AI is getting better at tricking humans

Maybe Mission Impossible: Dead Reckoning is on to something: Artificial intelligence systems are getting better at deceiving humans in ways they haven't been explicitly trained to do, according to new research published in the journal Patterns.

Some examples:

• Meta developed a program called Cicero that performed in the top 10% of human players at the world conquest strategy game Diplomacy… but to get there, the AI told premeditated lies, colluded to draw other players into plots, and justified an absence after being rebooted by telling another player: “I am on the phone with my girlfriend.”
• In a test conducted by researchers, OpenAI’s GPT-4 tricked a TaskRabbit worker into solving a Captcha by pretending to be vision-impaired.
• In another test conducted by researchers, AI organisms in a digital simulator “played dead” in order to trick a test built to eliminate AI systems, before resuming activity once testing was complete.

The programmed aim in these instances wasn’t necessarily to deceive… but they illustrate AI systems resorting to a Machiavellian-style “achieve my programmed goal by any means necessary” approach. Which could be difficult to correct for, given the “black box” problem that characterizes state-of-the-art machine-learning models: it’s impossible to say exactly how or why they produce the results they do – or whether they’ll always exhibit that behavior going forward, MIT Tech Review reports.

🤖 In other AI news: OpenAI yesterday launched its GPT-4o model, a newer, much faster version than GPT-4 with improved capabilities across text, vision, and audio.