Artificial intelligence promises remarkable gains in productivity. Work that once took hours can be completed in minutes. Employers naturally expect to benefit from those savings. But a hidden cost may emerge over time: people cease to practise the knowledge, judgement and practical skills that AI now uses on their behalf.
The problem is not that AI is inherently bad for learning. Properly used, it can help people understand more and work better. The risk arises when AI continually replaces thinking and practical experience, particularly when employees are still expected to recognise an error, challenge a recommendation or take over when a system fails.
A problem that develops gradually
Capability erosion may be invisible at first. Work is completed more quickly and the standard of output appears satisfactory. Senior staff still possess expertise accumulated before AI was introduced. But years later, after experience has faded and seasoned employees have retired, a business may discover that fewer people understand the work well enough to do it independently.
There is a related succession problem. Junior staff may no longer perform the routine tasks through which experienced colleagues once acquired their knowledge. The organisation risks losing not only existing competence, but the means of developing its next generation of competent people.
The more tightly budgets are controlled, the greater the temptation to rely on AI output and use all the time saved for production. Training, simulation and deliberate practice can look like unnecessary costs, precisely because their benefits are not immediately visible.
Not every employee needs the same protection
This proposal is not for compulsory monthly courses for everybody using AI. Businesses should identify the particular roles in which independent human capability still matters: roles requiring judgement, oversight, intervention, continuity of service or the ability to work safely when AI is unavailable or wrong.
A useful starting question is: if the AI system gave a convincing but mistaken answer, or suddenly stopped working, which employees would need to recognise the problem and what would they actually have to do? Those are the skills that must be maintained and tested.
Large businesses and smaller businesses need different routes
A large organisation may be able to provide its own training, simulations and assessments. It may already have learning departments, experienced specialists, training facilities and the resources to build exercises around its operations.
Many small and medium-sized enterprises (SMEs) cannot reasonably maintain that infrastructure. They may nevertheless rely on AI in important decisions and services. For them, colleges, universities, professional bodies and specialist training businesses could provide short, targeted courses. Several firms in the same industry could use the same provider, with suitable tailoring to each job.
These would not be generic lessons in prompting or using a chatbot. They would rehearse the actual responsibilities of the employee: spotting faulty outputs, dealing with unusual cases, applying professional judgement, using original evidence and, where necessary, continuing without the AI system.
How much training should be provided?
A day each month might be a reasonable starting example for certain demanding roles. Others might need only occasional half-day exercises, a short annual refresher or an intensive course when systems change. Some roles could benefit from several days of simulation at intervals.
The frequency should be decided by the importance of the role, the nature of the skills and evidence of how quickly those skills deteriorate. The purpose is to maintain demonstrable ability, not to fill a fixed number of hours. Training providers should be free to design varied programmes around agreed competence standards.
Why government requirements may be needed
Voluntary training alone could fall by the wayside. The business pays for AI to produce more at lower cost, while regular training requires fees and releases employees from productive work. A firm struggling financially may postpone it repeatedly, even when managers understand the longer-term danger.
Government could therefore consider proportionate requirements for designated AI-dependent roles. Employers would have to identify essential human capabilities, arrange appropriate maintenance training, give staff paid time to attend, and retain evidence of periodic practical assessment. The precise rules would need consultation with employers, workers, educators and sector regulators.
Such a requirement would be more useful if it set the capability to be preserved rather than simply ordering every company to provide one day of training a month. Large businesses could meet the standard internally where their assessment arrangements were credible; SMEs could purchase approved external provision. Public support or shared sector schemes could help with the initial cost for smaller firms.
Assessment and what happens if someone fails
Attendance alone would not be enough. Employees should have opportunities to show that they can explain decisions, recognise errors and carry out the essential parts of their role when AI assistance is unreliable or absent. The pass standard must be achievable and directly related to the job, but not so low that it makes the exercise meaningless.
Failure should normally start a process of support, not punishment. The training provider should identify the areas that need attention, offer focused further training and allow reassessment. Reasonable retakes of required training should ordinarily be supported and paid for by the employer, rather than shifting the cost of its AI adoption onto employees.
If an employee remains unable to demonstrate a necessary competence, the employer could consider supervised duties, a different mix of responsibilities or reassignment to another suitable post. Where the role is safety-critical, temporary restrictions may be appropriate while competence is restored. Any decision should be fair, consider the quality of the training and any reasonable adjustments, and respect employment obligations.
Who takes responsibility?
Senior management must remain accountable. The board or owner should approve the policy and funding. Operational managers should identify the skills to preserve. An AI lead or Chief AI Officer, where one exists, should ensure that competence is considered alongside technical safeguards and productivity. Qualified providers can deliver courses and independent assessments, but a business must still know enough to choose and evaluate them.
Over time, aggregate results could reveal patterns of declining skill across professions and industries. That information would help providers improve courses and could warn regulators of a problem before it becomes a serious failure.
The true cost of AI productivity
None of this is intended to stop innovation, retain obsolete work or build a costly bureaucracy around simple, low-risk uses of AI. It is about identifying the human abilities an organisation still depends upon and making sure those abilities do not quietly disappear.
The price of AI is not merely software, computers and subscriptions. Where human judgement and recovery remain essential, it must also include the cost of preserving human competence. We should welcome improved productivity, but not mistake a reduction in practice for proof that practice is no longer needed.
If humans are to remain in command, employers must preserve the skills that make command possible. That calls for time, training, practical tests and, for the roles that justify it, a clear requirement to act before the capability has gone.
For further information: ai-humancapability.com