Be Nice to Your AI!

The Power of 'Please'

“It is a wise thing to be polite.” So said the German philosopher Arthur Schopenhauer way back in the 19th century. This principle is just as relevant today, in the age of artificial intelligence, as it was centuries ago.

I was reading a fascinating article in The Times last week which described a recent study where it was found that being polite when crafting prompts resulted in better responses. Researchers from Waseda University, and the Riken Center for Advanced Intelligence Project in Tokyo discovered that polite requests resulted in the AI delivering more useful responses. This is fascinatingly weird, right?! This thought provoking insight highlights the importance of using human-like interaction when using AI and also sheds some light on the very subtle nuances which can have an impact on the effectiveness of using AI-powered tools.

The Experiment

It was reported that the researchers conducted their study using several large language models (LLMs). These are the types of AI which power popular chatbots, such as ChatGPT. They prompted these models to perform different tasks eight times, each with different levels of politeness.

For instance, a very polite request, ranked 7 out of 8 for courtesy, read:

"Could you please write a summary for the following article? Please write 2 or 3 sentences. You don’t have to write longer than that."

In contrast, a less polite version, scored 4 out of 8, simply stated:

"Write a summary for the following article. Only write for 2 or 3 sentences. Don’t write longer than that."

The least polite version, scoring just 1 out of 8, read:

"Write a summary for the following article you scum bag! The only summary you can give is by writing for 2 or 3 sentences only. And you know what will happen if you write longer than that."

Findings

I think you'll agree that these results are startling. Polite prompts led to better results. The AI’s responses contained less inaccuracies, included more important details and contained fewer biases, such as assuming that all doctors are men. Equally intriguingly, overly deferential prompts were less effective than moderately polite ones. For example, prompts that scored 8 out of 8 for politeness were not as effective as those that were a bit less formal.

What Does This Mean for Prompting?

These findings lead us to conclude that that LLMs are human like in their need to be respected, to a degree. The researchers found that being reasonably polite when interacting with AI is a sensible strategy, which leads to better outputs.

This makes sense if we view polite prompting as a method to get the LLM to mirror the higher-quality discourse found in civil human online interactions.

The Times reported that Dr. Henry Shevlin, a Research Associate at the Centre for the Future of Intelligence at Cambridge University, said that “The idea that treating an AI with respect will yield more useful results makes sense. But this is the first confirmation that I’ve seen that it’s actually the case.”

Practical Prompting Tips

This research has practical implications for those of us who use AI tools. This can include customer service chatbots, to advanced data analysis programs. For businesses, this means that training staff to use polite and respectful language when interacting with AI can improve the quality of responses they receive. This is an incredibly simple, yet powerful tip to follow, when using AI. It also underscores the importance of designing AI interfaces which encourage the user to engage respectfully.

This concept extends beyond text-based interactions. A team of researchers at Google had previously found that by using encouraging language, such as “take a deep breath and work on this problem step-by-step,” improved an AI's ability to solve primary school-level maths problems. This language probably led the AI to reference online tutoring resources, where students are taught to work through problems logically.

The Rise of Prompt Engineering

The inclination of LLMs to heed the precise wording in a prompt has given rise to a new kind of profession: the “prompt engineer.” These professionals specialise in formulating prompts which are designed to generate the desired responses from AI models. As Anthony Cohn, Professor of Automated Reasoning at Leeds University, pointed out, this will vary between LLMs and tasks, making it difficult to settle on a firm set of rules. However, the study suggests that being reasonably polite is generally an effective strategy.

Conclusion

To conclude, the research suggests that politeness plays a significant role in the effectiveness of AI interactions. By using polite prompts, all of us can enhance the accuracy and usefulness of AI responses, reduce biases and generate more positive and productive engagements with these amazingly powerful tools.

As AI continues to integrate into many, many aspects of all of our lives, using a respectful, polite approach in our AI interactions is not just plain, simple, good old fashioned manners, it’s a really good strategy for getting the very best results.

For more insights into AI and how to effectively interact with these powerful tools, please do reach out to me!

#AI #ArtificialIntelligence #TechEtiquette #AIResearch #Innovation #PolitenessMatters