Ari Daniel
👤 SpeakerAppearances Over Time
Podcast Appearances
Researchers found that numerous AI models affirmed someone's problematic or wrong behavior about half the time.
To consider the impact, they had 800 participants interact with an AI about a real conflict from their lives.
Those who'd interacted with an affirming AI were 25% more convinced they were right and 10% less willing to do something to fix the situation.
Stanford University computer science Ph.D.
student Myra Chang says that companies and policymakers should work together to fix the problem.
Especially the tough conversations.
Researchers found that numerous AI models affirmed someone's problematic or wrong behavior about half the time.
To consider the impact, they had 800 participants interact with an AI about a real conflict from their lives.
Those who'd interacted with an affirming AI were 25% more convinced they were right and 10% less willing to do something to fix the situation.
Stanford University computer science Ph.D.
student Myra Chang says that companies and policymakers should work together to fix the problem.
especially the tough conversations.
For NPR News, I'm Ari Daniel.
But at the same time, I think maybe the biggest recommendation is to not use AI to substitute conversations that you would be having with other people.
Researchers found that numerous AI models affirmed someone's problematic or wrong behavior about half the time.
To consider the impact, they had 800 participants interact with an AI about a real conflict from their lives.
Those who'd interacted with an affirming AI were 25% more convinced they were right and 10% less willing to do something to fix the situation.
Stanford University computer science PhD student Myra Chang says that companies and policymakers should work together to fix the problem.
Especially the tough conversations.