What is a skill, anyway?

This is the third panel in my triptych about skill hierarchies. The first two are here:? Don't rely on skill hierarchies and Beyond Skill Hierarchies. Two of the key problems that were pointed out in these posts revolve around the murky, mangled, or missing concept of a skill, our focus here.?

Skills first initiatives, skill-based hiring, bias-free matching of candidates to jobs, and objectively evaluating workers for progression all crucially involve identifying and measuring skills. So, we urgently need to understand and align on a definition for a skill or at least identify criteria for what a good definition will look like. Otherwise these initiatives are all just so many words that evaporate into thin air, leaving nothing behind but dashed hopes.?

I hasten to point out that creating and curating definitions is not always just an endless, frustrating academic exercise, as a common stereotype would have it. Clear, actionable definitions provide significant procedural guidance and dramatically increase the effectiveness of a very wide range of non-academic activities. "Definition" is simply another name for the explicit criteria that we use – the same criteria that enable categorization, analytics, modeling, workflows, and quality management – in research, manufacturing, and especially in the development of new products and technologies. Defining "done", for example, is an on-going challenge for many projects. Standards, like those curated and promoted by the International Standards Organization, are one example of shared definitions with significant practical impact, particularly in manufacturing. Queries for retrieving information from databases are definitions, too:? they specify the features that define what we want to focus on so they form one of the? foundations of data science.?

In spite of their importance in both theory and practice, I have been unable to find any clear guidance on what constitutes a complete definition or what its requisite components should be.? As a result of this lack of guidance or standardization, the definitions that we depend on for so many processes are subjective, often simplistic, and very variable. One way to improve this situation is to focus on delimiting and developing concepts instead of offering succinct sentence fragments to define terms.

Delimiting and Developing the concept of a skill

In place of a typical, laconic definition of the term "skill" like the ability to do something well, we need a richer, more informative characterization of the concept of a skill like the one developed below. To delimit a concept, we have to be explicit about how it is distinguished from its neighbors – the differentia of classical definitions. To develop it, we need to add much more information than a terse sentence fragment allows for.

Consider how to delimit a target concept. We can create a table of closely related terms and concepts that will force us to identify exactly which differences justify the inclusion of new concepts or help us to identify synonyms. These conceptual differences also call attention to the information that will have to be present in a fuller characterization of the target concept. Here's an initial delimitation of the neighborhood of the concept of a skill.?

Skills vs

• competencies -- Skills and competencies seem to be synonyms.
• abilities -- Skills are those abilities that someone can can do well
• steps -- Steps are those skills that are sequenced parts of a procedure
• methods -- Methods are those skills that are alternatives to a given step within a procedure
• complete skills -- Skills include an action; complete skills must also include a "focus"

As for how to develop a concept, we can use a template or checklist to create an array of features (or "triples") with the kinds of information that we would like to see in a more informative characterization of a concept – in particular the kinds that are used to delimit a concept from its neighbors plus other key information.?

Note that this approach is based on one way to define the graph neighborhood around a node for skill in a knowledge graph – in terms of the predicates or features that it usually appears with, i.e., in the style of an intensional definition. The template (or typical list of predicates) then constitutes clearer guidance for what is a complete characterization of a concept. As shown below, a richer characterization includes a longer list of predicates or features for a concept – which means more ways that we can compare one concept to another, e.g., for more detailed and flexible comparisons and inference. Experience shows that more, more granular features enable more robust, more detailed search and reasoning.?

What's a skill?

The view from a dictionary: the ability to do something well

The view from a?taxonomy: Skill subcategoryOf action

The view from a?knowledge graph:

• Skill subcategoryOf action
• Step subcategoryOf skill
• Procedure subcategoryOf skill
• Procedure hasPart step
• Skill hasAgent <person with skill>
• Skill hasPart action
• Skill hasPart focus
• Skill hasPart method
• Skill hasPurpose (broader) skill
• (skill, person) hasProficiency high
• etc.

Parent concept of skills.? A skill is a kind of action that (usually) people can initiate (as opposed to some event that might happen to them) and that they can complete with above average proficiency (i.e., more accurately, more quickly, more efficiently, more stylishly, etc. than most people, as opposed to an ability that most people can do successfully). We talk about people who are skilled or skillful when they are better than most at a given action. We don't talk about people who are skilled at breathing, falling, getting hurt, bleeding, freezing, sleeping, etc. because these are the wrong types of actions. And it's rare to talk about people who are skilled at walking, listening, or eating, because almost everyone does these things with comparable proficiency.

This way of delimiting skills excludes (for a more detailed delimitation) characteristics like positive attitude, hard-working, and awesomeness, as well as topics or areas of knowledge like biochemistry, Mongolian history, and fashion – because they are not actions. This of course does not mean that these are unimportant or less useful concepts, just that they are not skills and need to be processed, represented, and interpreted differently from skills. In fact, especially for terms related to crimes like murder, rape, and child abuse, there's a huge difference between showing murder as a skill (i.e., something a person does particularly well) vs murder as a topic (i.e., something a person knows about). Thousands of people on LinkedIn cite terms like these among their "skills" but are not confessing their roles as perpetrators:? they're simply law abiding detectives, counselors, or lawyers who know about these topics.

Parts of skills. Skills generally involve two parts:? an action and a "focus". For example, baking lasagne, making a soufflé, and firing ceramics are skills that share a similar action involving a hot oven but differ with respect to the focus or kind of object that is affected or created.? Similarly, teaching math, sales training, and coaching soccer share a similar action that involves helping people improve their abilities but differ with respect to the focus – which ability the action focuses on.?

This way of viewing the parts of skills allows us to define a complete skill as one that mentions an action as well as a "focus", which is the specific knowledge of the topic or object that the action focuses on. So the examples above are all complete skills, whereas sales, teaching, marketing, and research are incomplete skills, because although they mention an action they are so general that they do not mention a focus. This distinction is particularly useful because incomplete skills are more vague and far less useful in recommending specific candidates to actual jobs or in suggesting learning materials for specific tasks. In practice, a notion of completeness is also helpful for identifying vague or incomplete items and based on that to trigger different kinds of review, research, and inference, as well as to assign different levels of reliability. For example, if we identify "sales" as an incomplete skill, then instead of accepting and tallying it as is, we would trigger a process to identify, as possible, key missing information – like what kinds of products the skill mention refers to or which step in the sales process is intended in a particular context.?

Documenting parts of skills is particularly useful when we store the parts of a complete skill separately instead of storing only the skill name as a blob. When we store teaching math as [action:teaching, focus:math], we can search for each part or feature independently and find skills that are related to the same action (teaching biology, teaching geography, etc.) as easily as skills that are related to the same focus (learning math, applying math, etc.).

Parts of complex skills. Skills are often complex actions like procedures made up of steps (which are actions or skills themselves) that are ordered (or simultaneous) in time. To make a soufflé (a procedure), for example, we need to go through steps like: source the ingredients, mix them according to a recipe, bake them correctly, and plate the result attractively – all of these are actions that we may or may not be skilled at, and they make up the steps of make a soufflé.? A step, then, is one part of a procedure.

Many procedures can be carried out in different ways, with different methods.? This is to say that there can be variations in or options for a particular step of a procedure, and these variations are also actions or skills themselves.? For example, we can increase product awareness (a procedure) through product advertising (one step) and for that we can consider several methods for the same step:? print advertising, digital advertising, radio advertising, product placement in movies, celebrity endorsements, etc. These methods are all related but we do not do them in order, so they are not steps of product advertising:? they are different methods for executing the same step. So one skill can be a step or a method of another, more complex, skill:? radio advertising is a method of product advertising and product advertising is a step of increasing product awareness.?

Identifying the parts of a single skill and the more granular steps and methods of a procedure creates more granular, more detailed features that delimit these skill concepts more precisely, yield more detailed representations of them, and enable automated knowledge validation as well as far more robust search, comparison, and inference over skills than simple labels or human-readable definitions.

Progeny or types of skills. We can create collections (or types) of skills in many ways, based on different criteria:? the same actions, focus, shared steps or methods, common goals or procedures … or by grouping them by their relations to other things: for example, to the job roles, industries, companies, etc. that they are most relevant for.?

We might say that "marketing" as a skill includes the steps of marketing like planning campaigns, buying advertising, lead generation, lead qualification, etc. But we might also create collections like real estate marketing, software marketing, automobile marketing, service marketing, etc. where "marketing" emphasizes the different kinds of products and services that you are marketing – the foci or domains. And categories like "advertising" as a skill with digital advertising, TV advertising, print advertising, influencer marketing, product placement, etc. emphasize the different methods we use to complete a given action.??

All of these options (and many others) for grouping types of skills are valid and useful, and this highlights why simple hierarchies – which only encode one single set of skill types – distort reality in dramatic ways. This is one reason why so many (often incompatible) taxonomies are possible for the same domain – there is simply no one "true" or "best" way to group them. Including all of these types indiscriminately into a vague collection like "marketing" leads to confusion because no one can tell for sure how they're related. The key to encoding this variety of types is to build a knowledge graph for skills (which allows for multiple relations between skills) rather than a simplistic hierarchy (which only allows for one relation (parent) with one value). We need different predicates in the graph to indicate each criterion:? steps, domains, methods, etc., not just "parent" or subcategoryOf.

This template or checklist for developing the concept of a skill includes information about neighbors, parents, parts, relations between parts, and progeny or types.? It is already far more informative than simply the ability to do something well.? We can continue to extend the template and include additional information about the purposes, properties, and products of skills, as well as the reminder to include individual examples and other key information. This is natural in a knowledge graph but difficult to do with simple hierarchies. Delimiting and developing the concept of a skill with nodes in a knowledge graph creates much richer, more informative representations that can enable a wider variety of operations and more reliable applications in HR tech.

Would you use the dictionary definition of "building" as a blueprint to build a skyscraper? Certainly not. But we are doing exactly that when we build large-scale knowledge architectures, data pipelines, and technologies on a foundation of subjective categories defined by terse sentence fragments. There simply is not enough information to guide us.? Skills first initiatives, skill-based hiring, bias-free matching of candidates to jobs, and objectively evaluating workers for progression simply cannot succeed without a richer, better-delimited concept of a skill. Better concepts play a crucial role in developing technology everywhere.

What other information do you think we need to include in our concept of skill?