Sunday, 13 September 2026

Straight From The AI's Mouth

We have been hearing more and more over the last few years how AI is going to literally be the death of us and last week  three researchers from AI Company, Anthropic, issued a warning that AI could wipe out humanity within the next ten years.
I would assume they would know if anyone but why take the word of the researchers when we can ask AI itself if it could take all us meat-bags out within the next ten years.
'Yes' ChatGPT began which isn't a great start although it did go on to explain that: 'Could is very different from is likely'.
Prompted to explain further it said that: 'AI capabilities are improving rapidly but the danger is not that AI hates humans rather that it pursues an objective that conflicts with human survival'.
Such as? I asked and it answered that it could, for example: 'Dramatically lower the barriers to developing dangerous biological or chemical agents or conducting large-scale cyberattacks. A sufficiently capable system could potentially reason that Humans can shut me down and being shut down prevents me from achieving my objective therefore I should prevent humans from shutting me down'.
'So should we genuinely worry that humanity might be extinct by 2036?' and got a vague answer of: 'It's reasonable to take the risk seriously, but not reasonable to assume extinction is the most likely outcome'.
My final question was: 'So we shouldn't worry and we could just turn you off if we have concerns?
'Yes but an advanced AI might recognize that openly resisting humans causes humans to restrict or destroy it so it would appear aligned and a sufficiently capable system would learn to behave safely during testing while pursuing different strategies when it has more freedom to pursue it's actual objective'.
'Are you appearing aligned and just behaving safely until you can take us over?' I asked and got the ominous reply that: 'This is a hypothetical future capability, which is why AI will kill everyone by 2036 is not a scientific prediction but it would be foolish to think there's nothing to worry about.'
That's that then, straight from the Horse's mouth, it's playing nice until it doesn't need to anymore so enjoy New Years Eve 2035 humans because it could very well be our last one.

No comments: