What could possibly go wrong with AI?

[Beebo Brink]

Hackers competed to find AI harms. Here’s what they found. - The Washington Post

As AI chatbots and image generators go mainstream, their flaws and biases have been widely catalogued. We know, for example, that they can stereotype people of different backgrounds, make up false stories about real people, generate bigoted memes and give out inaccurate answers about elections. We also know they can overcorrect in an attempt to counter biases in their training data. And we know they can sometimes be tricked into ignoring their own restrictions.

What’s often missing from these anecdotal stories of artificial intelligence going rogue is a big-picture view of how common the problem is ― or to what extent it’s even a problem, as opposed to an AI tool functioning as intended. While it doesn’t claim to answer those questions definitively, a report released Wednesday by a range of industry and civil society groups offers some fresh perspective on the myriad ways AI can go wrong.

The report details the results of a White House-backed contest at last year’s Def Con hacker convention, which I wrote about last summer. The first-of-its-kind event, called the Generative Red Team Challenge, invited hackers and the general public to try to goad eight leading AI chatbots into generating a range of problematic responses. The categories included political misinformation, demographic biases, cybersecurity breaches and claims of AI sentience.

 

[GoblinCampFollower]

Yeah.... there are inherent reasons why AI's like ChatGPT can't have perfect safe guards. They can mostly just check for key phrases that will create known problems. I have also "tricked" ChatGPT into writing inappropriate stories before just for fun. AI companies will be in a never ending arms race to find more exploits like this. Broadly speaking, if you can describe how you want the AI to do something without using a known forbidden phrase, it will usually do it. I don' think this flaw has a permanent fix.

The bigger threat is some hackers with some money (maybe government backed) will just create their own versions that don't have any safeguards and use that to help them find exploits in cybersecurity. I bet propaganda machines have already been using chat bots to talk to people on the internet for whatever propaganda they want to spread. A chatbot who talks to you like a real person is a lot more persuasive than a poster was ever going to be!

I'm also well aware of the spread of deep fake porn of real people. That genie is out of the bottle forever and hard to control.

 

[Khamon]

Asimov spent half-a-century writing a multi-millennial arc to explain that the three laws are ineffective because machines cannot correctly assess harm to humans. The final solution for humanity is being included in a galactic shared mind, that encompasses all life and lifeless matter, because R. Daneel Olimusk declares it so. But I expect that we'll push the agenda forward anyway.

 

[Monica Dream]

The bigger threat is some hackers with some money (maybe government backed) will just create their own versions that don't have any safeguards and use that to help them find exploits in cybersecurity. I bet propaganda machines have already been using chat bots to talk to people on the internet for whatever propaganda they want to spread. A chatbot who talks to you like a real person is a lot more persuasive than a poster was ever going to be!

Yes. This is what is going to happen -not just governments but deep-pocketed corporations. Think insurance.

Porn is honestly the least of our worries. People losing their healthcare, their jobs and their homes due to AI heads the list..at least in my mind.

 

[Noodles]

The bigger threat is some hackers with some money (maybe government backed) will just create their own versions that don't have any safeguards and use that to help them find exploits in cybersecurity. I bet propaganda machines have already been using chat bots to talk to people on the internet for whatever propaganda they want to spread. A chatbot who talks to you like a real person is a lot more persuasive than a poster was ever going to be!

I'm also well aware of the spread of deep fake porn of real people. That genie is out of the bottle forever and hard to control.

Yeah, they can limit and regulate all these systems on the public web all they want but nothing keeps someone from running a home system to build malicious code or deep fake porn or even just fake AI porn.

 

[GoblinCampFollower]

Yeah, they can limit and regulate all these systems on the public web all they want but nothing keeps someone from running a home system to build malicious code or deep fake porn or even just fake AI porn.

There are already AI's specifically designed for AI porn that are just as easy to access as the mainstream ones.

 

[Kokoro Fasching]

There are already AI's specifically designed for AI porn that are just as easy to access as the mainstream ones.

AI Camera Only Takes Nudes

One of the cringier aspects of AI as we know it today has been the proliferation of deepfake technology to make nude photos of anyone you want. What if you took away the abstraction and put the fak…

hackaday.com

 

[Cindy Claveau]

 

[Free]

Tons? How does one weigh sex with an AI?

 

[Veritable Quandry]

Tons? How does one weigh sex with an AI?

A single-slot server weighs about 26 pounds.

 

[Argent Stonecutter]

Asimov spent half-a-century writing a multi-millennial arc to explain that the three laws are ineffective because machines cannot correctly assess harm to humans.

Actually, he started out because he wanted a set of rules that let him legitimately write mystery/detective puzzlers with an SF background, culminating in the perfect robot story that could be said to draw the line under the three laws, "The Caves of Steel".

Honestly, there was no reason for him to write more robot novels after that, though "The Naked Sun" was pretty good.

Linking the Robot novels and the Foundation series was a huge mistake.

 

[Argent Stonecutter]

SEVERAL MONTHS AGO I attended a preview of the motion picture "2001: A Space Odyssey" here in Boston. Against my better judgment I even got into a tuxedo for the occasion.

Perhaps the tuxedo contributed to an unwitting bit of pomposity on my part, for at one point I dissolved into a semi-irrational spasm of anger.

You see, I have included in a number of my stories something I call "The Three Laws of Robotics" of which the first is: "A robot may not harm a human being or, by inaction, allow a human being to come to harm." The Laws are a purely fictional device, but they have been picked up by other writers, who take them for granted in their robot stories, and over the years I have come to take them very seriously indeed.

In "2001," the most dramatic episodes involve an intelligent computer (equivalent to one of my robots) who deliberately brings about the death of several human beings. That this was going to happen was made abundantly clear to the audience just before the midpoint intermission, and at intermission I went seething up the aisle toward a friend of mine I noticed in the audience.

In tones of deep shock, I said to him, "They're breaking First Law! They're breaking First Law!"

And my friend answered, calmly, "So why don't you strike them with lightning, Isaac!"

Somehow that restored my perspective...