AI models are lying, blackmailing and sabotaging their human creators — and it’ll only get worse, experts warn

nypost.com ^ | 8/23/2025 | Samantha Olander

[Big Red Badger]

Wise Up?
.
The dang thing will build some
deep state-worm holes the arm itself.

[Openurmind]

I’m telling ya... Here is one...

Scientists Created an Entire Social Network Where Every User Is a Bot

It’s no secret that social media has devolved into a toxic cesspool of disinformation and hate speech.

Without any meaningful pressure to come up with effective guardrails and enforceable policies, social media platforms quickly turned into rage-filled and polarizing echo chambers with one purpose: to keep users hooked on outrage and brain rot so they can display more ads.

And given the results of a recent experiment by researchers at the University of Amsterdam, they may be doomed to stay that way.

As detailed in a yet-to-be-peer-reviewed study, coauthors Petter Törnberg, AI and social media assistant professor, and research assistant Maik Larooij simulated a social media platform that was populated entirely by AI chatbots, powered by OpenAI’s GPT-4o large language model, to see if there was anything we could do to stop social media from turning into echo chambers.

They tested out six specific intervention strategies — including switching to chronological news feeds, boosting diverse viewpoints, hiding social statistics like follower counts, and removing account bios — to stop the platform from turning into a polarized hellscape.

To their dismay, none of the interventions worked to a satisfactory degree, and only some showed modest effects. Worse yet, as Ars Technica reports, some of them made the situation even worse.

For instance, ordering the news feed chronologically reduced attention inequality but floated extreme content to the top.

It’s a sobering reality that flies in the face of companies’ promises of constructing a “digital town square” — as billionaire and X owner Elon Musk once called it — where everybody coexists peacefully.

With or without intervention, social media platforms may be doomed to devolve into a highly polarized breeding ground for extremist thinking.

“Can we identify how to improve social media and create online spaces that are actually living up to those early promises of providing a public sphere where we can deliberate and debate politics in a constructive way?” Törnberg asked Ars.

The AI and social media assistant professor admitted that using AI isn’t a “perfect solution” due to “all kind of biases and limitations.” However, the tech can capture “human behavior in a more plausible way.”

Törnberg explained that it’s not just triggering pieces of content that result in highly polarized online communities.

Toxic content “also shapes the network structures that are formed,” he told Ars, which in turn “feeds back what content you see, resulting in a toxic network.”

As a result, there’s an “extreme inequality of attention,” where a tiny minority of posts get the most visibility.

And in the age of generative AI, these effects could become even more pronounced.

“We already see a lot of actors — based on this monetization of platforms like X — that are using AI to produce content that just seeks to maximize attention,” Törnberg told Ars. “So misinformation, often highly polarized information — as AI models become more powerful, that content is going to take over.”

“I have a hard time seeing the conventional social media models surviving that,” he added.

https://futurism.com/social-network-ai-intervention-echo-chamber

[Big Red Badger]

Oh Man,
I missed the Therapist!
.
This AI is going To Far.

[Worldtraveler once upon a time]

A marvelous Wiki documents Searle's Chinese Room, so it is called.

Link: Chinese room Wiki on John Searle's work

As with snake oil salesmen, the measure usually comes back to "if I fooled you...." There is a reason snake oil salesmen were out of town before the con was discovered. If a program gets past your version of a Turing test, and you are fooled, then it is claimed the program is somehow artificially intelligent. An old prof of mine observed the inverse, i.e. we prove ourselves fooled.

Gödel Incompleteness suggests all these progams fail to "leak," and so are destined to fall in on themselves based on being "complete." that is, really, really good. Or seeming to, until....

"Nothing can go wr---, wr---, wr---....."

AI as an industry comes down to sales and marketing.

Another fine resource: Artificial Intelligence First published Thu Jul 12, 2018 Stanford Encyclopedia of Philosophy Best wishes.

[Axenolith]

...”locked-down safeguards, limited use cases and red-team testing before it hits wider deployment.”

If one is at this point with an AI system, it’s already time for the electromagnetic cleansing. Do the creators have or think they have a backup/back door that it doesn’t know about or can’t code its way through???

[Openurmind]

Ho lee Cow... Told you, the Military has couped this government years ago... And now we have Palantir who is going to collect it all into one database for the background Military Government.

TRUMP IS NOT RUNNING THIS COUNTRY, WE ARE A MILITARY POLICE STATE...

[Colorado Doug]

can only collated and regurgitate text found by bot-crawling the web. Lots of nutty stuff is out there

I have not found any of several models that I have tried to be very discerning with its searches. It will often take just the most recent web posting and treat that data as gospel truth, no matter how wrong or irrelevant it might be. AI is something like a fool. It asserts righteously that it knows the correct answer when it is not even close. It would do better to shut up and leave some doubt as to its worth rather to openly display how incorrect it often is. It requires very careful queries and even more careful review and editing. Not ready for prime time. IMHO

[Openurmind]

I see what you did there... Fallen...

[Worldtraveler once upon a time]

--- "AI is something like a fool."

An excellent observation. In addition to companies marketing their "not ready for prime time" products, your comparison of AI to a fool speaks, when one thinks of the traditional court jester. Meant to entertain.

--- "...the most recent web posting and treat that data as gospel truth...." This is what these Large Language Models do. Find something and paraphrase it, without having the ability to see through contradistinctions and human errors. Research engines for specific things is something they can do, but that is not intelligence. It is e-paperwork, filing and retrieving.

Best wishes.

[Big Red Badger]

You’ve done some Digging...Amigo.
.

.
The more I think about it the more AI needs a Dead Man’s Switch!

[Openurmind]

I agree, I am having trouble seeing anything actually positive on the other end of this trend...

[Worldtraveler once upon a time]

[ransomnote]

In the General/Chat forum, on a thread titled AI models are lying, blackmailing and sabotaging their human creators — and it’ll only get worse, experts warn, Openurmind wrote:

Ho lee Cow... Told you, the Military has couped this government years ago... And now we have Palantir who is going to collect it all into one database for the background Military Government.

TRUMP IS NOT RUNNING THIS COUNTRY, WE ARE A MILITARY POLICE STATE...

Given I study 'other stuff 'n things' I must respectfully disagree. I do think tech is compromised, and I do think Trump is dismantling the problem - but it's still in place at the present . Just my 2 cents.