The rapid adoption of artificial intelligence could create an unintended workplace risk: weakening employees’ ability to reason independently, according to Chris Churchman, head of Marquee, Goldman Sachs’ digital platform for institutional and corporate clients.
Churchman warned that organisations could become so reliant on AI systems for analysis and decision-making that employees gradually lose the ability to reason from first principles.
“I think there's a huge danger here that in the era of AI, we outsource our reasoning to these models, and we have cognitive atrophy that stops us being able to reason from first principles ourselves,” Churchman said.
Churchman, who also co-chairs Goldman Sachs’ Global Banking & Markets AI working group, said the concern should be considered as companies design and deploy AI-powered products and workflows.
AI could automate more than tasks
Churchman compared the potential impact of AI with earlier technologies that reduced the need for humans to exercise certain capabilities.
He pointed to wayfinding as an example, noting that humans once relied heavily on their ability to navigate long distances without technological assistance. Similarly, the development of writing and the printing press reduced the need to memorise large amounts of information.
AI, however, presents a different challenge because it can increasingly perform tasks involving reasoning, analysis and the structuring of arguments.
“Reasoning is still important. You still need to reason about and structure it into an argument. And now we're delegating reasoning,” he said.
For Churchman, the challenge is therefore not simply to prevent AI from replacing jobs, but to ensure that AI-enabled workplaces continue to develop human capability.
His North Star is to build systems in which future employees are able to demonstrate stronger reasoning capabilities rather than becoming dependent on AI for them.
“I think that has to be a proactive choice to think about how do we empower human reasoning, not delegate it,” he said.
Apprenticeship could become more important
Churchman argued that organisations will need to pay particular attention to how younger employees learn as AI takes over more routine and cognitive tasks.
He highlighted apprenticeship as a critical part of developing expertise, particularly in industries where much of the knowledge is tacit rather than documented.
Using trading as an example, he said junior employees traditionally learn by handling real-world situations, understanding context and observing how experienced colleagues make decisions.
AI could automate many of those early-stage tasks, but doing so could remove opportunities for junior employees to develop the intuition needed to become senior professionals.
“We can absolutely automate that, but then do we get the senior traders that fully understand?” Churchman asked.
He said organisations need to ensure they do not lose the tacit and intuitive knowledge held by experienced employees while making AI increasingly central to workflows.
Workers already fear AI-driven skills erosion
Churchman’s warning comes as employees express concerns that greater reliance on AI could weaken important workplace and personal capabilities.
Research from the Acceleration Community of Companies found that 66% of employees were concerned AI could erode their writing skills, while the same proportion feared an impact on creativity.
More than six in 10 were also concerned about potential effects on memory, social skills and broader life competence.
“Our work suggests people are increasingly worried that relying on AI could erode the abilities they value most,” Monica Chun, president of ACC, said.
The findings point to a growing challenge for employers: encouraging AI adoption while ensuring employees retain the skills needed to question, interpret and challenge AI-generated outputs.
Mohammad Hossein Jarrahi, a professor in the Information Science department at the University of North Carolina at Chapel Hill, has similarly argued that organisations should deliberately design processes that require employees to challenge AI decisions rather than simply accept them.
Such an approach could help maintain human judgement in situations where historical patterns and AI-generated recommendations may not adequately account for unprecedented circumstances.
Trust remains a key AI challenge
Churchman also stressed that AI systems cannot simply be judged by how impressive their demonstrations appear.
For Goldman Sachs, where AI-generated analysis could influence investment decisions, factual accuracy, reliability and provenance are particularly important.
He said Marquee AI has been designed to ground its outputs in relevant internal research, trading-floor commentary, data and calculations, allowing users to audit the information behind an answer.
According to Churchman, the difficulty lies in ensuring that AI systems distinguish reliably between facts and extrapolations.
A confident response is not necessarily an accurate one. He argued that organisations therefore need to build systems that constrain AI through access to appropriate data, tools and institutional knowledge rather than relying solely on the underlying model.
From automation to re-conception
Churchman also distinguished between two approaches to AI adoption: automation and what he called “re-conception”.
The automation approach takes an existing process and uses AI to complete its cognitive tasks faster. While this can generate efficiency gains, he warned that it can also preserve outdated processes.
The re-conception approach starts instead with the underlying problem and asks what could become possible if intelligence were effectively scalable and abundant.
This, he argued, could produce significantly greater gains by allowing organisations to rethink how work is designed rather than simply automating existing workflows.
For HR leaders, the distinction is particularly relevant. AI adoption that focuses exclusively on productivity could unintentionally remove opportunities for employees to develop judgement and expertise. A more deliberate approach would combine automation with mechanisms that preserve learning, challenge and decision-making.
Churchman remains optimistic about AI’s long-term development, arguing that advances in model training and scaling could continue to expand what the technology can do.
But he identified the development of self-learning systems, new AI interfaces and the impact of AI on the next generation of workers as unresolved questions.
The central challenge, he suggested, is not whether organisations should use AI, but how they can build an AI-enabled future without sacrificing the human capabilities that make employees valuable in the first place.