The killer application of skill-based analysis lies in its ability to sort out how things get done in the emerging era of integrated human-AI knowledge work. Frankly, this will make a lot of the debate about text-based skill inference vs. assessment and credentialing irrelevant, because these are questions about the specific relationship between humans and skills, not between work and skills.
To achieve the systematic roll-out of AI into existing workplaces, the attention has to be on the task that needs to be accomplished, and what kind of intelligence needs to be applied to that task. Ambiguous job descriptions or abstract skill categories simply don’t make that possible, because they don’t provide enough resolution of what the work actually looks like.
So if I am saying that neither inference nor assessment are sufficient, what am I calling for?
I am calling for the actual observation and classification of the work that is already happening in organisations. This requires a second great era of work analysis, similar to that which occurred in the manufacturing sector during the robotics revolution. People are not good at understanding and reporting on their own skills, partly because they have no incentive to do so, but also because those skills are mostly performed unconsciously and become invisible to the person performing them. Skills which most humans can perform unthinkingly are also not recognised as skills, but it is critical to recognise them when implementing artificial intelligence…because AI can’t perform them unless explicitly taught to do so (and perhaps not even then).
In the age of AI, skills need to be thought of not as something that only humans have, but as a quality of systems.