Overreliance on AI can Increase Cognitive Workload, Says Study

02.07.2023

Changing role of 'human in the loop'

A 2023 paper published in Management Science has found that while AI can improve the accuracy of human decision-making, it can actually increase the cognitive burden of the decision-maker. The paper, 'Human and Machine: The Impact of Machine Input on Decision Making Under Cognitive Limitations', is an investigation into the effect of machine-based predictions on the human decision-making process, as well as outcomes for the 'human in the loop'. Using a machine learning model, the paper's authors compared human decisions to machine-supported choices, identifying differences in tendencies, accuracy and the amount of cognitive effort required by humans.

The researchers say that incorporating AI can generate an increased workload during instances of cognitive restrictions, such as time pressure or the need to multitask. Despite this, organisations are increasingly using AI in the hope that they can alleviate workers' burdens or spend less money on payroll. Further, while AI virtually always improves the overall accuracy of choices, it can make certain mistakes more frequent, such as false positives. For example, in the case of a medical diagnosis, the findings of the paper indicate that under most circumstances, the use of AI will improve overall accuracy and reduce the risk of misdiagnosis. Yet, if the disease is rare and a medical practitioner is working under time constraints, using AI would result in a higher incidence of misdiagnosis and require more cognitive effort on the part of the human as a result of the need to resolve the 'ambiguity' or inaccuracy that using AI may result in.

The timely research comes amid questions regarding the role of AI in the workplace, widely believed to spell the end of a host of jobs performed by humans. The findings of the paper imply that attempting to streamline processes or downsize an employee population through the use of AI for certain tasks can fail, generating a need for more cognitive effort, not less. This being said, the researchers found there is value in using AI to support humans in certain tasks. The researchers believe that AI is best used to complement human tasks and decision-making, rather than to replace them outright. While AI can generate very accurate outputs, humans must still exert cognitive effort in assessing such outputs and comparing them to their own human conclusions before making a choice. The level of effort increases when humans must make decisions under pressure.

According to the researchers, machines are good at performing tasks with accuracy as a result of their computing power. Meanwhile, humans are flexible and adaptive decision-makers, but are constrained by our cognitive capacity. According to the researchers, this indicates that AI should be used to complement – not replace – human decision-making, and humans must learn when incorporating AI is useful versus when it is inefficient.

The findings of the researchers are relevant to Wittenborg, which is taking action to educate students as well as staff on the merits and drawbacks of AI in the workplace. “Many people have this perception that AI will replace humans,” says Hanna Abdelwahab, Wittenborg's Education Support Administrator. “The strength of AI lies in its speed, accuracy and consistency – dimensions which cannot be emulated by humans. However, the weaknesses of AI are soft skills such as creativity, innovation, collaboration and leadership,” she explains. “These are skills which human workers can complement. Until and when AI has mastered these soft skills, humans will always need to put in that extra cognitive effort to evaluate the information churned out by AI.”

Meanwhile, Zijian Wang, Senior Lecturer of Applied Sciences at Wittenborg, offers a slightly different approach. “Soft skills are for sure what AI could be lacking, but also in certain cases not relevant anymore,” he says. “For example, a tax advisor could be replaced, regardless of their soft skills, by a tax robot programme that is integrated into the accounting system and connected with the tax authority,” he explains.

“To further define the possible relations between humans and machines, or specifically AI and jobs, there are three types: AI-leading, AI-assisting, and AI-replacing. Once we understand if a certain job belongs to one of the three types, then we can choose the 'right' skills to learn.” Wang continues: “Learning the 'wrong' hard skills might be worse than not having sufficient soft skills. Hard skills in the fields of data, cloud, cyber security will be relevant for years to come.” He says that, considering the rapid pace of development in the AI world, lifelong learning will become the norm within the professional world. “AI will be everywhere, and it is probably everywhere already, because it is such a container word that can refer to the applications such as search engines, social media, servant robots and so many more.”