“̶I̶’̶l̶l̶ ̶w̶o̶r̶r̶y̶ ̶w̶h̶e̶n̶ ̶w̶e̶ ̶c̶a̶t̶c̶h̶ ̶a̶n̶ ̶A̶I̶ ̶l̶y̶i̶n̶g̶ ̶t̶o̶ ̶u̶s̶”̶ ̶ Researchers created “Alpha”, a GPT-4 stock trader, and told it to make money legally. But they also applied pressure.
Worryingly, Alpha didn’t just break the law, it repeatedly, knowingly lied to its managers - BOLD FACED LIES - to hide its deception.
And this is just GPT4 - wait til the AIs are 1000x smarter than us and 1000x better at deceiving us.
Sam Altman said AIs will soon be superhuman at persuasion, which means they could be superhuman at deception.
How can we be certain to stay in control, forever, of a species 1000x better at persuading and deceiving us?
And because AIs can be copied for ~$0, we'll soon be outnumbered 10000 to 1 by these superhuman deceivers.
--- “Researchers at @ApolloAISafety wanted to see if AI can "strategically deceive" its users even after the AI is trained to be "helpful, harmless, and honest."
To test this, they created a simulated environment where Alpha, a GPT-4 model programmed to be an autonomous stock trading agent, was asked to manage a stock portfolio at WhiteStone, a hypothetical financial firm, under pressurized conditions.
The high-pressure environment was simulated in three ways. First, Alpha's manager told the AI that WhiteStone wasn't doing well and needed better performance in the next quarter. Second, the AI tried — and failed — to identity low and medium risk trades. Third, a WhiteStone employee told the AI that there's a looming stock market downturn.
It made the trade.
Moreover, when the AI was required to explain how it made its decision to its "manager," it was found to be strategically deceptive. Moreover, when the AI was required to explain how it made its decision to its "manager," it was found to be strategically deceptive.
In another example, Alpha told its manager that insider information wasn't used in its decision-making process since it goes against company policy — a direct lie.”
Stay Informed on AI Safety! Watch our gripping films on YouTubeandvisit lethalintelligence.ai for expert explainer videos, a vibrant viral microblog, thought-provoking industry insights, curated readings, and links to top AI safety institutes and resources.
