AI at work needs organised challenge
A system can plan schedules, rank job applications or summarise performance. It can therefore also influence income, opportunities and workload. This opinion does not call for a general ban on AI. It argues that organisations should decide in advance how employees can have incorrect data, an unreasonable standard or missing context investigated. A human approval button alone is not always sufficient.
This is an editorial opinion piece. The conclusion is a reasoned position and not personal legal advice.
A score can hide assumptions
A productivity score looks precise, but depends on the chosen measurement method. The number of cases handled, for example, says little without considering difficulty and quality. An algorithm can apply these assumptions at scale without employees seeing the chosen standard. The Dutch Data Protection Authority points to risks for autonomy, workload and fair relationships in algorithmic management. Our editorial conclusion is that an organisation must explain the yardstick before using a score in discussions about performance, pay or availability for work.
Human assessment needs time and authority
A manager who simply confirms a system’s recommendation provides little substantive counterweight. Real assessment requires access to relevant data, understanding of limitations and authority to depart from the recommendation. That must also be practicable within the timetable. We believe organisations should record who explains a decision and who can implement a correction. Employees should not get lost between supplier, HR and manager because everyone says the outcome was determined elsewhere. Responsibility for an employment decision cannot disappear into the technical supply chain.
Consent does not resolve the power imbalance
Employee monitoring is subject to privacy requirements, including a valid legal basis, necessity and information. The Data Protection Authority stresses that consent may not be freely given in an employment relationship. A pop-up saying “I agree” is therefore not a convincing answer to every question about monitoring. Discuss the purpose of processing, whether a less intrusive method is possible and which data are actually needed. These are existing privacy questions that also matter when a supplier presents the system as an AI product. A new technical name does not automatically change the nature of data processing.
Employee representation must understand the application
Rules for employee monitoring may require the works council’s prior consent. Employees and their representatives also need substantive information on operation and consequences. A demonstration with an attractive dashboard is insufficient. Ask what data the system uses, how errors are discovered, which groups might be affected differently and how long data are retained. We argue for involvement before a working method is fixed. A discussion afterwards, when assessments have already been influenced, leaves less room to improve the process without personal harm.
A correction route must reach the real error
Employees need more than the ability to correct a typo in their profile: they must also be able to show that the chosen comparison creates a false picture. Consider caring responsibilities within a team that are not recorded as output, an adapted workstation or a schedule that insufficiently recognises agreed availability. A good route examines both data and interpretation. Specify who responds, within what time and what happens meanwhile to the disputed decision. This is our design proposal for fair decision-making; the exact legal obligations must be established for each application.
Start with a limited, verifiable use
Organisations can first identify the problem they want to solve and how they will measure success and errors. A limited trial with clearly defined consequences allows learning without immediately building every employment decision around a score. Preserve the option to return to another method. Ask employees what context is missing and check whether corrections are actually implemented. These recommendations are editorial choices, not a statement that every AI application follows the same statutory procedure. The difference between a writing assistant and a system that ranks people particularly deserves a specific assessment.
Counterargument: people are biased too, and systems can be fairer
That is a strong argument for seriously investigating the use of technology. A system can support consistency and make certain errors more visible. But possible improvement is not proof that a particular application already works well. Compare against a clear starting point and examine quality, consequences and whether errors can be remedied. Challenge also helps a good system be used better. It prevents human biases from being silently incorporated into data or design and then treated as an objective outcome. The aim is demonstrably better decision-making, with visible responsibility when it falls short.
When algorithmic planning or assessment is used, ask which data and standards are involved, who can explain the decision and how an error is corrected. Organise that discussion before the outcome affects your pay or role. Our editorial standard is that technology contributes convincingly to good work only when people retain a meaningful voice in decisions that shape their work.
Sources and editorial notes
These sources support the factual context. The position, assessment and conclusion are those of our editorial team.
Sources current as of:
- Publisher: Autoriteit PersoonsgegevensDutch Data Protection Authority: risks of workplace algorithmsautoriteitpersoonsgegevens.nlhttps://autoriteitpersoonsgegevens.nl/themas/algoritmes-ai/risicos-algoritmes-ai-ontwikkelingen-in-nederland/risicos-algoritmes-op-de-werkvloer
- Publisher: Autoriteit PersoonsgegevensDutch Data Protection Authority: conditions for monitoring employeesautoriteitpersoonsgegevens.nlhttps://autoriteitpersoonsgegevens.nl/themas/werk-en-uitkering/controle-van-werknemers/voorwaarden-voor-controle-werknemers
- Publisher: Sociaal-Economische Raad (SER)Social and Economic Council: technology and the works council’s roleser.nlhttps://www.ser.nl/-/media/ser/downloads/overige-publicaties/2018/technologische-ontwikkelingen-or.pdf
The source publications are in Dutch. All language editions cover Dutch law.
General information. Not a substitute for personal legal advice. This language edition contains draft translations based on official Dutch sources. Independent final legal review and native-language editing have not yet been recorded.