In my previous post, I argued that ISO 10667 can provide an important framework for maintaining quality in assessment as AI becomes increasingly integrated into recruitment and selection.
I think there may be an even broader role for the standard as the EU AI Act is implemented. Under the EU AI Act, the use of AI tools in selection is classified as a high-risk area.
It is repeatedly suggested that AI tools may potentially be involved throughout the recruitment and selection process: in the job analysis and the specification of requirements, in the sourcing of candidates, but also in the selection, as when screening candidate CVs, in assessments including reference checking. Such tools may also support integration of information and make conclusions as support for selection decisions.
However, the EU AI Act as well as European harmonised standards are being developed through CEN/CENELEC to support the implementation of the AI Act, are put forward to regulate the use of AI tools in recruitment and selection. These standards address areas such as risk management, data governance, transparency, human oversight, accuracy, robustness, cybersecurity, documentation and quality management.
This is where I believe ISO 10667 – the standard for Assessment service delivery - Procedures and methods to assess people in work and organizational settings could provide an valuable bridge in deciding when and how the use of AI tools may be suitable in Assessment processes.
To be clear, ISO 10667 is not a standard for the use of AI tools, and certification against. ISO 10667 does not in itself demonstrate compliance with the AI Act. But the standard provides something that the horisontal AI standards cannot necessarily provide on their own: a well-established framework for how assessment should be carried out in working life to correspond to the interests of the stakeholders involved. Such stakeholders may be the organisation where a position resides (aka “the client”), candidates involved in the process and the provider of the assessment service.
This ISO-standard covers the process from defining the assessment need and requirements, through selecting and using appropriate assessment methods, to integrating information, providing recommendations and reviewing outcomes. It also establishes responsibilities for both the client and the service provider.
The ISO 10667 can therefore be a useful link between the technical and regulatory requirements placed on AI systems and the professional context in which those systems are actually used.
For example, the AI Act may require appropriate risk management, transparency, accuracy and human oversight. ISO 10667 can help place these requirements within a broader assessment process: What is the purpose of the assessment? What should be assessed? Are the methods appropriate for that purpose? How should different sources of information be combined? What should the recommendation contain? And how should the client use that recommendation when making the final decision?
The two perspectives therefore address different but potentially complementary levels:
EU AI Act: establishes the legal requirements for high-risk AI.
Harmonised European AI standards: translate those requirements into more operational standards for AI systems and their use.
ISO 10667: can provide a bridge to the professional recruitment and assessment process in which the AI system is actually applied.
There are also interesting questions about responsibilities. ISO 10667 distinguishes between the client (i.e., the organisation where the position resides) and the service provider, while the AI Act distinguishes between roles such as provider and deployer. These roles may not map perfectly onto each other, but they are analogous, and connecting them could help clarify responsibilities when AI is introduced into assessment and selection services.
ISO 10667 will not cover every aspect of AI-based recruitment. Activities such as targeted job advertising and some forms of sourcing may fall outside its main assessment focus. But from job analysis and assessment planning through assessment, integration, recommendation and client decision-making, there is considerable potential for connection.
Perhaps this is one of the opportunities ahead: not to reinvent quality principles for recruitment and assessment simply because AI tools have entered the process, but to connect the new AI framework with standards and professional practices that already exist.
ISO 10667 was developed to improve the quality, fairness and transparency of assessment in working life.
Those principles seem like a very good starting point for building a bridge to the AI era.