Want to use AI to improve your work? Have it disagree with you - The Conversation

Knowledge workers such as scientists, teachers, lawyers and business leaders are increasingly using generative artificial intelligence applications for tasks that demand imagination, creativity and problem-solving. Ironically, perhaps, recent research shows that people who interact with generative AI for work tasks can end up not thinking critically. They can surrender to AI’s quick and confident answers, which can lead to speedy but low-quality solutions.

Scholars refer to this increasingly prevalent problem as cognitive offloading. And they are showing that it can erode the very skills that make knowledge workers valuable.

My research at the intersection of artificial intelligence and knowledge work provides a different view, however. In one study, my colleagues and I interviewed 45 knowledge workers at a U.S. public university. They all were using generative AI applications in their core tasks: generating new ideas, solving complex problems and designing novel solutions. We also analyzed the prompts and outputs from their iterative AI chats.

We found that when knowledge workers intentionally use AI to challenge their ideas – to generate friction – they can significantly improve their performance. Previous research on human-AI collaboration for loan evaluations had similar results.

How to create friction

Using AI to find information that contradicts your own can reveal unlikely perspectives that disrupt your familiar logic and spark new insights. This is particularly important because AI models quickly become familiar with how a user prompts the model and interprets its output, and the model presents information according to these patterns.

For example, a marketing professional in our study asked AI to create virtual customers with personas that would lead them to dislike a product under development, which was an accounting certificate program. The contrary feedback from the AI model revealed customer preferences that the accounting professional designing the program had not considered, such as using multiple cases from different industries to illustrate a principle, rather than cases from the same industry.

An attorney used AI to find obscure legal loopholes and to consider how companies might exploit them. That exposed surprising but realistic scenarios of hard-to-detect unethical behavior in employees.

Exploring unlikely and opposing ideas presented by AI can feel disorienting. To help, knowledge workers should lean on their own expertise to question surprising information that an AI system might provide.

An operations research scientist who works with supply chains described adapting unfamiliar computing programs from other fields suggested by AI, such as signal processing and wireless communication, which helped him create a stronger program of his own. Although the programs were new to him, his strong understanding of supply network modeling helped him see how he could apply them.

A study I was not involved with found that creative writers produce better copy when they use AI as a sounding board rather than to ghostwrite.

Knowledge workers can also use AI’s outputs to examine topics from multiple viewpoints, deepening their knowledge.

A research scholar in our study probed an AI system for journal papers that used a concept he was using, but in contrasting settings, which helped him understand the concept from different angles. A lawyer followed up on legal citations suggested by an AI model by asking for different types of illustrative cases. That helped her fully understand the original case and write a strong legal brief.

Helping knowledge workers help themselves

Leaders of organizations can encourage knowledge workers to develop friction-generating queries of AI models.

A straightforward approach is to use prompts that challenge rather than agree with a person’s thinking. People can also draw on their own knowledge, work experiences and professional ethics to scrutinize AI outputs. Sometimes people realize that a response lacks originality, even if it is factually correct.

Organization leaders can also create programs for knowledge workers to share their experiences and insights. Regular sessions – led by cross-functional AI task forces or AI champions – can also help knowledge workers keep up with rapidly evolving AI tools.

Providing resources for knowledge workers is important as well. Internal AI boot camps and workshops led by AI experts or experienced users can help workers understand what AI tools can and cannot do.

A professor leading an initiative about AI use in teaching described how she developed internal competitions with prizes for professors who successfully completed AI boot camps. Organizations can also offer AI sandboxes – safe, private environments where employees can experiment and learn how to use AI tools.

In these efforts to support knowledge workers, an important detail is pointing out that adding friction can appear to make an AI system work against you, but – as new research shows – it actually helps it work for you.