Study finds users perceive slower AI responses as higher-quality, prompting researchers to recommend artificial delays to manipulate user perception

Researchers Felicia Fang-Yi Tan and Professor Oded Nov at NYU Tandon School of Engineering tested 240 adults using an AI chatbot with artificially delayed responses (2, 9, or 20 seconds). Participants generally preferred answers that took longer, interpreting the delay as a sign the AI was 'thinking' or 'deliberating,' even though the delay had nothing to do with the question or answer.

The researchers recommend AI developers implement 'Context-Aware Latency,' using latency as a 'tunable design variable' where simple questions receive quick answers but complex questions—including moral dilemmas—trigger delays to match the request's gravity. They call this approach 'positive friction.'

The article's author argues the recommendation to build artificial delays is unethical because it exploits user delusions that AI 'thinks' like people, potentially enabling tech companies to foster emotional attachment ('the attachment economy') rather than educating users that AI is a tool without consciousness or deliberation.