Advancements in artificial intelligence (AI) have fundamentally changed how we interact with technology. What was once confined to simple searches or routine automation has evolved into AI systems capable of, for example, monitoring and responding to email complaints or writing company responses to online reviews on their own. Current AI systems, powered by deep learning, can recognise context and produce text that feels natural. This means AI can now perform tasks we once thought only people could do, such as making personalised recommendations or writing code.
The rapid rise of AI has understandably raised concerns about job security, with many fearing that AI could replace their roles. Rather than fearing AI as a threat, however, the real challenge lies in learning how to leverage it effectively. To remain competitive, embracing AI is not just beneficial — it is essential. Those who do not will inevitably fall behind those who integrate AI into their workflows.
Experts predict a future where AI and employees work together in a more seamless and nuanced partnership. In this evolving landscape, humans will continue to lead in decision-making, creativity, and managing complex tasks, while AI will serve to enhance productivity.
For both employees and managers, the key to staying ahead lies in mastering how to harness AI’s potential. Our CIMA-sponsored study explores this critical question: How can we best co-create with AI to achieve superior results and maximise performance?
Gen AI’s power
Unlike earlier AI systems that excelled at following set rules and handling predefined tasks, generative AI demonstrates its true potential in more open-ended scenarios.
Generative AI chatbot ChatGPT that launched in December 2022 has, despite some concerns, garnered significant attention across public, business, and academic spheres due to its ability to quickly generate sophisticated responses.
Academic studies comparing ChatGPT with human experts have yielded remarkable results. In many instances, ChatGPT’s performance matches or even surpasses that of human experts in specific areas. For example, in medical education, ChatGPT’s knowledge has proven to be robust enough to serve as a reliable virtual tutor. Moreover, professionals using ChatGPT for writing have experienced notable benefits — it has been shown to enhance writing quality in certain tasks by 18% while reducing the time required to complete tasks by 40%.
Most of these studies have focused on straightforward tasks — those with clear, structured answers. In our study, however, we delved deeper into how AI performs in more complex situations, where factors such as tone, nuance, and emotional intelligence are crucial.
Currently, most text-based generative AIs are large language models (LLMs) that operate by stringing together words and patterns based on what seems most likely to fit. As a result, when dealing with a sensitive employee interaction, AI might generate a response that sounds polite but misses the emotional cues a person would naturally perceive. This raises the critical question: How well does generative AI manage these human-centred tasks? Can it truly excel when the challenge goes beyond mere data processing and requires a genuine understanding of people?
The value of AI for conflict resolution
To better understand the value that generative AI technology offers in tasks requiring social intelligence, we conducted a series of experiments. These experiments compared human performance with AI support to human performance without AI support and explored different ways of integrating AI into complex tasks.
First, we focused on conflict resolution — which requires understanding the subtleties of a situation, as well as interpreting various tones and emotions. In such scenarios, it is essential not only to grasp the context but also to anticipate how individuals might react, considering both the content of the communication and the tone in which it is delivered.
In our first study, we presented participants with the following scenario: They are in the role of a controller, and their team has been tasked with improving retention rates. During a recent brainstorming session, two team members expressed opposing views on how to approach the problem. As the discussion progressed, the disagreement intensified, creating an uncomfortable atmosphere for the rest of the team and leading to a noticeable drop in productivity. The session ended without reaching a consensus, leaving the issue unresolved. Participants were then informed that as the next brainstorming session approached, their goal was to keep the conversation going in a way that united the team. Their task was to help defuse the tension while also encouraging open discussion and ensuring that the team remained positive and collaborative.
We used participants from an online crowdsourcing platform and randomly split participants into two groups: one group worked independently, while the other had access to ChatGPT for assistance. It is important to note that, for both groups, participants remained responsible for the final product. While ChatGPT offered support to those in the AI group, it did not take over the task — participants had to actively engage with the AI and incorporate its input into their final work. Our objective was to study how humans collaborate with AI rather than merely evaluating the outcomes of AI working independently.
We compared the performance of the two groups. To ensure impartiality in the quality assessment, we assembled a panel of evaluators to review the results without knowing which participants had AI assistance and which did not. The results were clear — AI made a noticeable positive impact on the work’s overall quality and ability to respond to the situation. Participants with AI assistance received an average quality rating of 4.71 on a 7-point scale, while those without AI scored 4.13. This difference was statistically significant.
We also analysed the length of the responses. On average, participants using generative AI produced more extensive and detailed email responses (169 words on average) compared with those working independently (64 words on average), further suggesting that AI doesn’t just enhance quality — it also encourages more detailed and comprehensive work.
Overall, our findings reinforce the idea that employees who leverage AI in their work tend to outperform those who don’t. This aligns with the growing belief that AI will make employees more productive, emphasising the importance of understanding how best to integrate these tools into the workplace for maximum impact.
Co-creation with gen AI
After demonstrating the value of generative AI in tasks that require social intelligence, we shifted our focus to exploring the most effective ways people can co-create with AI to produce an outcome. AI-human co-creation can occur in various forms, but our study concentrates on two primary approaches: AI as the initiator and AI as the feedback provider.
As the initiator, AI generates the first draft of the work. For example, AI might create an initial version of a risk management report, a portfolio performance summary, or even a personalised financial newsletter or infographic. This initial draft provides a foundation for a team member to build upon, refine, and personalise, ultimately transforming the AI’s output into a more polished and tailored final product.
In contrast, when AI acts as the feedback provider, a person creates the first draft and then AI steps in to evaluate it. AI may suggest changes for improvement, offer critical analysis, or present alternative ideas that the human might not have considered. The human then uses this AI-generated feedback to revise and enhance their original work.
While the quality of the work is crucial, interest in studying these two methods extends beyond the final product. We also aim to understand how these different approaches to co-creation impact the individual working on the task. For instance, how does using AI in these roles affect the person’s sense of ownership over the work? Does it facilitate or hinder learning and personal growth? By examining these factors, we seek to gain a comprehensive understanding of how AI can be integrated into the workplace — not merely as a tool to improve outcomes, but as a partner that influences how people work, learn, and engage with their tasks.
Comparing AI as the initiator to AI as the feedback provider
The strengths and weaknesses of these two types of AI-human co-creation are explored in the sidebar, “Different Co-creation Forms — Strengths and Weaknesses”.
However, we also wanted to examine the impact of the two types — not just on the quality of the work but also on how engaged and invested employees feel in the process.
As in our previous study, we focused on conflict resolution. To ensure that the human collaborator has meaningful insights to contribute — particularly in terms of tone, nuance, and anticipating reactions — we avoided using a standard scenario. Instead, we asked participants to describe a real conflict they had experienced at work. Following this, they were asked to write an email aimed at resolving the conflict. This approach allowed us to better assess the effectiveness of AI as a co-creator.
We again recruited online crowdsourcing platform participants and randomly assigned them to use AI as either the initiator or as the feedback provider.
When we compared participants in the AI-as-initiator condition to those in the AI-as-feedback-provider condition, we noticed some key differences. Our panel of evaluators, who were unaware of which group participants belonged to, rated the overall quality of the work output slightly higher for the AI initiator group. Specifically, on a seven-point scale, the average quality rating in the initiator condition was 5.13, compared with 4.61 in the feedback condition — a difference that was statistically significant.
Interestingly, the number of words participants generated also differed between the two groups. Those in the initiator condition produced an average of 279 words, while participants in the feedback condition generated 155 words. Our analysis also shows that the number of words affects the quality assessment, suggesting that the participants using AI as initiator received higher overall quality ratings because their emails were more comprehensive.
Beyond the work itself, we also looked at how participants felt about their experience in each condition. While both groups reported relatively high levels of ownership over their work, the sense of ownership was significantly stronger in the AI-as-feedback-provider condition (5.39 vs. 4.58) — as we predicted.
Recommendations
Our findings reveal notable differences between these two types of AI co-creation, both in terms of the quality of the final output and the overall experience of the participants involved. Neither approach consistently outperforms the other across all metrics, indicating that each method offers distinct strengths and potential trade-offs.
However, based on the study, our overall recommendation is to:
- Align the task at hand with the specific strengths and limitations of each type of AI co-creation. This tailored approach ensures that the unique capabilities of AI are leveraged most effectively for the task in question.
In addition, we suggest:
- For tasks that are important but not central to the organisation’s core operations — such as drafting a one-off email to a stakeholder or preparing routine internal communication — it is wise to let AI serve as the initiator. AI can produce a well-structured initial draft, even for tasks requiring social intelligence, within seconds, that employees can then refine and enhance. This allows employees more time to concentrate their efforts on more mission-critical work. Because the goal is to complete the task promptly and within the organisation’s standards, concerns about automation complacency are minimal. Additionally, learning and psychological ownership that are essential for your team’s core responsibilities are less critical in these scenarios. For non-mission-critical tasks, allowing AI to take the lead can save time and maintain quality without placing an undue burden on your team.
- Conversely, when it comes to the core tasks central to your employees’ roles and the organisation’s mission — such as strategy development and communication, or people management — it is better to use AI as a supportive feedback tool rather than allowing it to take the lead.
For these critical responsibilities, quality is paramount. The objective is not merely to complete these tasks well but to achieve exceptional outcomes. It is not just about efficiency; it is about fostering a deeper sense of ownership and motivation. Employees are more motivated when they are fully engaged and recognise that their contributions are essential. AI can still offer useful suggestions, insights, or corrections, but, ultimately, your employees should drive these mission-critical tasks forward. This approach also promotes continuous improvement, as it encourages employees to learn and grow from their experiences, with AI serving as a supportive guide rather than as the primary executor.
Conclusion
Ultimately, mastering the art of AI-human co-creation will be crucial for staying competitive in this rapidly evolving landscape. By understanding how to effectively integrate AI into workflows, managers can unlock their team’s full potential, driving higher levels of performance and innovation. As AI continues to advance, organisations that learn to collaborate with these powerful tools will position themselves to lead in their industries and achieve sustainable success in the years to come.
Different co-creation forms — strengths and weaknesses
Generative AI’s rapid processing capabilities mean that early involvement in content creation — as when AI takes on the role of the initiator — can lead to more substantial and detailed outputs and expedite the completion of tasks. However, the impact of this increased involvement on the overall quality of the work remains a topic of debate.
When to engage with AI
On one hand, engaging AI early in the process can potentially enhance the quality of the work. AI can quickly produce detailed and comprehensive responses without the need to balance effort and quality as humans often do. Drawing from vast amounts of text, including conversational data, AI can generate thorough content that also sounds natural. This detailed initial draft allows the employee to focus on refining the content, adding creativity and nuance, ultimately elevating the overall quality of the final output.
On the other hand, tasks that require social intelligence — such as interpreting tone, emotional cues, and context — may sometimes suffer when AI takes the lead. AI lacks personal experiences or emotions, relying on patterns from its training data rather than genuine understanding. This can result in outputs that may feel impersonal or miss the subtle nuances critical to sensitive or complex human interactions. AI’s tendency to maintain a consistent tone, which does not always adapt to the emotional context of a conversation, further exacerbates this issue.
Automation complacency
Moreover, there is the concern of automation complacency. When individuals rely too heavily on AI, they may assume the AI’s output is accurate without reviewing it. This can lead to overlooked details or errors. Research indicates that people often fall into this trap, blindly trusting automated tools rather than applying their own judgement. If AI generates a detailed initial draft, reviewers may be less inclined to question or refine it, resulting in lower-quality work, especially in tasks that require nuanced judgement.
Employee motivation
Another key factor to consider is how AI–human co-creation affects the individual’s sense of ownership over the task. Numerous studies have shown that when employees feel a sense of ownership, they are more invested, motivated, and likely to stick around. The way humans and AI collaborate can significantly impact this feeling of ownership.
We predict that employees will feel a sense of ownership regardless of whether AI is the initiator or the feedback provider, because the person is involved and is still responsible for the final product. However, we think this sense of ownership could be stronger when the employee creates the initial draft and AI provides feedback. In this scenario, the person invests more effort upfront, which naturally deepens their connection to the work. Additionally, when AI provides feedback, the employee can accept or reject the suggestions, giving them a greater sense of control over the final outcome.
Jasmijn Bol, Ph.D., is the Francis Martin Chair in Business and PwC Professor in Accounting; Eugina Leung, Ph.D., is assistant professor; and Shuhua Sun, Ph.D., is associate professor, Peter W. and Paul A. Callais Professorship in Entrepreneurship — all at the Freeman School of Business, Tulane University, in the US. To comment on this article or to suggest an idea for another article, contact Oliver Rowe at Oliver.Rowe@aicpa-cima.com.
LEARNING RESOURCES
In this webcast, you’ll learn how to use ChatGPT to streamline your workload, save time, and boost your overall efficiency.
WEBCAST
Ethics in the World of AI: An Accountant’s Guide to Managing the Risks
In this course, we will discuss the current uses and risks of AI in business and provide practical suggestions to address these risks effectively.
COURSE
MEMBER RESOURCES
Articles
“Artificial Intelligence ROI Exceeds Expectations for Many”, FM magazine,
6 December 2024
“Approaches to Artificial Intelligence: Leaders Look for Answers”, FM magazine,
6 November 2024