AI and Human Expertise: AI Workforce Impact series | Part one of two

The hidden cost of AI productivity: Who verifies the work?

Authors

Sofia Kyriakopoulou

Chief Technology Data and AI Officer, SCOR

Cécile Deman-Enel

Chief Human Resources Officer, SCOR

Artificial intelligence is transforming knowledge-intensive work. It can process information at speed, generate outputs in seconds and allow professionals to accomplish more in less time.

Yet productivity tells only part of the story.

As AI systems become more capable, professionals may spend less time creating outputs and more time reviewing, validating and taking responsibility for AI-generated work. The nature of the workload changes. Production effort declines while the demands of oversight, judgement and accountability grow.

This dynamic is central to a research collaboration between the Chair of Technology and Innovation Management at ETH Zurich and SCOR. Drawing on the scientific literature and interviews with SCOR experts, researchers analysed what is known about how generative and agentic AI affect human cognition, professional judgement and expertise development. The resulting report, Integrating Artificial Intelligence with Human Cognition, brings behavioural and cognitive science into the discussion about AI adoption.

The verification burden

AI processes large volumes of information through pattern matching and statistical inference. Experienced professionals contribute contextual understanding, tacit knowledge, critical thinking and the ability to navigate ambiguity. These capabilities remain essential when an AI-generated output must be interpreted, challenged or applied in a complex business environment.

This creates a trade-off. AI can substantially reduce the effort required to produce an output, while increasing the attention required to determine whether that output is accurate, appropriate and safe to use.

We describe this as the verification burden.

The burden extends beyond checking facts. Professionals may need to:

• understand how an output was produced;
• identify missing or misleading context;
• resolve ambiguity and conflicting information;
• assess whether the recommendation is appropriate for the specific situation;
• recognise when escalation is necessary; and
• assume accountability for the final decision.

Verification is therefore a cognitively demanding form of work. It requires sustained attention, healthy scepticism, domain expertise and a willingness to challenge an output that may appear fluent and credible.

Productivity can rise alongside cognitive demand

This creates a counterintuitive trade-off for organisations. Employees may complete more work, while carrying greater responsibility for assessing its quality and consequences.

The most experienced professionals may feel this shift most acutely. Their expertise makes them the natural point of escalation for exceptions, ambiguities and high-consequence decisions. As AI increases the volume and pace of production, these professionals may face a growing stream of outputs requiring human scrutiny.

Productivity metrics may therefore show a positive result while overlooking changes in attention, cognitive demand and accountability.

This matters because the sustainability of AI-enabled productivity depends on the human capacity required to support it. An initiative that accelerates production while creating excessive verification demands may shift pressure within the organisation rather than remove it.

A broader lens for AI performance

Building on the review, SCOR considers every AI initiative across three dimensions:

These dimensions form part of a wider set of choices about human judgement, accountability, learning and the design of work. Together, they will shape the capabilities organisations retain and strengthen over time.

Designing oversight deliberately

This wider perspective changes the questions organisations need to ask.

Who should review an AI-generated output? At what point should that review happen? When should a professional form an independent view before seeing the AI’s recommendation? Which exceptions require escalation? Where does final accountability sit?

These choices influence the quality of decisions and the cognitive demands placed on employees. They therefore need to be treated as core elements of AI design.

The objective is purposeful oversight: matching the level and form of human engagement to the cognitive demands, uncertainty and consequences of the task.

For leaders, this means assessing AI performance through the complete human-AI system. Output speed matters. The quality and sustainability of the surrounding decision process matter just as much.

SCOR’s approach: AI as a force multiplier

This perspective is central to SCOR’s AI strategy. As we elevate our ambition, we will continue exploring how AI can strengthen our business, support our employees and create the conditions for expertise to grow.

SCOR's work on agentic AI in P&C claims provided illustrative examples for the review, which also drew on interviews with two SCOR experts. The review points to a clear principle: the design of the interaction between people and AI is as consequential as the capability of the technology itself.

As advanced AI becomes widely accessible, sustainable advantage will increasingly come from the quality of the decisions organisations make around it and from the expertise of the people applying it.

Based on a research review by, and discussions with, Stefano Brusoni, Daniella Laureiro-Martinez and Tanja Müller (ETH Zurich)