Chatbots reward polite users with better answers

As artificial intelligence (AI) chatbots continue to become integral to newsrooms and media companies, a new debate has emerged regarding the nature of interactions with these digital entities. Recent research by Waseda University and the RIKEN Center sheds light on how courteous interactions could enhance AI performance and reduce inherent bias, challenging the notion that these systems are merely inanimate algorithms.

AI chatbots are increasingly used in newsrooms, performing a range of tasks including text generation, transcription, and summarisation. However, how one interacts with these systems might significantly impact their functionality. The research illustrates that using polite prompts can notably enhance the output of large language models like GPT-3.5, which showed a 15.6% improvement in language understanding benchmarks when polite language was utilised. Similarly, GPT-4 exhibited reduced bias with moderately polite prompts.

Beyond simply improving performance, polite prompts also influenced AI to draw from a more extensive array of linguistic sources, enhancing creativity. These benefits, however, appeared to vary across different languages, suggesting cultural influences on communication norms.

Robert Caulk, CEO of Emergent Methods, explained, “These AI models are sophisticated pattern recognition devices trained on a vast range of human interactions. The politeness of user input often leads to more thoughtful and respectful responses.”

Yet the notion of treating AI with respect is not universally accepted. Some experts and practitioners in fast-paced industries argue for efficiency over etiquette. The argument is that direct communication, without unnecessary verbosity, leads to faster results. Josh Brandau, CEO of NOTA, cautioned against anthropomorphising AI, explaining, “An AI doesn’t have feelings to hurt. Efficiency should be the priority.”

Despite differing opinions, digital content producers at TV stations like KOTV and KWTV treat their AI tools as useful equipment, respected but unemotional, a view echoed by Richard Cox, director of digital services at Griffin Media.

The research highlights that polite language can activate different aspects of AI’s learning, resulting in more thoughtful responses. Politeness may trigger patterns in the AI’s training data associated with comprehensive answers, mitigate noise by reducing irrelevant input, and help decrease the levels of bias in responses.

As AI continues to evolve, conversations around its treatment are surfacing, with historical references from pop culture cautioning against the mistreatment of non-human entities. Founder of The Media Copilot, Pete Pachal, noted that, while AI lacks feelings, human interactions with it might influence real-world behaviour, stating: “More importantly, humans have habits, and the more we interact with AI, the more our behaviours in that realm will leak into ‘meatspace.’”

Finding a digital middle ground appears essential, as excessive politeness can deteriorate AI performance. Instead, moderate politeness may be the optimal approach, fostering respectful interactions and preparing for future human-AI collaborations. Ashish Agrawal, founder of Eon Media, emphasised that polite interaction with AI is about optimising performance and setting precedents for AI integration into daily life.

Despite the increasing importance of courteous interactions, a 2024 survey by Talker Research revealed that opinions are divided, with 32% of Americans surveyed feeling that politeness towards AI is unnecessary and some even resorting to swearing during interactions. This suggests that while kindness might enhance effectiveness, the personal approach to AI remains diverse.

The implications of these findings are significant as they suggest current interactions will form the basis of future AI training, reinforcing the values of respectful communication for upcoming generations of AI systems. Whether one perceives AI as a tool or a potential collaborator, how we interact with these digital companions could influence the nature of future human-AI relationships.

Source: Noah Wire Services