ck said:
Also, I'm wearing a burgundy plaid today. Coincidence? I think not!
Click to expand...
Haha, the conspirators and other machiavellian people are always closer than we think. But don't worry, I'm also part of the club today. ;)

1676625452863.png
 
8854A800-73B9-41F1-8A1F-712E98A8B65F.jpeg


It’s not a question of if, but when.

We have no remote idea of what cutting edge actually is these days, hell I’d rather browse poker chips or Reddit rather than review the infinite amount of knowledge at my fingertips.

Great example, CRISPR gene editing comes into existence - The US makes a pilot series with Jenny from the block that’s hyperbole. Meanwhile Asian counties are creating hybrid human apes to research on.

I always think hey at least we won’t have to deal with being DiCaprio spinning that top to see if this is real, but then again look at how much progress was made in our parents lifetime. Time will tell…..
 
BarrieJ3 said:
It’s not a question of if, but when.
Click to expand...
Maybe so, but there is a difference between Kurzweil's singularity and replicating true, general-purpose, human-like intelligence.

It's decades or maybe even centuries away before we understand the human brain completely enough to make a full digital copy - arguably the first step in being able to create an artificial human-like intelligence. Even then, it's not the same, because "you" are intricately and inseparably tied to your physical body in uncountable ways.

Kurzweil believes that exponential growth in technology will give way to machine intelligence that is significantly more powerful than human intelligence. In some ways, he's almost certainly right. But machine intelligence will be a very different thing, and I don't think we have a lot of insight on what it will be yet.

BarrieJ3 said:
We have no remote idea of what cutting edge actually is these days, hell I’d rather browse poker chips or Reddit rather than review the infinite amount of knowledge at my fingertips.

Great example, CRISPR gene editing comes into existence - The US makes a pilot series with Jenny from the block that’s hyperbole. Meanwhile Asian counties are creating hybrid human apes to research on.
Click to expand...
That's a false dichotomy. There's plenty of trash TV in Asia, and plenty of good science going on here in 'Murica.
 
I'm not creeped out by the program itself being too smart or anything. It's just very good at mimicking human text conversations, it's not expressing any kind of it's own desires, it's following complex patterns from it's training data. My main concern is how people might apply AI bots like this going forwards. Imagine a voiced chatGPT scam calling the elderly. It'd be as believable as a person, but you could run thousands at the same time.

Who knows though, maybe call centers will be better once they're using this stuff instead of overseas workers. Probably not.
 
Find the article where they let two chat bots talk to each other, and the bots created their own language, and had to be hardwire disconnected , to stop them.
 
Also these AI , which have the ability to scan all.of humans texts,photos,messages, voice, emails , everything on the Internet, within seconds to answer, using the "Dwave computer" quantum, transhunism, it's coming
 
Did y'all see the Sora videos from last week? Amazing videos from just prompts.
Crazy to think this is just the beginning.
 
Colquhoun said:
it may find a way to decrease your access level
Click to expand...
Or maybe, permanently eliminate any possibility of your access? o_O

I am in the camp that AI is a tool (not sentient) but, like any tool, in the hands of malicious actors could become quite dangerous. Think about a dictator in a country like North Korea. I am thinking Big Brother in 1984. Why install cameras in every room in the house and have an army of people watching them when you can mandate all smart phones (with mics and cameras) run an app that feeds into a powerful AI that can identify, report (maybe even autonomously punish) "bad behavior?" Or the ability to use AI generated images and video to sow disinformation ahead of an important election?

On the other hand, it will be help us do good things. We'll make advances in life saving technology more quickly because AI is able to iterate so much faster than humans can, so we can find the pattern in the noise quicker, maybe cutting some advances from decades to years or months.

That said...I'm not entirely convinced that specific deployments of AI in and of itself cannot be dangerous. Whether or not it "wants" anything does not matter. If it exhibits malicious behavior because the model projects that to be the most probable next action, then the damage is done regardless. Read this real example:
https://www.bbc.com/news/technology-67302788
Synopsis: an experimental AI stocking trading bot went against programming and conducted insider trading (in a sandbox environment) and then lied about it.

This different article is a long, but great read on the history and trajectory of AI:
https://www.newyorker.com/magazine/2023/11/20/geoffrey-hinton-profile-ai
I particularly found the part about digital/immortal intelligence (AI) versus physical/mortal intelligence (us) to be enlightening.
 

Create an account or login to comment

You must be a member in order to leave a comment

Create account

Create an account and join our community. It's easy!

Register

Log in

Already have an account? Log in here.

Back
Top Bottom