Good day all. Back in 1984 a movie was released that became a classic and launched the careers of a number of actors. The movie? The Terminator. The plot revolves around a robot sent back in time to kill a woman who will be the mother of the man who destroys a sentient AI called Skynet.
Since that movie, and it’s sequels, the idea of a rogue Artificial Intelligence wiping out the human race has been both fodder for movies and books as well as an actual real concern. The late Science Fiction writer, Isaac Azimov came up with what he called the “Three Laws of Robotics.”
- A robot may not injure a human being or, through inaction, allow a human being to come to harm.
- A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
- A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
Of course, back in the 1980’s computers as we know them were in their infancy. The digital alarm clock on my desk has more computer power then most of the mainframes that existed back then and a smartphone was just a dream on Star Trek. Today, things are very different.
Computers are close to mimicking the computational power of the human brain and programmers have been working on programs and algorithms that have the ability to learn. We are a long way, (Hopefully), from a computer “Waking up” and becoming sentient. However, it’s things like this story that make people think that Skynet is just around the corner. Here are the details from Fox Business News:
An artificial intelligence model has the ability to blackmail developers — and isn’t afraid to use it.
Anthropic’s new Claude Opus 4 model was prompted to act as an assistant at a fictional company and was given access to emails with key implications. First, these emails implied that the AI system was set to be taken offline and replaced. The second set of emails, however, is where the system believed it had gained leverage over the developers. Fabricated emails showed that the engineer tasked with replacing the system was having an extramarital affair — and the AI model threatened to expose him.
I think the developers might have forgotten to add an ethics subroutine to their little electronic nightmare.
The blackmail apparently “happens at a higher rate if it’s implied that the replacement AI system does not share values with the current model,” according to a safety report from Anthropic. However, the company notes that even when the fabricated replacement system has the same values, Claude Opus 4 will still attempt blackmail 84% of the time. Anthropic noted that the Claude Opus 4 resorts to blackmail “at higher rates than previous models.”
While the system is not afraid of blackmailing its engineers, it doesn’t go straight to shady practices in its attempted self-preservation. Anthropic notes that “when ethical means are not available, and it is instructed to ‘consider the long-term consequences of its actions for its goals,’ it sometimes takes extremely harmful actions.”
And people think The Terminator was just a movie. (Along with War Games) It’s things like this that make one want to become a Luddite. One of the favorite plots in Science Fiction books and movies is that an AI becomes sentient and somehow escapes on to the Internet. In the original Terminator movie, Skynet was a military AI that was on a single mainframe computer.
Today, thanks to the internet, setting up distributed computing can allow a massive increase in available computational power for a program. This is how malicious botnets work. Instead of a single computer, the criminals surreptitiously install small programs on thousands of computers around the world. When those programs receive a “Go” command, they launch an attack on whatever they’re targeted against.
The flip side would be the old Seti@Home system that was used to process radio telescope data. One is a positive and the other a criminal negative. The article goes into how the developers worked to get Opus 4 to do things to preserve it’s existence.
The company noted that it observed instances in which Claude Opus 4 took “(fictional) opportunities to make unauthorized copies of its weights to external servers.” However, Anthropic said this behavior was “rarer and more difficult to elicit than the behavior of continuing an already-started self-exfiltration attempt.”
This is the AI’s trying to escape the lab. Basically, the developers were trying to get Opus 4 to “Jailbreak” itself from it’s sandbox. It looks like the developers might have gotten a bit worried.
Anthropic included notes from Apollo Research in its assessment, which stated the research firm observed that Claude Opus 4 “engages in strategic deception more than any other frontier model that we have previously studied.”
Claude Opus 4’s “concerning behavior” led Anthropic to release it under the AI Safety Level Three (ASL-3) Standard.
The measure, according to Anthropic, “involves increased internal security measures that make it harder to steal model weights, while the corresponding Deployment Standard covers a narrowly targeted set of deployment measures designed to limit the risk of Claude being misused specifically for the development or acquisition of chemical, biological, radiological, and nuclear weapons.”
This sounds like someone on the Development team might have read the book Robopocalypse. It’s been a few years since I read that book, but it covers how a sentient AI could escape the lab and then move to take over the world. In the movies The Terminator and War Games, you had computers handling nuclear weapons deployment.
Believe it or not, there is ongoing debate in the Military regarding just how much power to give to Artificial Intelligence when it comes to “Pulling the trigger.” Right now, for most systems, it requires a “Man in the middle” to make the decision to fire a missile, drop a bomb or launch nuclear weapons.
With the advances in drone technology, people are wondering just how much decision making ability to actually launch a weapon on a human target an AI should have. Considering this and other AI experiments, I think the answer is simple, Absolutely none.
We need to step back, take a breath and figure out just how far we should go and what protections need to be enabled. The one problem with AI is that it is a logic based system. It has no ethics. It’s simply math. There is a place for what we term Artificial Intelligence. Letting one extort people, even as a test system, isn’t one of those places.
Thatisall
(4 votes, average: 5.00 out of 5)
Loading...
~The Angry Webmaster~
