Forum: Create conditions for staff to oversee AI decisions properly

At the DBS Asia Future Forward Conference, DBS chief executive Tan Su Shan stressed that businesses must look beyond simply adopting artificial intelligence, and rethink their workflows to achieve meaningful outcomes (Human judgment and intentional outcomes are vital in AI use: DBS CEO, Oct 12). TikTok chief executive Chew Shou Zi also highlighted the importance of human judgment in AI use.

These are important points. But can employees genuinely exercise their judgment when their employers also expect them to maximise speed, productivity and cost savings?

Imagine an employee using AI to assess customer applications. He is expected to check its recommendations, but is also measured on how many cases are completed and how quickly.

A careful review takes time. Questioning a recommendation may mean missing a target. Approving it and moving on may be the easiest way to meet expectations.

The employee is nominally in control. But is that control meaningful if exercising independent judgment comes at a professional cost?

This is the gap between human oversight in principle and human oversight in practice. When an AI-assisted decision goes wrong, it may be tempting to ask why the employee approved it. But we should also ask whether the employee had enough time, information, expertise and authority to challenge it.

Businesses that want employees to take responsibility for AI-assisted decisions must create the conditions for genuine oversight. Employees need time to check recommendations, training to recognise errors and the freedom to reject or escalate questionable decisions without being unfairly penalised for doing so.

Accountability must also extend to the organisations that choose these systems, determine how they are used and set the performance targets that shape employees’ behaviour.

As Singapore encourages wider AI adoption, businesses should look beyond whether AI makes work faster or cheaper. They should also consider whether it improves decisions in practice, and whether employees can challenge its recommendations when something appears wrong.

AI can help us work faster. But a faster process is not necessarily a better one. A human who is held responsible for an AI-assisted decision must have more than a button to approve it. That person must have the real ability to question it.

Ives Tay