AI Tools Perform Better When You’re Nice to Them, Research Shows

Recent research by Google engineers discovered that the “nicer” you are to Artificial Intelligence (AI) systems, and the more you treat them like you would a real teammate, the better they perform on the tasks you ask them to do.

Beyond the implications of this in our everyday interactions with AI, how might we use these learnings to collaborate better not just with our AI tools, but also with our colleagues? Put differently: what can AI teach us about ourselves?

The Power of Kindness in Boosting AI Performance

Google’s engineers set out to find optimised prompts to help Google’s Palm Model (similar to ChatGPT) accurately execute tasks, a bit like writing the most optimal clues for players when designing a crossword puzzle. They found some surprising results, instead: the positive impact of kindness has on AI’s performance.

When phrases like “take a deep breath” and “go through this step by step” were used to encourage the AI to solve math equations, it demonstrated an astonishing 80% accuracy in its answers, compared to a mere 34% accuracy achieved with less supportive language. This research reaffirms that AI systems, while not sentient, perhaps benefit from the same kind of encouragement and guidance that humans do.

How does this work (given AI surely don’t have feelings)? While it’s not possible for an AI to literally take a breath, it is possible to make their jobs easier. AI systems do not reason like humans, but instead use Large Language Models (LLMs), a vast dataset of words, to produce results. Prompts are the key to helping LLMs navigate these datasets to improve their problem-solving (you may have heard roles like “Prompt Engineer” surface at organisations in recent years for this very reason…)

Google’s engineers suspect that AI responds more accurately to “kind” prompts because the datasets they drew data from may have included phrases like “let’s take a deep breath” or “think step by step” before showing more carefully reasoned solutions. Adding encourage phrasing to prompts therefore allows AI to better navigate vast datasets, and as a result, produce a more accurate answer.

This tells us two things. First, we may hypothesise that encouragement and kindness are highly effective to include into learning design, generally, because they are closely correlated with more careful reasoning in the LLMs AI (or at least Google’s AI) are programmed on. Second, that kindness may be a more foundational behavioural norm then we may give it credit for. It is worth exploring this second one, in particular.

Applying Lessons Beyond AI: Building Positive Work Cultures

The lessons learned from AI’s response to kindness extend beyond the realm of technology; studies like Google’s demonstrate how kindness is an important tool in collaboration of all kinds and is increasingly becoming a cultural expectation and norm, not just in AI, but also in our interactions at work.

There’s a reason here at TST we list Kindness as one of our values. Gallup’s extensive research into employee wellbeing concludes the same results year-on-year: incorporating kindness into your leadership and HR practices helps reduce employee burnout and absenteeism, and helps improve employee wellbeing.

Much like the AI tools researched by Google, receiving words of reassurance can help your colleagues feel more fulfilled, boost their self-esteem, and improve their confidence in navigating difficult tasks. Here’s how you might apply these lessons to your everyday interactions at work:

1. Positive Affirmations: Just as AI systems thrive on positive affirmations, workplace colleagues and team members appreciate recognition and encouragement. Recognise and appreciate the efforts of your team members regularly. Research has found that a simple “I appreciate your inputs” can go a long way in boosting morale and performance, particularly in today’s burnout culture.

2. Support and Guidance: Providing support and guidance in a constructive and friendly manner can make a significant difference in your team’s performance. Encourage your team to “take a deep breath” and approach challenges “step by step,” just as you would with AI, and watch their problem-solving skills flourish.

3. Constructive Feedback and Continuous Learning: Just as AI models improve with guidance, your team members can excel when they receive constructive input and have opportunities for growth. Encourage a culture of constructive feedback and continuous learning. Create an environment where feedback is seen as a means of improvement rather than criticism. This approach will help your team thrive, adapt, and innovate.

AI as a Reflection of Ourselves

As we strive for better interactions with AI tools, we can also take these lessons into our everyday interactions with our colleagues. So, the next time you notice a colleague struggling with a problem, try to foster a better working environment by encouraging them to “take a deep breath” and “work on it step by step.” If it works for AI, it might just work for us too!