Trust in AI: A Human-Centric Approach to Maximize Performance

Implementing AI in your business is not merely a technical issue; it's a fundamental shift in your business processes, affecting your employees' daily work life. For business leaders, innovators, and decision-makers, building and deploying trustworthy AI systems is crucial. Properly calibrated trust in AI enhances decision-making, boosts product adoption, streamlines operations, and opens new avenues for innovation, ultimately driving business growth.

The benefits of AI in work environments, such as enhanced productivity, improved customer service, and increased innovation, are well-known. AI automates repetitive tasks, enabling employees to focus on complex work, and AI-driven analytics support data-driven decisions. Despite these advantages, the risks of human-AI interaction are less discussed due to their complexity. However, simple rules for implementation and comprehensive employee training can mitigate these risks. 

When discussing the topic of AI implementation for productivity with other people in the industry, I notice that we often speak about its reliability or accuracy. I repeatedly stumble about a false belief: that the implementation of superior AI automatically translates into better performance. This assumption overlooks a critical factor: the users’ trust in AI. 

Research reveals that human-AI collaboration can dramatically diminish the performance of a user if the collaboration fails. Correctly calibrated trust is essential for successful collaboration. AI that isn't trusted correctly not only struggles with adoption but can also lead to decreased performance due to improper use, because users might over- or under rely on AI’s recommendations. 

AI Adoption: Over-Reliance and Under-Utilization

When implementing AI in workplaces, wrongly calibrated trust can decrease performance in several ways. Too much trust may cause employees to overly rely on AI and overlook its errors, thus decreasing decision-making quality. Conversely, too little trust can result in underutilization of AI, where employees ignore accurate AI recommendations, missing out on efficiency gains.

Risk 1: Accepting false AI recommendations.

A commonly used LLM tool, ChatGPT, demonstrates that using AI isn't always straightforward. People often misuse ChatGPT for financial information or metrics that require structured data analysis. However, ChatGPT’s primary functions are content generation, restructuring, brainstorming, and summarizing language-based data. While it provides answers to all queries, the quality varies, leading to over-reliance on low-quality AI suggestions and diminished performance. This demonstrates that AI tools perform poorly outside their intended purpose and should only be used for well-defined tasks.

→ AI tools should be strictly used for their developers’ intended purpose.

An article by Ness found that junior developers experienced reduced productivity with Generative AI because they accepted AI-generated code without thorough evaluation, leading to issues. Task expertise is crucial for judging AI output quality and avoiding over-reliance on incorrect suggestions. Contrary to common belief, AI tools enhance existing expertise rather than elevating less skilled workers. They are not magic wands that make inexperienced workers perform above their levels.

→ AI tools should be used to support skilled workers who are able to judge the outcome.

Risk 2: Rejecting correct AI recommendations.

Unlike humans, AI cannot reciprocate trust, making the interaction fundamentally different. Trusting AI involves accepting certain risks and relying on its perceived reliability. AI's opacity, especially in machine learning systems, means users cannot see how decisions are made, leading to lower trust and adoption. This "black box" phenomenon causes users to question AI decisions and outcomes, resulting in decreased decision-making efficiency and performance due to under-reliance on AI.

→ Users should be educated about the functionality of an AI tool. 

Moreover, users tend to abandon AI recommendations after observing failures, unlike with human colleagues who are given chances to improve. Research suggests that once trust in AI is broken, it is incredibly hard to repair. This leads to too little trust in AI, resulting in missed gains in performance and efficiency.

→ Users’ expectations towards AI capabilities should be correctly set to avoid damages of trust.

Addressing the Risks: Strategies for Effective Trust Calibration

Correct calibration of trust in AI ensures users understand the AI's capabilities and limitations, fostering appropriate reliance on its outputs. This transparency helps mitigate the "black box" phenomenon, increasing confidence in AI decisions. Additionally, by educating users and managing their expectations, trust repair becomes more achievable, reducing the likelihood of complete abandonment after failures.

The human aspect: How can you calibrate your trust in AI? 

Trust calibration is an ongoing process that occurs throughout the user's interaction with the AI system. This continuous engagement helps build and maintain trust over time. Just as one would take the time to understand and build trust with a new team member through feedback and iteration, fostering a collaborative relationship with AI requires a thoughtful approach.

1. Understanding AI Purpose, Capabilities, and Limitations: The AI system should be viewed as a new team member. Similar to learning a colleague’s strengths and weaknesses, it is essential to spend time understanding how to interact with the AI system and what it is intended to do. This helps set realistic expectations and avoids over-reliance or undue skepticism.
2. Leveraging Complementary Strengths: Combining human expertise and intuition with the AI’s data processing power achieves the best outcomes, much like working with a team member who complements one's skills. AI should be used as a tool to support decisions, not replace them.
3. Maintaining a Critical Mindset: It is important to continue questioning or challenging AI outputs that don’t seem right. Just as one would critically assess a colleague’s suggestions, the same scrutiny should be applied to AI recommendations.

Implementing AI effectively requires more than just integrating technology; it necessitates a deep understanding of trust dynamics between users and AI systems. By correctly calibrating trust through continuous engagement, understanding AI capabilities, and fostering a collaborative relationship, businesses can unlock the full potential of AI. This human-centric approach not only enhances decision-making and boosts productivity but also ensures sustainable and innovative growth in the long run.

How can MVPF help? 

At MVP Factory, we’ve seen how trust - or the lack of it - can make or break the success of AI adoption. One project that comes to my mind is our work on WiseWater AI, where we helped Niterra Venture Lab design a platform for the water sector to predict and compare infrastructure investments. 

It wasn’t just about building a functional AI system; it was about making sure the people using it understood how it worked and could rely on it to make critical decisions.

Through workshops, collaboration, and a lot of listening, the team bridged the gap between technical complexity and human understanding and ensured the AI served its purpose, not just in terms of accuracy but also by fostering trust among its users. The result? A tool that empowered professionals to make smarter, more sustainable decisions—confidently.

Projects like this showcase that AI isn’t just about the algorithms; it’s about the people who use them. I believe that taking a human-first approach in every project, helps teams not only adopt AI but trust it enough to fully leverage its potential.

If there’s one thing I have learned, it’s that trust in AI isn’t built overnight - it’s built over time, through understanding, collaboration, and clear communication.

About the author: As a product manager, Hannah's journey in artificial intelligence (AI) has been both exhilarating and enlightening. Her deeper understanding of this field comes from extensive research on trust in AI, specifically comparing AI to human decision support agents and analyzing the impact of risk on trust dynamics.

Do you have thoughts about AI adoption or are curious about how MVP Factory can help with your AI journey? Feel free to reach out to Hannah directly or message her to access the full article.