Why making AI too easy to use is a risk for business
AI can boost efficiency at work, but it may also weaken the experimentation and critical thinking that drive innovation. Leaders should introduce strategic friction to help preserve judgement, learning and creativity.
Associate Professor of Strategy & Behavioural Science
Article at a glance
-
Leaders should design AI workflows to preserve the human ability to evaluate, challenge and build on AI-generated outputs, new research reveals
-
Deliberate friction, requiring human input and thinking, can help avoid an AI-driven degradation in critical thinking within organisations.
As the saying goes, necessity is the mother of invention: a lack of easy solutions forces us to come up with innovative ways to tackle complex problems. Artificial intelligence (AI) offers an alternative approach, providing instant solutions with little human effort required. While this can certainly increase productivity, it may also reduce necessity in the sense of needing to innovate, think creatively and come up with novel approaches to get things done.
Is this a problem? Perhaps not in the short term, as less work means lower costs and more resource capacity for businesses. But in the long term, it may pose an existential threat by atrophying the critical faculties we need to produce new thinking and develop innovative solutions. We call this the productivity trap.
looks at this from an organisational perspective. It shows that when good enough solutions become free for anyone in the organisation to access, facilitated by AI, the rate of reuse goes up, while the rate of independent exploration and experimentation goes down. Over time, this sees teams increasingly converge on a limited number of approaches.
Avoiding the productivity trap
This creates a significant tension for business leaders: how to balance the cost and efficiency benefits of AI with the erosion of creative thinking among individuals and teams.
One option is to simply limit or slow AI integration, but that comes with clear opportunity costs in short-term savings and long-term efficiency gains. At the same time, ignoring the dangers of the productivity trap could see an organisation lose its USP and cutting edge in the eyes of the market, or be caught off guard by an unexpected challenge.
The solution lies in what we call strategic friction adding a small hurdle to individual use of AI in order to raise the total level of knowledge at the organisational level. This works because when people are required to invest some effort in order to use ready-made findings, they naturally learn and produce new knowledge for others to build on.
As a result, more distinct approaches survive, fewer individuals give up on exploration and experimentation, and the organisation is less at risk of getting stuck with a limited range of solutions. This relies on a mechanism known as absorptive capacity, which is the ability to build expertise by stacking new learning onto existing knowledge. This helps people to grow understanding, allowing them to evaluate, adapt and improve ideas rather than just copying them.
"When people are required to invest some effort in order to use ready-made findings, they naturally learn and produce new knowledge for others to build on."
Friction with purpose
The key to using strategic friction is not to create bureaucratic hurdles for the sake of it. Instead, its about keeping enough friction in the system so that people need to understand ideas before reusing them, helping them to continue to learn and develop rather than stagnate.
Leaders should calibrate this carefully; too little friction can limit innovation, while too much might slow everything down. Equally, not every task requires friction; routine, simple tasks where speed is the goal will benefit less than those where the organisation needs staff to think, learn and develop.
At the organisational level, this means moving from a model where users ask AI questions and receive simple solutions, to one where users have to contribute their own understanding to receive useful outputs. To achieve this, leaders could consider requiring staff to submit documented independent work before being allowed to use AI for a task.
At the system level, AI platforms can be customised to ask for a user-generated input before providing an output. For example, an analyst asking AI for a risk assessment could first be required by the interface to upload their own hypothesis and list of variables. The AI would then build on this human input in a reciprocal, conversational process that sees the two parties collaborating, rather than the user simply delegating work to the AI.
Getting the right staff mindset in place can help too. Traits to look out for include an unwillingness to accept or rely on AI-generated work at face value. Instead, workers should take the time to see whats missing, test the assumptions and adapt it to fit the context. This requires the ability to spot errors, omissions and incorrect assumptions in AI outputs, as well as the ability to explain why these things are problems.
Building a workforce with these skills may see a shift from targeting those who can give the right answers to those who can identify the right processes. Leaders should be looking for builders rather than free-riders, choosing those who instinctively understand the limits of AI capability, question its outputs, and provide their own inputs to ensure the results are useful.
