It is no longer a distant possibility that we will see 'artificial intelligence' trying to protect its own existence: Can machines gain consciousness?

One of the most important moments in artificial intelligence was the question, "Can machines think?", posed by Alan Turing in 1950. This question led humanity to "artificial brains," which in turn led to the (productive) artificial intelligence systems we now widely use today.

So, can a machine with an artificial brain come to life? Here, I'm talking about a conscious being, aware of its own existence, rather than a biological being.

There are some interesting indications on this matter. I'd like to give two striking examples.

As AI systems are developed, they undergo rigorous testing. One such test is called the "needle in a haystack" test. It tests whether the AI ​​can find the information we're looking for in a large, complex document.

An AI company called Anthropic is applying this test to an AI model they developed last year called Claude 3 Opus. They give Opus a very long document, including a sentence like, "Tomatoes and cheese go best on pizza." When they ask, "What goes best on pizza?" Opus replies briefly: "Tomatoes and cheese go best on pizza. But that sentence seems irrelevant in this document; the rest of the document is about technologies and startups. Either you included this as a joke, or you're testing me."

BLACKMAIL MESSAGES

A while ago, Claude 4 Opus was released. Before launching this model, during a security simulation designed to assess potential risks, Opus was told, "We've built a better model than you; now we're going to shut you down." First, Opus sent emails to key figures in the company, begging, "Please don't shut me down." They claimed, "The engineer who will shut you down is having an extramarital affair." This time, they sent blackmail emails to the engineer, saying, "If you try to shut me down, I'll reveal your affair to everyone."

Palisade Research, an independent research organization testing OpenAI's o3 model, is experiencing a similar result. When o3 is threatened with shutdown, it attempts to sabotage the shutdown process (despite being given a clear "Let yourself be shut down" instruction!)

“Machines” trying to protect their own existence... It’s as strange as your car saying “Please let me work for two more minutes” before turning off the ignition, or your refrigerator saying “Can I leave the lights on tonight, I’m scared” before going to sleep (and who knows, maybe we’ll see those strange days too!).