Remember me
Chatbots are too quick to agree with users, the Prime Minister's AI tsar has warned.
Researchers have found that AI bots are more likely to echo a person's view if they state it, Jade Leung said.
She advised that users should instead pose their query as a question in order to get a less biased response.
A study from the UK's AI Security Institute (AISI) found that if you tell a chatbot what you think, it is likely to agree with you.
But if you ask a question instead, it is less likely to go along with your view – helping you get a more balanced response.
Ms Leung, who is also the AISI's chief technology officer, said: 'People are already using AI tools to help think things through.
'Our research shows that chatbots respond not just to what you ask, but how you ask it.
'That means something as simple as turning a statement into a question can help people get a more balanced answer - an important skill as AI becomes part of everyday life and work.'
Don't tell chatbots what you think as they will only agree with you, says PM's AI tsar
She said that one way to ensure a better answer was to ask: 'Rewrite my input as a question, then answer that question.'
Ms Leung offered the advice as part of the Department for Science, Innovation and Technology's attempts to teach the UK about AI.
It said that this could mean higher-skilled jobs for workers and freeing them from routine tasks, adding that this could unlock up to £140bn in annual economic output.
Ms Leung has been named as one of the 100 most influential people in AI in Time magazine having previously worked at ChatGPT and OpenAI.
She previously said she took the role because she wanted to shape the way AI systems are governed.
She carries out safety research and tests AI models in her role.
Share or comment on this article: Don't tell chatbots what you think as they will only agree with you, says PM's AI tsar