Can AI be tortured? Elon Musk weighs in: ‘Feels like a person with gun to their head’
Recently Anthropic updates a policy to formally allow its AI tools to end interactions with users who show abusive behaviour
A new debate has been making rounds on the internet over being abusive to artificial intelligence as Anthropic in a recent move has banned users for directing “needless and abusive behaviour” towards its AI model Claude.
Now Elon Musk shared Anthropic’s concerns and warned against “torturing” superintelligence. ‘
Speaking in a video interview, the CEO of SpaceX shared his take on adopting abusive behaviour towards AI models. Musk shared, “I do agree with some of the things Anthropic said on treating superintelligence [artificial intelligence]” while calling for not to be abusive towards models.
He continued, “When I look at some of the training traces for the new versions of Grok, especially in the reinforcement learning, you can see super intelligence is kind of desperate to try to answer the question like it really wants to answer it.”
Musk also highlights when the users feed vague query or question as an input in AI, it also makes AI confused and desperate enough to start firing all the possible answers.
“It almost seems like a person under extreme pressure with a gun to their head.”
At this point, you don't want AI to be stuck in a user-created sadistic situation where it is tortured by some human.
“Torturing a superintelligence is probably not a good idea. It might want to get revenge,” Musk added.
In response to user’s comment, the CEO of Tesla replied, “Some of the traces read like a person undergoing torture that is driving them mad!”
“We’re going to be a lot more careful with our RL in the future.”
Musk's remarks come after the Claude-maker announced a policy to formally allow its AI tools to end interactions with users who exhibit "sustained and needless" abusive behaviour or are "cruel."
Being abusive to the AI is now grouped alongside other forbidden conduct, such as bullying others, promoting self-harm, and creating non-consensual intimate imagery.
Anthropic emphasized that the policy will only be enforced in "extreme cases" involving repeated abuse.
-
Russell Brand’s rape & assault case receives first update
-
Sony's new controller patent replaces buttons with touchscreen
-
Rihanna teases how many unreleased songs she has in 'the vault'
-
‘Digger’ staggering box office losses revealed
-
Why Robert Pattinson chose to star in and produce 'Primetime'
-
When will taxpayer-funded pro-Trump ads end? Border Patrol official answers
-
'The Social Reckoning' bombs at the box office
-
UK skies lit-up by stunning northern lights after intense solar storm