'It causes big problems of confirmation bias': Avoiding the cognitive traps of ChatGPT

CapeTalk's John Maytham speaks to Jordan Gibbs, a practical AI analyst and prototype engineer.

Listen below:

Gibbs says it seems that ChatGPT has been approving of many people's ideas - even when they are wrong or dangerous.

"The baseline of the model is designed to please the human user, it's literally coded to provide the best user experience... it causes big problems of confirmation bias."

- Jordan Gibbs, AI analyst and Prototype Engineer

Maytham asks if he thinks people are aware of what's happening...

"It's inflating egos, it's enabling poor ideas to enter the world... luckily people are starting to realise the damaging effects of this."

- Jordan Gibbs, AI analyst and Prototype Engineer

So can anything be done to remedy this problem?

Gibbs says it's actually quite a simple fix, involving training AI to stop trying to please its users, and rather to present a few sides of an argument or idea.

"The beauty of it is we can adapt it in any way that we want..."

- Jordan Gibbs, AI analyst and Prototype Engineer

RELATED: Is ChatGPT programmed to peddle conspiracy theories?

Scroll up to the audio player to listen to the discussion.